Compare commits

..

1 Commits

Author SHA1 Message Date
Dai Ha bf2d940c26 fleetd #669: fix fleetd.example.yaml's false scan-exclusion claims; cover the shipped shape
CI / shell-tests (pull_request) Failing after 9s
CI / contract (pull_request) Successful in 51s
CI / build (pull_request) Failing after 2m10s
fleetd.example.yaml claimed member spaces are excluded from the lead-tab scan and that
a lead's workspace default is "leads" and must not match a member workspace. Both are
false: production always passes an empty excludedWorkspaceLabels set (CB-558), and a
lead's workspace defaults to the same shared "fleet" space the members use. Rewrite
both paragraphs to name the defenses that actually exist: the startup tabPrefix refusal
and CallerResolver's roster-first precedence.

Add the test nobody wrote: a lead is still discovered when its workspace label equals
the member workspace, with an empty exclusion set. Correct
aTabInAWorkerSpaceIsNeverALeadEvenWhenItsLabelMatches's javadoc, which overclaimed that
the excludedWorkspaceLabels parameter is the live production guard. Drop the stale line
number from FleetdAssemblyLeadTabScannerExclusionTest's assertion message and javadoc.
2026-10-04 06:31:20 +02:00
35 changed files with 425 additions and 4118 deletions
+7 -35
View File
@@ -159,29 +159,6 @@ fails.
`{action: "cancel", token}` drops a pending request without rolling.
**A change is coming: the roll will restart the process instead of sending `/clear` (fleetd #726
unit 2, written 2026-10-04).**
Today a roll types `/clear` into your pane. Your `claude` process keeps running, so a newer CLI on
disk is never loaded. Unit 2 replaces that: the daemon ends the old pane, launches a fresh one,
waits for the new terminal to be recognised as a lead, and only then sends the bootstrap text.
**Everything below about `/clear` is accurate while the old jar is running.** Unit 2 was not merged
when this note was written, and a merge is not a deployment.
**How to tell which one is live: read your own tool list.** If `fleet_handover`'s description says
it will "clear your pane", the daemon is serving the old behaviour. If it names a restart, the new
behaviour is live. The description comes from the running daemon, so it cannot disagree with the
code that is actually loaded.
Two things change for you once it is live. The `status` outcomes are different: three new failures
replace the `/clear` ones. And the "never observed as WORKING after 8 consecutive IDLE/DONE polls"
warning described below can no longer appear, because that wait is deleted — so if you still see
it, the old jar is running. `TURN_NEVER_SETTLED` does not change, and still means nothing was
touched.
**Delete this note and rewrite the `/clear` paragraphs once the new jar is live.**
**Things that will surprise you:**
- **`accepted` does not mean your pane has been cleared.** It means every gate passed and the roll
@@ -223,18 +200,13 @@ touched.
- **The roll can still refuse after `confirm` returns**, and by then there is no caller to tell.
Those outcomes are logged only, as `lead-rollover:` lines in the daemon log.
- **The bootstrap prompt works end to end. Measured 2026-09-22, re-measured 2026-10-04.** This used
to say the fix was unproven (fleetd #489) and told you to expect a failure. That is no longer
true. On 2026-09-22 the daemon log held four `lead-rollover: rolled` lines. On 2026-10-04 it holds
**20**, against a control of 86 `lead-rollover:` lines. Each roll cleared the old lead and started
a fresh session against the handover file, with the configured `bootstrapText` arriving as its
first message. No context was lost. The old `Unknown command: /clearFresh` failure from 2026-09-12
does not appear in the log at all.
19 of the 20 carry an `elapsedMs`: median 16507 ms, maximum 48261 ms, and two above 45000 ms. That
figure times the **whole** roll, and the wait for your own turn to end dominates it, so do not
read it as the cost of the clear. Expect a roll to take tens of seconds, and do not treat a slow
one as a failed one. Re-measure all of these with:
- **The bootstrap prompt works end to end. Measured 2026-09-22.** This used to say the fix was
unproven (fleetd #489) and told you to expect a failure. That is no longer true. The daemon log
now holds four `lead-rollover: rolled` lines, and three of them ran on 2026-09-22 at 10:01:43,
10:38:28 and 11:15:47. Each one cleared the old lead and started a fresh session against the
handover file, with the configured `bootstrapText` arriving as its first message. No context was
lost. The old `Unknown command: /clearFresh` failure from 2026-09-12 does not appear in the log
at all. Re-measure both numbers with:
```bash
grep -c "lead-rollover: rolled" fleetd/fleetd.out # successful rolls
+23 -44
View File
@@ -27,31 +27,26 @@ through its `fleet_*` tools. No session addresses a peer, a broker, or the netwo
**Every role reads this file.** A member runs in a git worktree of this same repo, so it inherits
this `CLAUDE.md` verbatim, and every rule below is role-conditional.
**Call `fleet_whoami`.** It returns `primary`, `worker`, `architect`, `collaborator`, or `observer`,
resolved by the daemon from your connection — unforgeable, and the same resolution its authorization
gate uses. A worker also carries its `sessionId`, `profile`, `worktree` and `branch`; an architect
carries the slot name it was bound to; a collaborator carries its registry name and its own
`sessionId`, and **no `leader` key** — a collaborator is a named peer, not a primary. An **observer**
carries only its own `sessionId`: a pane the daemon could not place as any of the above, authorized
to `READ`/`METRICS` and to `REPLY`/`ASK` on its own pane and nothing more — never `SEND`, never a
ticket. Don't infer what you can ask.
**Call `fleet_whoami`.** It returns `primary`, `worker`, `architect`, or `collaborator`, resolved by
the daemon from your connection — unforgeable, and the same resolution its authorization gate uses.
A worker also carries its `sessionId`, `profile`, `worktree` and `branch`; an architect carries the
slot name it was bound to; a collaborator carries its registry name and its own `sessionId`, and
**no `leader` key** — a collaborator is a named peer, not a primary. Don't infer what you can ask.
Only if that call is unavailable, fall back to these — each is one-way, so keep reading until one
fires: the reply charter in your system prompt (*"You are a spawned member in the
claude-bridge fleet"*) ⇒ **spawned member**; fleet tools prefixed `mcp__fleet__*` ⇒ **spawned
member** (the launcher fixes that mount name; a primary's mount is named by whoever wrote its
`.mcp.json`, so it varies — and a member spawned before CB-632 still says `mcp__bridge__*`); `ANTHROPIC_BASE_URL` set ⇒ **spawned member** (Claude-model members run
on a clean env, so its *absence* proves nothing). None of these separate a worker from an architect,
or a worker from an **observer** — an observer is just as unspawned as a collaborator and carries
none of these signals either, so only `fleet_whoami` tells the two apart. **And none of them fires
for a collaborator at all**: every signal in the ladder detects a *spawned* member, while a
collaborator is a tab a person opened by hand, so it has no charter, no fixed mount name and a
normal environment. A collaborator — or an observer — that cannot call `fleet_whoami` therefore falls
to the line below and acts as a worker. That is the safe direction — it under-privileges, and the
refusals are loud — but it means a collaborator or an observer has no way to learn what it is except
by asking. **Still unsure ⇒ act as a worker**, the most restricted member role this ladder can name.
The two mistakes are not symmetric: a primary acting as a worker is refused by the authorization gate
— loud and self-correcting — while a member acting as the primary ends its turn with no `fleet_reply`,
on a clean env, so its *absence* proves nothing). None of these separate a worker from an architect —
only `fleet_whoami` does. **And none of them fires for a collaborator at all**: every signal in the
ladder detects a *spawned* member, while a collaborator is a tab a person opened by hand, so it has
no charter, no fixed mount name and a normal environment. A collaborator that cannot call
`fleet_whoami` therefore falls to the line below and acts as a worker. That is the safe direction —
it under-privileges, and the refusals are loud — but it means a collaborator has no way to learn
what it is except by asking. **Still unsure ⇒ act as a worker**, the most restricted member role. The
two mistakes are not symmetric: a primary acting as a worker is refused by the authorization gate —
loud and self-correcting — while a member acting as the primary ends its turn with no `fleet_reply`,
and the sender silently receives nothing. Fail toward the recoverable error.
### Invariants — every role, no exceptions
@@ -105,11 +100,7 @@ below are the procedure — run them in order, every task, not only the big ones
5. **Collect** — `fleet_poll{ticket}` → `fleet_ack{target, msgId}`. Answer a worker's `fleet_ask`
with `fleet_send{turnId, content}` — **not** `sessionId`. A worker gone quiet is diagnosed with
`fleet_status`, never by reading its terminal; it also reports an open question and the `turnId`
that answers it — but **only to the caller that created that delegation**, so a question raised
under an architect's brief is invisible to you, and seeing none does not mean there is none.
**Only that same creator can answer it.** A `turnId` you came by any other way is refused, so an
architect's worker waits for that architect and not for you.
**A worker's ask waits ~55 seconds, and no nudge makes that longer** — so never
that answers it. **A worker's ask waits ~55 seconds, and no nudge makes that longer** — so never
brief a worker to "ask me". Decide before you delegate, or give it an explicit default.
**A correction cannot reach a busy member.** A `fleet_send` to a working member is *accepted* and
returns a ticket, and is then never delivered — measured here three times in one session, and the
@@ -174,10 +165,10 @@ you decide.
| See the fleet | `fleet_list` → `leads` (your peers) + `members` (each carries `agentSessionId` when its backend knows one) + `loopHealth` (`RUNNING`, `STALLED`, or `STOPPED` for `statusPoller` and `sessionReaper`) · one peer's state: `fleet_status{sessionId}` |
| Delegate (blocking) | `fleet_send{sessionId, content}` |
| Delegate (long task) | `fleet_send{sessionId, content, wait:false}` → ticket → `fleet_poll{ticket}` |
| Answer a member's `fleet_ask` | `fleet_send{turnId, content}` — **not** `sessionId`, and only the caller that created that delegation |
| Answer a member's `fleet_ask` | `fleet_send{turnId, content}` — **not** `sessionId` |
| Message a **peer lead** on this host | `fleet_send{sessionId: <their terminal>, content}` — `fleet_list` → `leads` reports it. Coordination only, **never** a task |
| Message a **peer lead** on another daemon or host | `fleet_send{coordId: <their coord-id>, content}` — needs a `coordinator:` block; your own coord-id is in `fleet_list`. Coordination only, **never** a task |
| Message a **collaborator** on this host | `fleet_send{sessionId: <their terminal>, content}` — `fleet_list` reports a `collaborators` array, and each row carries that peer's `name` and the `sessionId` you send to. It is visible to you, to an architect and to another collaborator, never to a worker. Coordination only, **never** a task |
| Message a **collaborator** on this host | `fleet_send{sessionId: <their terminal>, content}` — but **`fleet_list` does not report collaborators**, so you cannot discover one: it must tell you its `sessionId`, which its own `fleet_whoami` gives it. Coordination only, **never** a task |
| Answer a peer lead that messaged you | `fleet_send{coordId}` — or `{sessionId}` if they are on this host. **Not** `fleet_reply`: it has no peer route and the publish is refused |
| Read your own held lead-to-lead mail (no ack) | `fleet_poll{coordId: <your own coord-id, from fleet_list's coordinator.selfId>}` — primary-only; never acks, so `fleet_list`'s `held[]` still shows it after. `fleet_list`'s `held[]` gives only a truncated preview — this is the only way to read the full body |
| Collect a held reply | `fleet_poll{target}` · then `fleet_ack{target, msgId}` |
@@ -266,8 +257,8 @@ of those is refused at the gate, not queued.
Two limits worth knowing before you hit them. **You cannot reach a worker** — not even to help one —
because a worker belongs to the lead that spawned it, and routing around that would make you a
second orchestrator with no plan. Send to the lead instead. And **you cannot read a ticket**, so you
cannot collect a delegation's reply: `fleet_poll` refuses you at the role gate, and a ticket also
records the terminal that created it, so even a leaked id reads nothing.
cannot collect a delegation's reply; ticket ids are a plain counter with no owner check, so holding
one would let you walk every other session's answers.
Being named buys you a channel, not authority. Your `fleet_send` to a lead is coordination between
peers: the lead owes you no obedience, and you owe it none.
@@ -339,22 +330,10 @@ must obey belongs in the charter, not here.
(#362). **Read `plugin/` before designing anything about onboarding a project.** Two limits are
structural, not bugs: a plugin cannot carry the role agent files, because
`ClaudeCodeLauncher.java:371` requires `<cwd>/.claude/agents/<role>.md` in the member's own
worktree; and a plugin reaches a member only through `CLAUDE_CONFIG_DIR`, which
`ClaudeCodeLauncher.java:286` exports with `putIfPresent` — so only for a profile that sets
`configDir`. Every `claude-code` profile does set one (the four without are `opencode`, which
never reads that variable). **But measured 2026-10-04: two of them point at
`~/.ccs/instances/ltms`, which is the operator's own `CLAUDE_CONFIG_DIR` on this host.** So for an
`opus` or `sonnet` member, "a member never reads the operator's plugin store" is false — it reads
the same store, because that store is the one its `configDir` names. It stays true for `local` and
`local-direct`, which point at `~/.ccs/instances/gx10`. `ClaudeCodeLauncher`'s own javadoc names
the related hazard: that file is rewritten on every spawn, so for those two profiles fleetd and the
operator's live session write the same `.claude.json`, and its compare-and-swap "narrows the
lost-update window, it does not close it". Re-measure which profiles share the operator's dir with
`awk '/^profiles:/{i=1;next} /^[a-z]/{i=0} i&&/^ [a-z-]+:$/{p=$1} i&&/configDir:/{print p,$2}'
fleetd/fleetd.yaml` against `echo $CLAUDE_CONFIG_DIR`; delete this note once no profile names the
operator's dir. **Treat the plugin as the lead-side surface and put member-facing assets in the
worktree** — that conclusion holds either way, because a worktree asset does not depend on which
config dir a member reads.
worktree; and a plugin cannot deliver anything to members at all, because
`ClaudeCodeLauncher.java:285` exports `CLAUDE_CONFIG_DIR` and every Claude profile here sets it,
so a member never reads the operator's plugin store. **The plugin is the lead-side surface;
member-facing assets travel in the worktree.**
- **Never commit** `.mcp.json` (the primary's local copy, flagged `--skip-worktree`) or `wiki/`
(a submodule with its own remote).
- **A provisioned worktree neutralizes `.mcp.json`, `opencode.json` and `.autoenv`** — the repo's
@@ -226,8 +226,9 @@ public final class Fleetd {
* is the same way — a person's own tab, matched to a configured name, never spawned.
*
* <p>Neither a lead nor a collaborator is ever enrolled in {@link MemberPresence} — {@code
* FleetMcp} marks presence only for a worker, an architect, or the unconfigured-pane floor,
* never for a lead or a collaborator. So without the second and third disjuncts a lead or
* FleetMcp} marks presence for every spawned member (worker and architect), deliberately, since
* that map doubles as the member roster's availability signal and a lead or collaborator counted
* there would show up as an available member. So without the second and third disjuncts a lead or
* collaborator is permanently un-deliverable: every send to one sat on the gate for
* {@code READINESS_GRACE_POLLS} (~60s) and then failed having never been typed into the pane.
*
@@ -51,6 +51,7 @@ import dev.ltms.fleet.power.CaffeinateSleepAssertionMechanism;
import dev.ltms.fleet.power.IdleSleepGuard;
import dev.ltms.fleet.rest.FleetApp;
import dev.ltms.fleet.session.GitWorktrees;
import dev.ltms.fleet.session.MemberSession;
import dev.ltms.fleet.session.SessionManager;
import dev.ltms.fleet.session.SessionReaper;
import io.javalin.Javalin;
@@ -479,9 +480,13 @@ final class FleetdAssembly {
new PaneLocator(herdr, memberHerdr), new LsofPeerPidLookup(), new LsofProcessCwdLookup());
// fleetd #669 Unit D: a live spawned member resolves as its own role, whatever a tab map
// says about the same terminal. fleetd #702: SessionManager.spawnedMemberRole also answers
// for a pane mid-teardown, not only one still in the registry — see its javadoc.
Function<String, MemberRole> spawnedMemberRole = sessions::spawnedMemberRole;
// says about the same terminal — read from the roster meant for a hot path (SessionManager
// javadoc), never rosterResolved(), since resolve() runs on every request.
Function<String, MemberRole> spawnedMemberRole = terminal -> sessions.roster().stream()
.filter(s -> terminal.equals(s.terminalId()))
.map(MemberSession::role)
.findFirst()
.orElse(null);
// CB-501: one resolver behind both entry paths. Worker identity still comes from the
// connection and is never token-gated, so enabling token mode cannot lock the fleet out.
@@ -75,9 +75,6 @@ public final class Authz {
* {@code SEND} is refused, as if no terminal were a configured lead or collaborator — the
* same decision {@link #NO_KNOWN_LEAD_OR_COLLABORATOR} gives explicitly. Every other action's
* result is identical to the four-argument form's, since none of them consult the classifier.
*
* <p>Its default classifier denies every collaborator, so a caller enforcing authorization
* must use the four-argument form instead.
*/
public static boolean permits(Principal caller, Action action, String targetSession) {
return permits(caller, action, targetSession, NO_KNOWN_LEAD_OR_COLLABORATOR);
@@ -138,15 +135,13 @@ public final class Authz {
// fleet_whoami — and carries no secrets: no ticket reply, no pending question, and no
// other session's turn state. Those live under TASK_READ. METRICS is the separate
// Prometheus scrape. Both are open to every authenticated role, including a
// collaborator and the unconfigured-pane floor.
// collaborator.
case READ, METRICS -> caller.isPrimary() || caller.isWorker() || caller.isArchitect()
|| caller.isCollaborator() || caller.isObserver();
|| caller.isCollaborator();
// Ticket polling and session status, open to every role READ is open to except a
// collaborator or an observer. MessageService compares a ticket's creator to the
// caller on every read as well, so dropping this gate would not expose another
// session's reply — it would move the refusal later and widen what a caller that
// never orchestrates can probe.
// collaborator: ticket ids are a sequential counter with no owner check, so a holder
// could walk every ticket and read another session's delegation reply.
case TASK_READ -> caller.isPrimary() || caller.isWorker() || caller.isArchitect();
// fleetd #421: reading held lead-to-lead mail is the primary's alone. An architect
@@ -40,7 +40,7 @@ import java.util.function.Supplier;
* the case the previous step does not catch: a binding with no live spawned-member session.</li>
* <li>A loopback peer PID that maps to an operator-labelled collaborator tab ⇒
* {@link Role#COLLABORATOR}, carrying that collaborator's name.</li>
* <li>A loopback peer PID that maps to any other herdr pane ⇒ {@link Role#OBSERVER}. This is
* <li>A loopback peer PID that maps to any other herdr pane ⇒ {@link Role#WORKER}. This is
* unforgeable (the OS reports the PID, herdr owns the PID→pane map) and is honoured
* regardless of auth mode, so enabling auth never breaks the fleet.</li>
* <li>Otherwise, under {@code token} mode, a valid bearer token ⇒ {@link Role#PRIMARY}.</li>
@@ -249,6 +249,17 @@ public final class CallerResolver {
return leadTerminals.get();
}
/**
* The currently-recognised architect slots, {@code terminal_id → slot name} (CB-548).
*
* <p>Read from the same supplier {@link #resolve} consults, so a slot that is <em>listed</em>
* here but would not <em>resolve</em> (or the reverse) cannot drift apart. Live for the same
* reason as {@link #leads()}.
*/
public Map<String, String> members() {
return architectTerminals.get();
}
/**
* The currently-recognised collaborator tabs, {@code terminal_id → name}.
*
@@ -323,7 +334,7 @@ public final class CallerResolver {
// of the above keeps that stronger role.
return Principal.collaborator(collaborator, c.terminal(), c.pid());
}
return Principal.observer(c.terminal(), c.pid()); // unforgeable; never token-gated
return Principal.worker(c.terminal(), c.pid()); // unforgeable; never token-gated
}
if (tokenMode) {
@@ -88,13 +88,6 @@ public record Principal(Role role, String terminal, long pid, String name) {
return new Principal(Role.COLLABORATOR, terminal, pid, name);
}
/**
* The unconfigured-pane floor: a loopback caller whose pane matched no other role.
*/
public static Principal observer(String terminal, long pid) {
return new Principal(Role.OBSERVER, terminal, pid);
}
public boolean isPrimary() {
return role == Role.PRIMARY;
}
@@ -111,15 +104,11 @@ public record Principal(Role role, String terminal, long pid, String name) {
return role == Role.WORKER;
}
public boolean isObserver() {
return role == Role.OBSERVER;
}
/**
* Whether this caller is a spawned member with its own pane.
*
* <p>Both workers and architects are spawned members. A lead is not: it is a peer the
* operator started and named, never a pane this daemon spawned.
* <p>Both workers and architects are spawned members. A lead is excluded because recording it
* as present would count it as an available member in the roster.
*/
public boolean isSpawnedMember() {
return role == Role.WORKER || role == Role.ARCHITECT;
@@ -151,7 +140,6 @@ public record Principal(Role role, String terminal, long pid, String name) {
case WORKER -> "worker:" + terminal;
case ARCHITECT -> "architect:" + name;
case COLLABORATOR -> "collaborator:" + name;
case OBSERVER -> "observer:" + terminal;
case PRIMARY -> name == null ? "primary" : "leader:" + name;
case ANONYMOUS -> "anonymous";
};
@@ -45,17 +45,6 @@ public enum Role {
*/
COLLABORATOR,
/**
* A loopback pane that resolved to none of the roles above: not a live spawned member, not a
* configured lead, not a bound architect slot, not a configured collaborator tab. Unforgeable
* like a worker's — derived from the connection's pane, never from a request argument, and
* honoured regardless of auth mode. May {@code READ} and {@code METRICS}, and {@code REPLY}/
* {@code ASK} only as its own pane; may not {@code SPAWN}/{@code STOP}/{@code DRAIN}/
* {@code HANDOVER}, {@code SEND}, poll a ticket ({@code TASK_READ}), or reach the coordination
* broker ({@code COORD_SEND}/{@code COORD_READ}).
*/
OBSERVER,
/** Authenticated as nothing. Authorized for nothing but {@code /healthz}. */
ANONYMOUS
}
@@ -2776,10 +2776,9 @@ public record FleetConfig(
});
if (!bad.isEmpty()) {
throw new IllegalStateException("refusing to start: " + String.join("; ", bad)
+ ". A member labelled that way, while its pane carries no entry in the "
+ "spawned-member roster, is read back as a lead or collaborator and granted "
+ "that identity's authority. Change one of the two so member tabs cannot be "
+ "confused with a lead's or collaborator's tab.");
+ ". Every member labelled that way would be read back as a lead or "
+ "collaborator and granted that identity's authority. Change one of the two "
+ "so member tabs cannot be confused with a lead's or collaborator's tab.");
}
List<String> collisions = new ArrayList<>();
@@ -2878,8 +2877,7 @@ public record FleetConfig(
* the focused tab rather than its own, so it can land inside a lead's or collaborator's own
* labelled tab. {@link dev.ltms.fleet.herdr.LeadTabScanner} identifies a lead or collaborator
* purely by that tab's label — it does not exclude the member space — so a member that ends up
* there, while its pane carries no entry in the spawned-member roster, is read back as that
* lead or collaborator and granted that identity's authority.
* there would be read back as that lead or collaborator and granted that identity's authority.
*
* <p>Only an entry with a non-blank {@code tab} is in scope: one with no {@code tab} feeds
* nothing into {@link dev.ltms.fleet.herdr.LeadTabScanner}, so it creates no hazard here.
@@ -2910,9 +2908,8 @@ public record FleetConfig(
}
throw new IllegalStateException("refusing to start: profile(s) " + bad
+ " use placement: pane while fleet.leaders or fleet.collaborators names a tab. A "
+ "pane-placed member can land inside that labelled tab, and while its pane "
+ "carries no entry in the spawned-member roster, it is read back as the lead or "
+ "collaborator and granted that identity's authority. Set placement: tab for "
+ "pane-placed member can land inside that labelled tab and be read back as the "
+ "lead or collaborator, granted that identity's authority. Set placement: tab for "
+ "each named profile, or remove the tab from every fleet.leaders and "
+ "fleet.collaborators entry.");
}
@@ -1,146 +0,0 @@
package dev.ltms.fleet.herdr;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import java.nio.charset.StandardCharsets;
import java.util.List;
import java.util.function.IntFunction;
/**
* The three protections every {@code agent.start} caller needs against herdr's pane-typed launch
* surface (fleetd #220, #727): a byte-limit check on the assembled command line, a bounded retry
* on {@code agent_pane_busy} (the target pane's shell has not reached its prompt yet), and a
* bounded retry on {@code agent_name_taken} with a fresh name each attempt. One implementation —
* every caller of {@code agent.start}, lead or member, goes through this seam rather than carrying
* its own copy.
*/
public final class ResilientAgentLaunch {
private static final Logger log = LoggerFactory.getLogger(ResilientAgentLaunch.class);
private ResilientAgentLaunch() {
}
/**
* The pty line buffer herdr types a launch command into: BSD/macOS {@code MAX_CANON}. Not a
* fleetd choice and not configurable — see {@link #checkFits}.
*/
public static final int PANE_COMMAND_BYTE_LIMIT = 1024;
/** Per-argument allowance for the separating space and a shell quote pair fleetd cannot see. */
private static final int QUOTING_OVERHEAD_PER_ARG = 3;
/** herdr rejects a duplicate agent {@code name}; a caller retries a bumped name this many times. */
public static final int NAME_RETRIES = 8;
/**
* Retries for {@code agent.start} against a seed pane whose shell has not reached its prompt
* yet — {@code tab.create}/{@code pane.split} return as soon as the pane exists, and herdr
* refuses to start an agent in a pane that is not "an available shell" ({@code agent_pane_busy}).
*/
public static final int SHELL_READY_RETRIES = 20;
/** Raised by {@link #checkFits} when the assembled command cannot fit the pane line. */
public static final class TooLargeException extends RuntimeException {
public TooLargeException(String message) {
super(message);
}
}
/**
* Verify the assembled launch line fits the pane herdr types it into. herdr does not exec the
* launch command — it TYPES it into the pane as one line, and a pty line buffer holds only
* {@value #PANE_COMMAND_BYTE_LIMIT} bytes. Everything past that byte is dropped with no error
* anywhere: herdr answers "agent started", the backend exits on the mangled argument it was
* handed, the pane closes, and the only symptom is a readiness timeout with no reason. That is
* how fleetd #214 broke every claude-code spawn — one 50-byte flag pushed a 978-byte command to
* 1028, and the tail that got cut was {@code --autocompact 250000}.
*
* <p>So measure it here and refuse, loudly and immediately, rather than start something that
* cannot work. The estimate is deliberately conservative: fleetd cannot see herdr's quoting, so
* every argument is charged its own bytes plus a separator and a quote pair. An over-estimate
* costs a clear error at a length that was already unsafe; an under-estimate would let the
* silent truncation back in.
*
* @param label names the launch in the refusal message (a profile name)
* @param argv the full argv, including the executable at index 0
* @throws TooLargeException naming the limit, the estimate, and the longest argument
*/
public static void checkFits(String label, List<String> argv) {
int bytes = 0;
String longest = null;
int longestBytes = 0;
for (String arg : argv) {
int argBytes = arg == null ? 0 : arg.getBytes(StandardCharsets.UTF_8).length;
bytes += argBytes + QUOTING_OVERHEAD_PER_ARG;
if (argBytes > longestBytes) {
longestBytes = argBytes;
longest = arg;
}
}
if (bytes <= PANE_COMMAND_BYTE_LIMIT) {
return;
}
String culprit = longest == null ? "<none>"
: longest.substring(0, Math.min(longest.length(), 60)) + (longest.length() > 60 ? "…" : "");
throw new TooLargeException(
"launch command for " + label + " is about " + bytes + " bytes, over the "
+ PANE_COMMAND_BYTE_LIMIT + "-byte limit of the pane line herdr types it into. "
+ "The pty would drop the tail silently and the backend would exit on a mangled "
+ "argument. Longest argument is " + longestBytes + " bytes: " + culprit
+ " — move it off the command line (a file flag) or shorten it.");
}
/**
* Start an agent into {@code paneId}, retrying {@code agent_pane_busy} up to {@code retries}
* times with {@code sleeper} run between attempts.
*
* @throws HerdrException the last {@code agent_pane_busy} failure once {@code retries} is
* spent, or immediately for any other herdr failure
*/
public static Agent startAwaitingShellPrompt(AgentControl agents, String name, String kind,
List<String> args, String paneId,
int retries, Runnable sleeper) {
HerdrException busy = null;
for (int attempt = 0; attempt < retries; attempt++) {
try {
return agents.start(name, kind, args, paneId);
} catch (HerdrException e) {
if (!"agent_pane_busy".equals(e.code())) throw e;
log.debug("pane {} not at its shell prompt yet, retrying agent.start", paneId);
busy = e;
sleeper.run();
}
}
throw busy;
}
/**
* Start an agent under a freshly generated name each attempt, retrying {@code agent_name_taken}
* up to {@code nameRetries} times — herdr refuses a duplicate {@code name} outright, so a stale
* registry entry (a crashed session, a name the registry has not yet released) must not block a
* legitimate relaunch. Each attempt also carries its own {@link #startAwaitingShellPrompt} retry.
*
* @param nameForAttempt called once per attempt (0-based) to produce that attempt's name
* @throws HerdrException the last {@code agent_name_taken} failure once {@code nameRetries} is
* spent, or immediately for any other herdr failure
*/
public static Agent startUniquelyNamed(AgentControl agents, String kind, List<String> args,
String paneId, IntFunction<String> nameForAttempt,
int nameRetries, int shellReadyRetries, Runnable sleeper) {
HerdrException last = null;
for (int attempt = 0; attempt < nameRetries; attempt++) {
String name = nameForAttempt.apply(attempt);
try {
return startAwaitingShellPrompt(agents, name, kind, args, paneId,
shellReadyRetries, sleeper);
} catch (HerdrException e) {
if (!"agent_name_taken".equals(e.code())) throw e;
log.debug("agent name '{}' taken, retrying", name);
last = e;
}
}
throw last;
}
}
@@ -5,7 +5,6 @@ import dev.ltms.fleet.herdr.Agent;
import dev.ltms.fleet.herdr.AgentControl;
import dev.ltms.fleet.herdr.HerdrException;
import dev.ltms.fleet.herdr.PendingCloseMarker;
import dev.ltms.fleet.herdr.ResilientAgentLaunch;
import dev.ltms.fleet.herdr.Tab;
import dev.ltms.fleet.herdr.Workspace;
import dev.ltms.fleet.herdr.WorkspaceControl;
@@ -13,7 +12,6 @@ import dev.ltms.fleet.launch.ClaudeCodeArguments;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import java.security.SecureRandom;
import java.util.ArrayList;
import java.util.LinkedHashMap;
import java.util.LinkedHashSet;
@@ -21,7 +19,6 @@ import java.util.List;
import java.util.Map;
import java.util.Objects;
import java.util.Set;
import java.util.concurrent.atomic.AtomicLong;
import dev.ltms.fleet.peer.PeerLauncher;
/**
@@ -83,19 +80,9 @@ public final class LeadLauncher {
private static final Logger log = LoggerFactory.getLogger(LeadLauncher.class);
/** Attempts {@link #relaunch(String)} makes before giving up and returning {@code null}. */
static final int RELAUNCH_ATTEMPTS = 3;
private final AgentControl agents;
private final WorkspaceControl spaces;
private final FleetConfig cfg;
private final Runnable sleeper;
// Per-process token mixed into each lead agent name so a fresh daemon process (seq back at 0)
// cannot collide with a same-name lead that outlived a restart — the same scheme
// HerdrPeerLauncher uses for members (fleetd #727).
private final String nameNonce = String.format("%06x", new SecureRandom().nextInt(1 << 24));
private final AtomicLong nameSeq = new AtomicLong();
/**
* @param agents herdr agent control (start, list)
@@ -103,27 +90,9 @@ public final class LeadLauncher {
* @param cfg the loaded config — {@code fleet.leaders}, {@code profiles} and each lead's tab
*/
public LeadLauncher(AgentControl agents, WorkspaceControl spaces, FleetConfig cfg) {
this(agents, spaces, cfg, () -> sleepUninterruptibly(300));
}
/**
* Test seam: as above, plus an injectable {@code sleeper} for the {@code agent_pane_busy}
* retry (fleetd #727), so a test can prove the retry budget without a real sleep.
*/
LeadLauncher(AgentControl agents, WorkspaceControl spaces, FleetConfig cfg, Runnable sleeper) {
this.agents = agents;
this.spaces = spaces;
this.cfg = cfg;
this.sleeper = sleeper;
}
/** Uninterruptible sleep — the production {@link #sleeper} between {@code agent_pane_busy} retries. */
private static void sleepUninterruptibly(long ms) {
try {
Thread.sleep(ms);
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
}
}
/**
@@ -204,14 +173,24 @@ public final class LeadLauncher {
log.info("lead '{}': {} live, {} wanted — nothing to start", name, running, wanted);
continue;
}
if (!lead.isCreatable()) {
// A lead with a `tab:` but no `profile:` is recognise-only by design: the operator
// opens it by hand. Say so once rather than looking like a silent failure.
log.info("lead '{}' is not live, and names no profile — it can be recognised but not "
+ "launched. Add `profile:` under fleet.leaders.{} to have fleetd start it.",
name, name);
continue;
}
ResolvedLead resolved = resolveLaunchable(name);
if (resolved == null) {
FleetConfig.Profile profile = cfg.profiles().get(lead.profile());
if (profile == null) {
log.warn("lead '{}' names profile '{}', which is not configured — not launching",
name, lead.profile());
continue;
}
for (int i = running; i < wanted; i++) {
if (launch(name, resolved.lead(), resolved.profile()) != null) {
if (launch(name, lead, profile)) {
started++;
}
}
@@ -219,82 +198,6 @@ public final class LeadLauncher {
return started;
}
/** A declared lead paired with the profile it launches on — {@link #resolveLaunchable}'s result. */
private record ResolvedLead(FleetConfig.Leader lead, FleetConfig.Profile profile) {
}
/**
* The declared {@code Leader} and its {@code Profile} for {@code name}, read from the config
* snapshot this launcher was constructed with.
*
* @return the resolved pair, or {@code null} (having logged) if {@code name} is not declared
* under {@code fleet.leaders}, that lead names no {@code profile:} (a {@code tab:}-only,
* recognise-only lead), or its {@code profile:} is not configured. Shared by
* {@link #ensureLeads()} and {@link #relaunch(String)} so the three refusals and their
* wording live in one place.
*/
private ResolvedLead resolveLaunchable(String name) {
FleetConfig.Leader lead = cfg.fleet().leaders().get(name);
if (lead == null) {
log.warn("lead '{}' is not declared under fleet.leaders — not launching", name);
return null;
}
if (!lead.isCreatable()) {
// A lead with a `tab:` but no `profile:` is recognise-only by design: the operator
// opens it by hand. Say so once rather than looking like a silent failure.
log.info("lead '{}' names no profile — it can be recognised but not launched. Add "
+ "`profile:` under fleet.leaders.{} to have fleetd start it.", name, name);
return null;
}
FleetConfig.Profile profile = cfg.profiles().get(lead.profile());
if (profile == null) {
log.warn("lead '{}' names profile '{}', which is not configured — not launching",
name, lead.profile());
return null;
}
return new ResolvedLead(lead, profile);
}
/**
* Start the named lead from the config snapshot this launcher was constructed with — not a
* live read, so a lead's {@code profile:} or {@code tab:} edited in config needs a daemon
* restart to take effect here — outside of {@link #ensureLeads()}'s {@code instances}
* bookkeeping.
*
* @return the started {@link Agent}, or {@code null} if {@code name} is not declared under
* {@code fleet.leaders}, that lead names no {@code profile:} (a {@code tab:}-only,
* recognise-only lead), its {@code profile:} is not configured, or every attempt up to
* {@link #RELAUNCH_ATTEMPTS} failed to start it. Never throws.
*
* <p>Does not count how many instances of this lead are already live. {@link #ensureLeads()}'s
* count exists to avoid starting a second orchestrator; the caller of this method has already
* decided to replace the lead and owns that decision.
*
* <p>Retries the whole launch attempt — not only the {@code agent_name_taken}/
* {@code agent_pane_busy} cases {@link ResilientAgentLaunch} already retries inside one
* {@code agents.start} call — up to {@link #RELAUNCH_ATTEMPTS} times, sleeping via the
* injected sleeper between attempts, and returns the agent from the first attempt that
* succeeds.
*/
public Agent relaunch(String name) {
ResolvedLead resolved = resolveLaunchable(name);
if (resolved == null) {
return null;
}
for (int attempt = 1; attempt <= RELAUNCH_ATTEMPTS; attempt++) {
Agent started = launch(name, resolved.lead(), resolved.profile());
if (started != null) {
return started;
}
if (attempt < RELAUNCH_ATTEMPTS) {
sleeper.run();
}
}
return null;
}
/**
* How many live leads exist per configured name, and which of that name's labelled tabs are
* <em>not</em> live: a running agent in a tab labelled with that lead's exact {@code tab}
@@ -403,17 +306,8 @@ public final class LeadLauncher {
return null;
}
/**
* Start one lead. Returns null (having logged) rather than throwing on any failure.
*
* <p>Goes through the same {@link ResilientAgentLaunch} seam every member spawn uses
* (fleetd #727): the assembled argv is refused outright if it cannot fit the pane line herdr
* types it into, a stale {@code agent_name_taken} (a crashed session's name the registry has
* not yet released) is retried under a fresh per-attempt name rather than refusing the whole
* relaunch, and a seed pane whose shell has not reached its prompt yet ({@code
* agent_pane_busy}) is retried rather than failing on the first miss.
*/
private Agent launch(String name, FleetConfig.Leader lead, FleetConfig.Profile profile) {
/** Start one lead. Returns false (having logged) rather than throwing on any failure. */
private boolean launch(String name, FleetConfig.Leader lead, FleetConfig.Profile profile) {
String label = lead.tabLabel();
String cwd = (lead.cwd() == null || lead.cwd().isBlank())
? System.getProperty("user.dir") : lead.cwd();
@@ -429,12 +323,8 @@ public final class LeadLauncher {
// Same shape as the member launchers: herdr resolves the executable from `kind`, so
// argv[0] (the configured launcher, e.g. `ccs`) is dropped and only the rest is passed.
List<String> argv = leadArgv(profile);
ResilientAgentLaunch.checkFits(profile.profile(), argv);
List<String> args = argv.isEmpty() ? argv : argv.subList(1, argv.size());
Agent started = ResilientAgentLaunch.startUniquelyNamed(agents, herdrKind(profile), args,
tab.rootPaneId(),
attempt -> "lead-" + name + "-" + nameNonce + "-" + nameSeq.incrementAndGet(),
ResilientAgentLaunch.NAME_RETRIES, ResilientAgentLaunch.SHELL_READY_RETRIES, sleeper);
Agent started = agents.start("lead-" + name, herdrKind(profile),
argv.isEmpty() ? argv : argv.subList(1, argv.size()), tab.rootPaneId());
// Label AFTER the start succeeds. A label written before would survive a failed start
// and then read back as a live lead on the next boot, which is the exact staleness the
@@ -444,7 +334,7 @@ public final class LeadLauncher {
log.info("lead '{}' launched: profile={} tab={} pane={} terminal={} label='{}' cwd={}",
name, profile.profile(), tab.tab().tabId(), started.paneId(),
started.terminalId(), label, cwd);
return started;
return true;
} catch (RuntimeException e) {
log.warn("lead '{}' failed to launch on profile '{}': {}",
name, profile.profile(), e.getMessage());
@@ -456,7 +346,7 @@ public final class LeadLauncher {
tab.tab().tabId(), cleanup.getMessage());
}
}
return null;
return false;
}
}
@@ -164,13 +164,7 @@ public final class LeadRollover {
* The handover file's modified time is not after {@link #open}'s request timestamp, or is
* older than {@code maxDocAgeSeconds}.
*/
HANDOVER_STALE,
/**
* This lead terminal already has a roll running: an earlier {@link #confirm} call claimed
* it and that roll's continuation has not released it yet. {@code detail} names the lead
* terminal and the token that holds the claim.
*/
ROLL_ALREADY_RUNNING
HANDOVER_STALE
}
/**
@@ -311,15 +305,6 @@ public final class LeadRollover {
*/
private final Consumer<Runnable> continuationRunner;
private final Map<String, PendingRollover> pending = new ConcurrentHashMap<>();
/**
* Lead terminal → the token of the roll currently holding that terminal exclusive, for
* {@link #confirm}'s single-flight claim. {@link #confirm} claims an entry here with an
* atomic put-if-absent once every other gate has passed, refusing with {@link
* RefusalReason#ROLL_ALREADY_RUNNING} when a claim is already held; {@link #runRollover}
* releases it in a {@code finally}, on both the success and the thrown-exception path. A
* terminal absent from this map has no roll currently in flight for it.
*/
private final Map<String, String> rollingByTerminal = new ConcurrentHashMap<>();
/**
* Finished tokens → what actually happened, for {@link #status}. Bounded by {@link
* #OUTCOME_HISTORY_CAP}, oldest evicted first ({@code removeEldestEntry} on an insertion-order
@@ -502,16 +487,6 @@ public final class LeadRollover {
return docCheck;
}
// Single-flight claim: atomic put-if-absent, taken only after every other gate has
// passed, so a refused confirm() never takes it. A non-null previous value means a
// different, still-running roll already holds this lead terminal.
String holder = rollingByTerminal.putIfAbsent(p.leadTerminal(), token);
if (holder != null) {
return RollDecision.refused(RefusalReason.ROLL_ALREADY_RUNNING,
"lead terminal " + p.leadTerminal() + " already has a roll running under token "
+ holder);
}
// Record IN_PROGRESS BEFORE removing from `pending` — see RollState#IN_PROGRESS and
// OUTCOME_HISTORY_CAP's javadoc. This ordering means `token` is written into `outcomes`
// while it is STILL present in `pending`; status() checks `outcomes` first (see that
@@ -524,23 +499,7 @@ public final class LeadRollover {
pending.remove(token);
log.info("lead-rollover: confirmed token={} lead={} — roll scheduled once the calling turn ends",
token, callerTerminal);
try {
continuationRunner.accept(() -> runRollover(p, cfg));
} catch (RuntimeException e) {
// continuationRunner can reject the hand-off itself (e.g. a bounded executor's
// RejectedExecutionException) before runRollover ever starts, so runRollover's own
// finally — the only other place that releases rollingByTerminal — never runs either.
// Release the claim here and overwrite the IN_PROGRESS entry with a terminal outcome,
// or this lead terminal could never be rolled again and status() would report
// IN_PROGRESS forever for a roll that in fact never started.
log.warn("lead-rollover: continuationRunner rejected token={} lead={}: {} — the roll "
+ "never started; releasing its claim and reporting it as FAILED",
token, callerTerminal, e.toString(), e);
rollingByTerminal.remove(p.leadTerminal(), token);
outcomes.put(token, new RollStatus(RollState.FAILED,
"continuationRunner rejected this roll before it ever started: " + e.toString()
+ " — the roll never ran; open() a fresh rollover request"));
}
continuationRunner.accept(() -> runRollover(p, cfg));
return RollDecision.approved();
}
@@ -580,12 +539,6 @@ public final class LeadRollover {
"the roll's continuation threw " + e.toString() + " — the roll is dead and will "
+ "not retry itself; check the daemon log for the stack trace, then open() "
+ "a fresh rollover request"));
} finally {
// Release the single-flight claim on both the normal return and the thrown-exception
// path above — a release only on success would leave this lead terminal unrollable
// forever after one failure. The conditional two-argument remove only clears the
// entry this roll itself holds, never a different roll's claim on the same terminal.
rollingByTerminal.remove(p.leadTerminal(), p.token());
}
}
@@ -334,7 +334,7 @@ public final class FleetMcp {
* caller explicitly saying so — never by omitting a {@link CallerResolver} the way the old
* {@code callers == null} idiom allowed. {@code callers} itself is required either way: even
* under {@link #UNENFORCED}, the one real {@link CallerResolver} still resolves every caller's
* {@link Principal} (so {@code markTrackedCallerPresent}/{@code recordPrimarySingleton} see a
* {@link Principal} (so {@code markSpawnedMemberPresent}/{@code recordPrimarySingleton} see a
* real identity), and {@link #denyFor} is the only thing that changes.
*/
public enum AuthorizationMode { ENFORCED, UNENFORCED }
@@ -442,10 +442,10 @@ public final class FleetMcp {
// fall back to here — AuthorizationMode governs enforcement, not identity.
Principal p = callers.resolve(req.getRemoteAddr(), req.getRemotePort(),
req.getHeader("Authorization"));
// Guards on the ROLE, not on the terminal being null — this marks presence for
// a worker, an architect, or the unconfigured-pane floor, and excludes a lead
// or a collaborator even though each carries its own pane too.
markTrackedCallerPresent(p, presence);
// CB-532: guard on the ROLE, not on the terminal being null. This excludes a
// lead, which carries its pane too, while including every spawned member role.
// Enrolling a lead would count it as an available member in the roster.
markSpawnedMemberPresent(p, presence);
return McpTransportContext.create(Map.of(
CALLER_TERMINAL, orEmpty(p.terminal()),
CALLER_PID, Long.toString(p.pid()),
@@ -480,7 +480,7 @@ public final class FleetMcp {
// Answering a worker's fleet_ask (CB-205): resolve its blocked question and
// block for the worker's reply as it resumes the same turn. This is the same
// delegation, so ownership is left untouched (CB-548) — never re-recorded.
return answer(messages, turnId, content, timeoutMs(a), caller);
return answer(messages, turnId, content, timeoutMs(a));
}
// CB-548: delegator ownership (which lead's reply nudge this worker routes to,
// CB-532) is recorded only once the send is ACCEPTED — MessageService has won the
@@ -490,8 +490,8 @@ public final class FleetMcp {
Runnable onAccepted = () -> primaryRegistry.recordDelegation(target, caller);
// wait defaults to true (block for the reply); wait:false is fire-and-poll.
return Boolean.FALSE.equals(a.get("wait"))
? sendAsync(messages, target, content, onAccepted, workers.profiles(), caller)
: send(messages, target, content, timeoutMs(a), onAccepted, workers.profiles(), caller);
? sendAsync(messages, target, content, onAccepted, workers.profiles())
: send(messages, target, content, timeoutMs(a), onAccepted, workers.profiles());
};
// fleet_reply's identity is the CONNECTION, never an argument — so the authz check
// is "is this caller a worker at all", and it can only ever reply as itself.
@@ -514,7 +514,7 @@ public final class FleetMcp {
(exchange, req) -> {
McpSchema.CallToolResult denied = deny(exchange, toolAction("fleet_status", req.arguments()), null);
if (denied != null) return denied;
return status(messages, str(req.arguments(), "sessionId"), callerTerminal(exchange));
return status(messages, str(req.arguments(), "sessionId"));
};
BiFunction<McpSyncServerExchange, McpSchema.CallToolRequest, McpSchema.CallToolResult> pollHandler =
(exchange, req) -> {
@@ -524,7 +524,7 @@ public final class FleetMcp {
// The action depends on the ARGUMENTS, not on the tool name -- see pollAction.
McpSchema.CallToolResult denied = deny(exchange, toolAction("fleet_poll", a), target);
if (denied != null) return denied;
return poll(messages, leadChannel, str(a, "ticket"), target, coordId, callerTerminal(exchange));
return poll(messages, leadChannel, str(a, "ticket"), target, coordId);
};
// CB-307 Increment 3: per-msgId ack (not needed in v1 but supported by the inbox).
// Acking removes a reply from the inbox, so it is a drain, not a read.
@@ -560,10 +560,8 @@ public final class FleetMcp {
return listFleet(workers, sessions, messages, capacity, healthCoverage, loopHealth, quarantine, outage,
leadSeats, leadContextGauge, leadConfigDirs, callers.leads(),
callerTerminal(exchange),
callers.collaborators(), collaboratorsVisibleTo(principal(exchange)),
new CoordinationSource(leadChannel, peers),
coordinatorVisibleTo(principal(exchange)),
leadsVisibleTo(principal(exchange)), membersVisibleTo(principal(exchange)));
coordinatorVisibleTo(principal(exchange)));
};
BiFunction<McpSyncServerExchange, McpSchema.CallToolRequest, McpSchema.CallToolResult> stopHandler =
(exchange, req) -> {
@@ -745,53 +743,7 @@ public final class FleetMcp {
return caller.isPrimary();
}
/**
* fleetd #703: who may see {@code fleet_list}'s {@code collaborators} array — a roster of the
* human-opened tabs this daemon recognises as named peers. Visible to exactly the roles that
* may {@link Authz.Action#SEND} to a named peer ({@code Authz.java}'s {@code SEND} case):
* the primary, an architect, and a collaborator (reaching another collaborator or a lead). A
* worker holds {@code READ} but can never {@code SEND} to a collaborator, so listing them to a
* worker would expose which human tabs exist on the host with no use to that caller. Split out
* for the same reason as {@link #coordinatorVisibleTo}: the decision must be unit-testable
* without fabricating an SDK {@code McpSyncServerExchange}, and the handler must call this
* named predicate rather than inlining the check.
*/
static boolean collaboratorsVisibleTo(Principal caller) {
return caller.isPrimary() || caller.isArchitect() || caller.isCollaborator();
}
/**
* Who may see {@code fleet_list}'s {@code leads} array — exactly the roles that may
* {@link Authz.Action#SEND} to a lead: the primary, an architect, and a collaborator. A
* collaborator's own {@code fleet_whoami} carries no lead address, and {@code leads} is the
* only place this tool gives one, so a collaborator needs this array to use the send it
* already holds. A worker can never {@code SEND} at all, so it still sees neither this array
* nor {@code members}; a worker's own facts come from {@code fleet_whoami} instead. Split
* out for the same reason as {@link #coordinatorVisibleTo} and
* {@link #collaboratorsVisibleTo}: the decision must be unit-testable without fabricating an
* SDK {@code McpSyncServerExchange}, and the handler must call this named predicate rather
* than inlining the check.
*/
static boolean leadsVisibleTo(Principal caller) {
return caller.isPrimary() || caller.isArchitect() || caller.isCollaborator();
}
/**
* Who may see {@code fleet_list}'s {@code members} array — the primary and an architect,
* which may {@link Authz.Action#SEND} to a member. A worker can never {@code SEND} at all,
* and a collaborator may {@code SEND} only to a lead or another collaborator, never to a
* spawned member, so neither sees this array even though {@link #leadsVisibleTo} grants a
* collaborator the sibling one.
*/
static boolean membersVisibleTo(Principal caller) {
return caller.isPrimary() || caller.isArchitect();
}
/**
* The terminal of the caller on this call's connection, or {@code null} when that caller carries
* no terminal, which is only the unnamed primary. A named lead, an architect and a worker each
* carry one.
*/
/** The worker identity resolved from this call's connection, or {@code null} if the primary. */
private static String callerTerminal(McpSyncServerExchange exchange) {
Object v = exchange.transportContext().get(CALLER_TERMINAL);
String s = v == null ? null : v.toString();
@@ -833,14 +785,9 @@ public final class FleetMcp {
return identity;
}
/**
* Mark a caller present for the injector readiness gate, when its deliverability depends on
* proving a live MCP contact: a worker, an architect, or the unconfigured-pane floor. A lead
* or a collaborator is excluded — each is already deliverable through its own named-registry
* entry.
*/
static void markTrackedCallerPresent(Principal caller, MemberPresence presence) {
if (caller.isSpawnedMember() || caller.isObserver()) {
/** Mark a connected spawned member available for the injector readiness gate. */
static void markSpawnedMemberPresent(Principal caller, MemberPresence presence) {
if (caller.isSpawnedMember()) {
presence.markPresent(caller.terminal());
}
}
@@ -913,8 +860,7 @@ public final class FleetMcp {
* a BUSY interloper never claims a turn it did not win. {@code null} disables recording.
*/
static McpSchema.CallToolResult send(MessageService messages, String sessionId, String content,
Long timeoutMs, Runnable onAccepted, Set<String> profiles,
String callerTerminal) {
Long timeoutMs, Runnable onAccepted, Set<String> profiles) {
if (isBlank(sessionId) || isBlank(content)) {
return error("sessionId and content are required");
}
@@ -924,7 +870,7 @@ public final class FleetMcp {
}
long timeout = clamp(timeoutMs == null ? DEFAULT_TIMEOUT_MS : timeoutMs);
try {
return formatReply(messages.send(sessionId, content, timeout, onAccepted, callerTerminal), timeout);
return formatReply(messages.send(sessionId, content, timeout, onAccepted), timeout);
} catch (HerdrException e) {
return error("herdr error contacting session " + sessionId + ": " + e.getMessage());
}
@@ -934,15 +880,13 @@ public final class FleetMcp {
* {@code fleet_send} carrying a {@code turnId}: the primary's answer to a worker's
* {@code fleet_ask} (CB-205). Resolves the worker's blocked question and blocks for its reply as
* it resumes the same turn — surfaced to the primary identically to a normal send.
* {@code callerTerminal} must match the turn's recorded owner or this is refused.
*/
static McpSchema.CallToolResult answer(MessageService messages, String turnId, String content, Long timeoutMs,
String callerTerminal) {
static McpSchema.CallToolResult answer(MessageService messages, String turnId, String content, Long timeoutMs) {
if (isBlank(turnId) || isBlank(content)) {
return error("turnId and content are required to answer a worker's question");
}
long timeout = clamp(timeoutMs == null ? DEFAULT_TIMEOUT_MS : timeoutMs);
return formatReply(messages.answer(turnId, content, timeout, callerTerminal), timeout);
return formatReply(messages.answer(turnId, content, timeout), timeout);
}
/**
@@ -991,8 +935,6 @@ public final class FleetMcp {
+ "\" and content set to your answer; the worker resumes the same turn.");
case STALE_TURN -> error("that question is no longer open — it timed out or was already "
+ "answered (turnId stale)");
case NOT_TURN_OWNER -> error("this turn belongs to a different delegation — only the caller "
+ "that opened it may answer it");
case TIMED_OUT_WORKING, TIMED_OUT_QUEUED, BUSY -> text("[no reply within " + timeout + "ms — worker "
+ r.outcome().name().toLowerCase().replace("timed_out_", "") + "; retry or poll status]");
// fleetd #571: delivery is unknown here — agent.prompt pastes and submits in one call,
@@ -1014,17 +956,6 @@ public final class FleetMcp {
*/
static McpSchema.CallToolResult sendAsync(MessageService messages, String sessionId, String content,
Runnable onAccepted, Set<String> profiles) {
return sendAsync(messages, sessionId, content, onAccepted, profiles, null);
}
/**
* As above, recording {@code creatorTerminal} as this ticket's owner (the caller's own
* terminal, resolved from the connection) so a later {@code fleet_poll{ticket}} only hands the
* result back to that same caller — see {@link MessageService#poll(String, String)}.
*/
static McpSchema.CallToolResult sendAsync(MessageService messages, String sessionId, String content,
Runnable onAccepted, Set<String> profiles,
String creatorTerminal) {
if (isBlank(sessionId) || isBlank(content)) {
return error("sessionId and content are required");
}
@@ -1032,7 +963,7 @@ public final class FleetMcp {
if (targetError != null) {
return targetError;
}
String ticket = messages.sendAsync(sessionId, content, onAccepted, creatorTerminal);
String ticket = messages.sendAsync(sessionId, content, onAccepted);
return text("accepted — task delegated. Poll fleet_poll with ticket=" + ticket);
}
@@ -1211,24 +1142,9 @@ public final class FleetMcp {
* exclusively of {@code ticket}/{@code target} — see {@link #pollAction}'s javadoc for why this
* is a different inbox (this daemon's own {@link LeadChannel}) that authorizes differently
* ({@link Authz.Action#COORD_READ}, primary-only) from either of the original two branches.
*
* <p>Does not check who owns {@code ticket} — see the overload that takes {@code
* callerTerminal} for that. Callers that do not resolve a caller terminal (tests, or a surface
* with no connection-based identity) use this one.
*/
static McpSchema.CallToolResult poll(MessageService messages, LeadChannel leadChannel, String ticket,
String target, String coordId) {
return poll(messages, leadChannel, ticket, target, coordId, null);
}
/**
* As above, refusing a ticket lookup whose caller's terminal differs from the terminal that
* created it — see {@link MessageService#poll(String, String)}. {@code callerTerminal} is the
* CALLING session's terminal id, resolved by the MCP layer from the connection, never a
* client-supplied value.
*/
static McpSchema.CallToolResult poll(MessageService messages, LeadChannel leadChannel, String ticket,
String target, String coordId, String callerTerminal) {
if (!isBlank(coordId)) {
return pollHeldPeerMail(leadChannel, coordId);
}
@@ -1242,7 +1158,7 @@ public final class FleetMcp {
if (isBlank(ticket)) {
return error("ticket (or target) is required");
}
MessageService.TaskView v = messages.poll(ticket, callerTerminal);
MessageService.TaskView v = messages.poll(ticket);
if (v == null) {
return error("unknown ticket: " + ticket + " (never issued, or expired)");
}
@@ -1349,20 +1265,16 @@ public final class FleetMcp {
/**
* {@code fleet_status}: the live lifecycle status of a worker session, plus — when the worker
* is paused mid-turn in an async {@code fleet_ask} — the open question and how to answer it, so
* a lead on its normal poll cadence does not need the ticket to notice. The question, its
* {@code turnId} and its ticket id are shown only to the caller whose terminal created that
* delegation, or to a caller with no terminal at all (the unnamed primary); any other caller
* still sees the base status. {@code callerTerminal} is the CALLING session's terminal id,
* resolved by the MCP layer from the connection, never a client-supplied value.
* is paused mid-turn in an async {@code fleet_ask} (CB-582) — the open question and how to
* answer it, so a lead on its normal poll cadence does not need the ticket to notice.
*/
static McpSchema.CallToolResult status(MessageService messages, String sessionId, String callerTerminal) {
static McpSchema.CallToolResult status(MessageService messages, String sessionId) {
if (isBlank(sessionId)) {
return error("sessionId is required");
}
try {
String base = messages.status(sessionId).name().toLowerCase();
MessageService.PendingAsk ask = messages.pendingAsk(sessionId, callerTerminal);
MessageService.PendingAsk ask = messages.pendingAsk(sessionId);
if (ask == null) {
return text(base);
}
@@ -1417,14 +1329,6 @@ public final class FleetMcp {
}
return text(json(m));
}
if (caller.isObserver()) {
// No architect slot, collaborator name, or lead name to report — only the pane itself,
// so a peer that already knows this terminal can still address it.
if (caller.terminal() != null) {
m.put("sessionId", caller.terminal());
}
return text(json(m));
}
if (!caller.isWorker()) {
// CB-530: which lead, once more than one pane is configured as one. `role` deliberately
// still reads "primary" — the fallback ladder in CLAUDE.md keys on it, and a lead IS a
@@ -1857,8 +1761,7 @@ public final class FleetMcp {
Map<String, String> leads, String selfTerm,
CoordinationSource coordination) {
return listFleet(workers, sessions, messages, capacity, healthCoverage, loopHealth, quarantine,
OutageSource.none(), LeadSeatSource.none(), new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm,
Map.of(), false, coordination, false, true, true);
OutageSource.none(), LeadSeatSource.none(), new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm, coordination, false);
}
/** As above, plus fleetd #201 Unit 5 cool-off facts (see {@link OutageSource}). */
@@ -1867,8 +1770,7 @@ public final class FleetMcp {
QuarantineSource quarantine, OutageSource outage,
Map<String, String> leads, String selfTerm) {
return listFleet(workers, sessions, messages, capacity, healthCoverage, LoopHealthSource.none(), quarantine, outage,
LeadSeatSource.none(), new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm,
Map.of(), false, CoordinationSource.none(), false, true, true);
LeadSeatSource.none(), new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm, CoordinationSource.none(), false);
}
/**
@@ -1885,8 +1787,7 @@ public final class FleetMcp {
QuarantineSource quarantine, Map<String, String> leads, String selfTerm,
CoordinationSource coordination) {
return listFleet(workers, sessions, messages, capacity, healthCoverage, LoopHealthSource.none(), quarantine, OutageSource.none(),
LeadSeatSource.none(), new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm,
Map.of(), false, coordination, false, true, true);
LeadSeatSource.none(), new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm, coordination, false);
}
/** As above, plus fleetd #201 Unit 5 cool-off facts (see {@link OutageSource}). */
@@ -1895,8 +1796,7 @@ public final class FleetMcp {
QuarantineSource quarantine, OutageSource outage,
Map<String, String> leads, String selfTerm, CoordinationSource coordination) {
return listFleet(workers, sessions, messages, capacity, healthCoverage, LoopHealthSource.none(), quarantine, outage,
LeadSeatSource.none(), new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm,
Map.of(), false, coordination, false, true, true);
LeadSeatSource.none(), new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm, coordination, false);
}
/**
@@ -1918,8 +1818,7 @@ public final class FleetMcp {
LeadSeatSource leadSeats, Map<String, String> leads, String selfTerm,
CoordinationSource coordination) {
return listFleet(workers, sessions, messages, capacity, healthCoverage, LoopHealthSource.none(), quarantine, outage,
leadSeats, new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm,
Map.of(), false, coordination, false, true, true);
leadSeats, new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm, coordination, false);
}
/**
@@ -1936,23 +1835,14 @@ public final class FleetMcp {
* {@code false} (fleetd #463: a forgotten argument fails closed, not
* open), so a test that wants the {@code coordinator} row must pass
* an explicit {@code true}
* @param leadsVisible whether this caller may see the {@code leads} array (see
* {@link #leadsVisibleTo}) — unlike {@code callerIsPrimary}, this is
* also {@code true} for an architect or a collaborator, so it cannot
* be derived from {@code callerIsPrimary} alone
* @param membersVisible whether this caller may see the {@code members} array (see
* {@link #membersVisibleTo}); also {@code true} for an architect, but
* unlike {@code leadsVisible}, never for a collaborator
*/
static McpSchema.CallToolResult listFleet(PeerLauncher workers, SessionManager sessions, MessageService messages,
CapacitySource capacity, HealthCoverageSource healthCoverage,
QuarantineSource quarantine, OutageSource outage,
LeadSeatSource leadSeats, Map<String, String> leads, String selfTerm,
CoordinationSource coordination, boolean callerIsPrimary,
boolean leadsVisible, boolean membersVisible) {
CoordinationSource coordination, boolean callerIsPrimary) {
return listFleet(workers, sessions, messages, capacity, healthCoverage, LoopHealthSource.none(), quarantine,
outage, leadSeats, new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm,
Map.of(), false, coordination, callerIsPrimary, leadsVisible, membersVisible);
outage, leadSeats, new LeadContextGauge(), LeadConfigDirSource.none(), leads, selfTerm, coordination, callerIsPrimary);
}
/**
@@ -1962,17 +1852,6 @@ public final class FleetMcp {
* matter); the one caller that matters for caching, {@code fleet_list}'s MCP handler, passes
* its own single long-lived instance instead (see {@code FleetMcp}'s {@code leadContextGauge}
* field).
*
* @param collaborators terminal_id → collaborator name, live from the resolver
* ({@link dev.ltms.fleet.auth.CallerResolver#collaborators()})
* @param collaboratorsVisible whether this caller may see the {@code collaborators} array (see
* {@link #collaboratorsVisibleTo}); every wrapper overload above
* passes {@code false}, so a test that wants the row must call this
* overload with an explicit {@code true}
* @param leadsVisible whether this caller may see the {@code leads} array (see
* {@link #leadsVisibleTo})
* @param membersVisible whether this caller may see the {@code members} array (see
* {@link #membersVisibleTo})
*/
static McpSchema.CallToolResult listFleet(PeerLauncher workers, SessionManager sessions, MessageService messages,
CapacitySource capacity, HealthCoverageSource healthCoverage,
@@ -1981,56 +1860,32 @@ public final class FleetMcp {
LeadSeatSource leadSeats, LeadContextGauge contextGauge,
LeadConfigDirSource leadConfigDirs,
Map<String, String> leads, String selfTerm,
Map<String, String> collaborators, boolean collaboratorsVisible,
CoordinationSource coordination, boolean callerIsPrimary,
boolean leadsVisible, boolean membersVisible) {
CoordinationSource coordination, boolean callerIsPrimary) {
try {
// Neither row's assembly (leadView/memberCapacityView probing herdr for live status)
// runs unless at least one of them needs the live-agent lookup backing it.
Map<String, Agent> live = (leadsVisible || membersVisible)
? workers.list().stream()
.map(Agent.class::cast)
.filter(a -> a.terminalId() != null)
.collect(Collectors.toMap(Agent::terminalId, Function.identity(), (_, b) -> b))
: Map.of();
Map<String, Agent> live = workers.list().stream()
.map(Agent.class::cast)
.filter(a -> a.terminalId() != null)
.collect(Collectors.toMap(Agent::terminalId, Function.identity(), (_, b) -> b));
List<Map<String, Object>> leadRows = leads.entrySet().stream()
.sorted(Map.Entry.comparingByValue())
.map(e -> leadView(e.getKey(), e.getValue(), live.get(e.getKey()), selfTerm, contextGauge,
leadConfigDirs))
.toList();
// fleetd #209: this is the caller-driven fleet_list read that actually reports
// agentSessionId (via memberCapacityView -> SessionManager.rosterView), so it uses the
// resolving roster; the heartbeat/health/metrics timers stay on the plain sessions.roster().
List<MemberSession> roster = sessions.rosterResolved();
List<Map<String, Object>> out = roster.stream()
.map(s -> memberCapacityView(s, live.get(s.terminalId()), messages, capacity.clock().getAsLong()))
.toList();
Set<String> profiles = new java.util.TreeSet<>(capacity.configuredProfiles().get());
roster.stream().map(MemberSession::profile).forEach(profiles::add);
Map<String, Object> result = new LinkedHashMap<>();
// READ is permission to enter this tool, not permission to receive every field it can
// build -- gate BEFORE assembling each row, so the key is absent rather than
// present-and-empty; a caller without either row gets its own facts from fleet_whoami.
if (leadsVisible) {
List<Map<String, Object>> leadRows = leads.entrySet().stream()
.sorted(Map.Entry.comparingByValue())
.map(e -> leadView(e.getKey(), e.getValue(), live.get(e.getKey()), selfTerm, contextGauge,
leadConfigDirs))
.toList();
result.put("leads", leadRows);
}
if (membersVisible) {
List<Map<String, Object>> out = roster.stream()
.map(s -> memberCapacityView(s, live.get(s.terminalId()), messages, capacity.clock().getAsLong()))
.toList();
result.put("members", out);
}
result.put("leads", leadRows); result.put("members", out);
result.put("healthCoverage", healthCoverage.value().get());
result.put("loopHealth", Map.of(
"statusPoller", loopHealth.statusPoller().get().name(),
"sessionReaper", loopHealth.sessionReaper().get().name()));
// fleetd #703: a collaborator tab is a person's own tab, so the row is assembled and
// included only for the roles that may SEND to a named peer -- gate BEFORE assembling
// it, same reason as the coordinator row just below: the key must be absent for a
// worker, never present-and-empty.
if (collaboratorsVisible) {
result.put("collaborators", collaborators.entrySet().stream()
.sorted(Map.Entry.comparingByValue())
.map(e -> collaboratorRow(e.getKey(), e.getValue()))
.toList());
}
// fleetd #439: coordinator/coordinatorView is lead-to-lead coordination state and must
// never reach a worker or an architect -- gate BEFORE assembling it, not after, so the
// key is absent rather than present-and-empty.
@@ -2342,20 +2197,6 @@ public final class FleetMcp {
return m;
}
/**
* fleetd #703: one row of the {@code collaborators} array — a named peer's registry name and
* the herdr {@code terminal_id} a {@code fleet_send} must target to reach it. No status, no
* context gauge, no profile: a collaborator is never spawned and carries no profile, so those
* fields have no meaning for it, and the scan behind {@code terminal} only reports a tab it
* actually found in herdr, so a tab nobody has open does not appear here at all.
*/
private static Map<String, Object> collaboratorRow(String terminal, String name) {
Map<String, Object> m = new LinkedHashMap<>();
m.put("name", name);
m.put("sessionId", terminal);
return m;
}
/**
* Reads the lead context gauge for one lead. {@code configDir} is this lead's configured
* {@code CLAUDE_CONFIG_DIR} override (see {@link LeadConfigDirSource}), derived from
@@ -2550,14 +2391,7 @@ public final class FleetMcp {
private static McpSchema.Tool listTool() {
return tool(FleetTool.LIST.wireName(),
"List the whole fleet the bridge tracks, in two parts. 'members' is visible to "
+ "the primary and an architect only. 'leads' is visible to those two AND a "
+ "collaborator — exactly the roles that may fleet_send to a lead, so a "
+ "collaborator can learn a lead's sessionId before using the send it already "
+ "holds. A worker holds READ to call this tool at all, but gets neither "
+ "array, never an empty one; a worker reads its own session, "
+ "profile, state, worktree, branch and owner from fleet_whoami instead. "
+ "'leads' are your PEERS — other "
"List the whole fleet the bridge tracks, in two parts. 'leads' are your PEERS — other "
+ "orchestrators, each with its sessionId (the address to fleet_send to), "
+ "name, live status, and 'self': true on your own row; this is how you "
+ "discover a peer lead without being told its address. 'members' are the "
@@ -2588,12 +2422,7 @@ public final class FleetMcp {
+ "msgId/from/preview, never the full body), and one row per coordinator.peers "
+ "coord-id ('peers': coordId/reachable, plus pending/consumers when reachable) — "
+ "this is peer DISCOVERY for cross-host leads, distinct from the local 'leads' "
+ "array above. It is omitted entirely when no coordinator is configured. A "
+ "'collaborators' array, visible only to the primary, an architect, and a "
+ "collaborator (never a worker), reports every other named peer tab this daemon "
+ "recognises: each row is 'name' (its registry name) and 'sessionId' (the "
+ "terminal id a fleet_send targets to reach it). Absent entirely for a worker, "
+ "whatever is configured.",
+ "array above. It is omitted entirely when no coordinator is configured.",
objectSchema(Map.of(), List.of()));
}
@@ -6,7 +6,6 @@ import dev.ltms.fleet.herdr.AgentControl;
import dev.ltms.fleet.herdr.AgentStatus;
import dev.ltms.fleet.herdr.HerdrClient;
import dev.ltms.fleet.herdr.HerdrException;
import dev.ltms.fleet.herdr.ResilientAgentLaunch;
import dev.ltms.fleet.herdr.Tab;
import dev.ltms.fleet.herdr.Workspace;
import dev.ltms.fleet.herdr.WorkspaceControl;
@@ -68,6 +67,16 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
private static final Logger log = LoggerFactory.getLogger(HerdrPeerLauncher.class);
/** herdr rejects a duplicate agent {@code name}; we retry a bumped name this many times. */
private static final int NAME_RETRIES = 8;
/**
* Retries for {@code agent.start} against a seed pane whose shell has not reached its prompt
* yet — {@code tab.create}/{@code pane.split} return as soon as the pane exists, and herdr
* refuses to start an agent in a pane that is not "an available shell" ({@code agent_pane_busy}).
*/
private static final int SHELL_READY_RETRIES = 20;
private final String namePrefix; // label prefix: naming + reap scheme
private final AgentControl agents;
private final WorkspaceControl spaces;
@@ -757,26 +766,90 @@ public abstract class HerdrPeerLauncher implements PeerLauncher {
// Protocol 19 resolves the executable from the agent kind (== namePrefix here), so
// argv[0] — the configured executable — is dropped and only the extra args are passed.
List<String> args = argv.isEmpty() ? argv : argv.subList(1, argv.size());
try {
ResilientAgentLaunch.checkFits(cfg.profile(), argv);
} catch (ResilientAgentLaunch.TooLargeException e) {
throw new PeerUnreachableException(e.getMessage());
checkPaneCommandFits(cfg, argv);
HerdrException last = null;
for (int attempt = 0; attempt < NAME_RETRIES; attempt++) {
long seq = nameSeq.incrementAndGet();
String name = namePrefix + "-" + cfg.profile() + "-" + nameNonce + "-" + seq;
try {
return new Started(startAwaitingShellPrompt(name, args, paneId), seq);
} catch (HerdrException e) {
if (!"agent_name_taken".equals(e.code())) throw e;
log.debug("peer name '{}' taken, retrying", name);
last = e;
}
}
long[] lastSeq = {0};
Agent agent = ResilientAgentLaunch.startUniquelyNamed(agents, namePrefix, args, paneId,
attempt -> {
lastSeq[0] = nameSeq.incrementAndGet();
return namePrefix + "-" + cfg.profile() + "-" + nameNonce + "-" + lastSeq[0];
},
ResilientAgentLaunch.NAME_RETRIES, ResilientAgentLaunch.SHELL_READY_RETRIES, sleeper);
return new Started(agent, lastSeq[0]);
throw last;
}
/**
* fleetd #220: herdr does not exec the launch command — it TYPES it into the pane as one line,
* and a pty line buffer holds only {@value #PANE_COMMAND_BYTE_LIMIT} bytes (BSD/macOS {@code
* MAX_CANON}). Everything past that byte is dropped. Nothing reports it: herdr answers "agent
* started", the backend exits on the mangled argument it was handed, the pane closes, and the
* only symptom is {@link #waitUntilInjectableOrThrow} timing out 20 seconds later with no
* reason. That is exactly how #214 broke every claude-code spawn — one 50-byte flag pushed a
* 978-byte command to 1028, and the tail that got cut was {@code --autocompact 250000}.
*
* <p>So measure it here and refuse, loudly and immediately, rather than spawn something that
* cannot work. The estimate is deliberately conservative: fleetd cannot see herdr's quoting, so
* every argument is charged its own bytes plus a separator and a quote pair. An over-estimate
* costs a clear error at a length that was already unsafe; an under-estimate would let the
* silent truncation back in.
*
* @throws PeerUnreachableException when the command cannot fit — the same failure the spawn
* would have hit anyway, named at the point it is still
* explainable
*/
private void checkPaneCommandFits(FleetConfig.Profile cfg, List<String> argv) {
int bytes = 0;
String longest = null;
int longestBytes = 0;
for (String arg : argv) {
int argBytes = arg == null ? 0 : arg.getBytes(java.nio.charset.StandardCharsets.UTF_8).length;
bytes += argBytes + QUOTING_OVERHEAD_PER_ARG;
if (argBytes > longestBytes) {
longestBytes = argBytes;
longest = arg;
}
}
if (bytes <= PANE_COMMAND_BYTE_LIMIT) {
return;
}
String culprit = longest == null ? "<none>"
: longest.substring(0, Math.min(longest.length(), 60)) + (longest.length() > 60 ? "…" : "");
throw new PeerUnreachableException(
"launch command for profile " + cfg.profile() + " is about " + bytes + " bytes, over the "
+ PANE_COMMAND_BYTE_LIMIT + "-byte limit of the pane line herdr types it into. "
+ "The pty would drop the tail silently and the backend would exit on a mangled "
+ "argument. Longest argument is " + longestBytes + " bytes: " + culprit
+ " — move it off the command line (a file flag) or shorten it.");
}
/**
* The pty line buffer herdr types a launch command into: BSD/macOS {@code MAX_CANON}. Not a
* fleetd choice and not configurable — see {@link ResilientAgentLaunch#checkFits}.
* fleetd choice and not configurable — see {@link #checkPaneCommandFits}.
*/
static final int PANE_COMMAND_BYTE_LIMIT = ResilientAgentLaunch.PANE_COMMAND_BYTE_LIMIT;
static final int PANE_COMMAND_BYTE_LIMIT = 1024;
/** Per-argument allowance for the separating space and a shell quote pair fleetd cannot see. */
private static final int QUOTING_OVERHEAD_PER_ARG = 3;
/** Start the agent into {@code paneId}, waiting out the seed shell's boot with the sleeper. */
private Agent startAwaitingShellPrompt(String name, List<String> args, String paneId) {
HerdrException busy = null;
for (int attempt = 0; attempt < SHELL_READY_RETRIES; attempt++) {
try {
return agents.start(name, namePrefix, args, paneId);
} catch (HerdrException e) {
if (!"agent_pane_busy".equals(e.code())) throw e;
log.debug("pane {} not at its shell prompt yet, retrying agent.start", paneId);
busy = e;
sleeper.run();
}
}
throw busy;
}
// --- discovery + reap ----------------------------------------------------------------------
@@ -87,7 +87,7 @@ public final class MessageService {
/**
* The worker paused mid-turn to ask the primary a question (CB-205); {@code text} is the
* question and {@code turnId} correlates the answer. Not terminal — the primary answers with
* {@link #answer(String, String, long, String)} and the turn resumes.
* {@link #answer(String, String, long)} and the turn resumes.
*/
QUESTION,
/** Timed out after the message was delivered — the worker is still working. */
@@ -120,16 +120,10 @@ public final class MessageService {
/** Another send to this session was in flight for the whole window. */
BUSY,
/**
* An answer ({@link #answer(String, String, long, String)}) referenced a {@code turnId}
* that is no longer open — the worker's {@code fleet_ask} already timed out or was answered.
* An answer ({@link #answer(String, String, long)}) referenced a {@code turnId} that is no
* longer open — the worker's {@code fleet_ask} already timed out or was answered.
*/
STALE_TURN,
/**
* An answer ({@link #answer(String, String, long, String)}) named a {@code turnId} that is
* still open, but the answering caller is not the caller whose accepted delegation opened
* it. Distinct from {@link #STALE_TURN} so a refusal is never reported as a lapsed turn.
*/
NOT_TURN_OWNER
STALE_TURN
}
/**
@@ -139,7 +133,7 @@ public final class MessageService {
* {@link Outcome#COMPLETED_UNREPLIED}), or the question for {@link Outcome#QUESTION},
* else {@code null}
* @param turnId correlation id for a {@link Outcome#QUESTION} (answered via
* {@link #answer(String, String, long, String)}), else {@code null}
* {@link #answer(String, String, long)}), else {@code null}
*/
public record Reply(Outcome outcome, String text, String turnId) {
/** A reply with no correlation id (the common terminal outcomes). */
@@ -281,19 +275,10 @@ public final class MessageService {
* {@link #abandon}) can never match again regardless of this flag's value.
*/
private volatile boolean askTimedOut;
/**
* The terminal of the caller whose {@code fleet_send{wait:false}} created this ticket, or
* {@code null} when that caller had no terminal (the unnamed primary) or the ticket was
* created through an overload that does not record one. {@link #poll(String, String)}
* compares a polling caller's own terminal against this field before handing back the
* ticket's state.
*/
private final String creatorTerminal;
private Task(String ticket, String target, LongSupplier nowNanos, String creatorTerminal) {
private Task(String ticket, String target, LongSupplier nowNanos) {
this.ticket = ticket;
this.target = target;
this.creatorTerminal = creatorTerminal;
this.createdNanos = nowNanos.getAsLong();
future.whenComplete((reply, ex) -> completedNanos = nowNanos.getAsLong());
}
@@ -355,14 +340,6 @@ public final class MessageService {
*/
private final ConcurrentHashMap<String, Boolean> queuedDeliveries = new ConcurrentHashMap<>();
private final AtomicLong ticketSeq = new AtomicLong();
/**
* Minted once per {@code MessageService} instance and folded into every ticket id (see
* {@link #sendAsync(String, String, Runnable, String)}). {@link #ticketSeq} alone restarts at
* zero for every instance, so without this a ticket id can be reused across instances and
* resolve to an unrelated {@link Task} with no error; this nonce makes that impossible, because
* an id minted by one instance can never match the id space of another.
*/
private final String ticketBootNonce = UUID.randomUUID().toString().substring(0, 6);
private final ExecutorService asyncExecutor = Executors.newThreadPerTaskExecutor(
Thread.ofVirtual().name("bridge-async-", 0).factory());
@@ -679,7 +656,7 @@ public final class MessageService {
case TIMED_OUT_WORKING, TIMED_OUT_QUEUED, TIMED_OUT_UNCONFIRMED, BUSY -> "timeout";
case WORKER_FAILED -> "failed";
case BACKEND_EXHAUSTED -> "backend_exhausted";
case STALE_TURN, QUESTION, NOT_TURN_OWNER -> null; // not a completed delegation
case STALE_TURN, QUESTION -> null; // not a completed delegation
};
}
@@ -938,17 +915,14 @@ public final class MessageService {
/**
* Deliver {@code content} to {@code target} (a herdr {@code terminal_id}) and block until the
* worker replies via {@link Rendezvous} or {@code timeoutMillis} elapses. {@code callerTerminal}
* is the terminal of the caller making this call — {@code null} for the unnamed primary — and is
* recorded as the turn's owner, the only caller {@link #answer(String, String, long, String)} will
* later accept an answer from if the worker pauses mid-turn to ask.
* worker replies via {@link Rendezvous} or {@code timeoutMillis} elapses.
*/
public Reply send(String target, String content, long timeoutMillis, String callerTerminal) {
return send(target, content, timeoutMillis, null, callerTerminal);
public Reply send(String target, String content, long timeoutMillis) {
return send(target, content, timeoutMillis, null);
}
/**
* As {@link #send(String, String, long, String)}, but with an accepted-delivery hook.
* As {@link #send(String, String, long)}, but with an accepted-delivery hook.
*
* <p>{@code onAccepted} is invoked exactly once, once this send has won {@code target}'s send
* lock and so become the <em>accepted target turn</em> — it runs <em>before</em> delivery is
@@ -959,13 +933,12 @@ public final class MessageService {
* acceptance means a concurrent sender that times out {@code BUSY} can never steal ownership it
* never earned. {@code null} disables the hook.
*/
public Reply send(String target, String content, long timeoutMillis, Runnable onAccepted, String callerTerminal) {
return send(target, content, timeoutMillis, onAccepted, null, callerTerminal);
public Reply send(String target, String content, long timeoutMillis, Runnable onAccepted) {
return send(target, content, timeoutMillis, onAccepted, null);
}
/** Run a send, optionally stopping an async task that teardown already failed before acceptance. */
private Reply send(String target, String content, long timeoutMillis, Runnable onAccepted, Task task,
String callerTerminal) {
private Reply send(String target, String content, long timeoutMillis, Runnable onAccepted, Task task) {
long deadlineNanos = System.nanoTime() + timeoutMillis * 1_000_000L;
ReentrantLock lock = sessionLocks.computeIfAbsent(target, _ -> new ReentrantLock());
@@ -985,7 +958,7 @@ public final class MessageService {
// open race). Opening first also means a throwing onAccepted (fired before enqueue) or an
// enqueue failure is safely closed by the finally below: nothing is left queued, and the
// failed send leaves no stale waiter behind.
CompletableFuture<Rendezvous.Resolution> reply = rendezvous.open(target, Rendezvous.Owner.of(callerTerminal));
CompletableFuture<Rendezvous.Resolution> reply = rendezvous.open(target);
// CB-640: this send now owns target's delivery, so any earlier stranded-reply or
// still-queued fact no longer describes the live state — clear both rather than let
// them outlive the send that supersedes them.
@@ -1164,30 +1137,19 @@ public final class MessageService {
* mid-turn (already picked up), so the answer flows back through its own open {@code fleet_ask}
* call, not a new status-gated delivery. The forward waiter is opened <em>before</em> the worker
* is unblocked so a reply that lands the instant it resumes is not lost.
*
* <p>{@code callerTerminal} is the terminal of the caller making this call — {@code null} for
* the unnamed primary. It is checked against the turn's recorded owner (the caller whose
* accepted delegation opened it, see {@link #send(String, String, long, String)} and
* {@link #sendAsync(String, String, Runnable, String)}) before anything else runs: a mismatch,
* including a turn with no owner on record at all, returns {@link Outcome#NOT_TURN_OWNER}
* without touching the rendezvous, the session lock, or any async task bookkeeping.
*/
public Reply answer(String turnId, String content, long timeoutMillis, String callerTerminal) {
public Reply answer(String turnId, String content, long timeoutMillis) {
String workerSession = rendezvous.askSession(turnId);
if (workerSession == null) {
return new Reply(Outcome.STALE_TURN, null); // the ask lapsed (timed out or already answered)
}
Rendezvous.Owner owner = rendezvous.askOwner(turnId);
if (!Rendezvous.Owner.permits(owner, callerTerminal)) {
return new Reply(Outcome.NOT_TURN_OWNER, null);
}
long deadlineNanos = System.nanoTime() + timeoutMillis * 1_000_000L;
ReentrantLock lock = sessionLocks.computeIfAbsent(workerSession, _ -> new ReentrantLock());
if (!tryLock(lock, remainingMillis(deadlineNanos))) {
return new Reply(Outcome.BUSY, null);
}
try {
CompletableFuture<Rendezvous.Resolution> reply = rendezvous.open(workerSession, owner);
CompletableFuture<Rendezvous.Resolution> reply = rendezvous.open(workerSession);
// fleetd #575: this try used to open below, AFTER the Task lookup/registration and the
// STALE_TURN early return that follows it — so that return was covered only by a
// hand-rolled copy of the finally's own cleanup pair, not the finally itself. Widening the
@@ -1317,20 +1279,8 @@ public final class MessageService {
* @return the ticket to poll for the eventual result
*/
public String sendAsync(String target, String content, Runnable onAccepted) {
return sendAsync(target, content, onAccepted, null);
}
/**
* As {@link #sendAsync(String, String, Runnable)}, recording {@code creatorTerminal} as this
* ticket's owner. {@link #poll(String, String)} refuses a later caller whose own terminal
* differs from this one; {@code null} records no owner (a caller with no terminal — the
* unnamed primary — is always allowed to poll the result regardless).
*
* @return the ticket to poll for the eventual result
*/
public String sendAsync(String target, String content, Runnable onAccepted, String creatorTerminal) {
String ticket = "task-" + ticketBootNonce + "-" + ticketSeq.incrementAndGet();
Task task = new Task(ticket, target, nowNanos, creatorTerminal);
String ticket = "task-" + ticketSeq.incrementAndGet();
Task task = new Task(ticket, target, nowNanos);
tasks.put(ticket, task);
if (pushLoop != null) {
// CB-588: task.future only ever completes on a terminal phase (DONE or a failure) — a
@@ -1361,7 +1311,7 @@ public final class MessageService {
}
asyncExecutor.submit(() -> {
try {
Reply result = send(target, content, ASYNC_TIMEOUT_MS, onAccepted, task, creatorTerminal);
Reply result = send(target, content, ASYNC_TIMEOUT_MS, onAccepted, task);
if (result.outcome() == Outcome.QUESTION) {
// Keep the accepted owner until answer() finishes it. markAsyncQuestion may run
// just after resolveQuestion wakes this thread.
@@ -1393,31 +1343,15 @@ public final class MessageService {
}
/**
* As {@link #poll(String, String)}, with no caller terminal — the ticket's ownership is never
* checked, so this overload must only be used where the caller's identity is otherwise
* irrelevant (a test, or a surface that does not resolve a caller terminal at all).
*/
public TaskView poll(String ticket) {
return poll(ticket, null);
}
/**
* Snapshot the state of an async delegation. Returns {@code null} for an unknown/expired ticket.
* Refuses a {@code callerTerminal} that differs from the terminal that created the ticket (see
* {@link #sendAsync(String, String, Runnable, String)}) with a {@link Phase#FAILED} view that
* carries no reply text — a caller with no terminal (the unnamed primary) is never refused.
* Otherwise returns a {@link Phase#PENDING} view (with the live worker status as detail), a
* Snapshot the state of an async delegation. Returns {@code null} for an unknown/expired ticket;
* otherwise a {@link Phase#PENDING} view (with the live worker status as detail), a
* {@link Phase#DONE} view carrying the reply, or a {@link Phase#FAILED} view with the reason.
*/
public TaskView poll(String ticket, String callerTerminal) {
public TaskView poll(String ticket) {
Task task = tasks.get(ticket);
if (task == null) {
return null;
}
if (!ownsTicket(task, callerTerminal)) {
return new TaskView(ticket, Phase.FAILED, null, null,
"forbidden: this ticket was created by a different session", null);
}
CompletableFuture<Reply> f = task.future;
if (!f.isDone()) {
Reply question = task.question;
@@ -1453,17 +1387,6 @@ public final class MessageService {
return new TaskView(ticket, Phase.FAILED, null, null, detail, null);
}
/**
* Whether {@code callerTerminal} may read {@code task}'s state. A caller with no terminal
* always may — that is the unnamed primary, resolved by token or loopback trust, which never
* carries a herdr pane and must keep reading every ticket. Otherwise the caller's terminal must
* equal the terminal recorded on the task; a task with no recorded terminal matches no
* terminal-bearing caller.
*/
private static boolean ownsTicket(Task task, String callerTerminal) {
return callerTerminal == null || callerTerminal.equals(task.creatorTerminal);
}
/**
* Test seam only — carries no production behaviour, and nothing in this class calls it;
* {@link #pruneTerminalTickets} still reads {@link Task#completedNanos} directly.
@@ -1783,17 +1706,16 @@ public final class MessageService {
}
/**
* The question {@code workerSession} is currently paused on via {@code fleet_ask}, if any —
* {@code fleet_status} uses this to show a pending question without the caller needing the
* ticket. {@code null} when the session has no open async question (including a session mid a
* <em>blocking</em> {@code fleet_ask}, which has no {@link Task} to look up — see
* {@link PendingAsk}), or when {@code callerTerminal} does not own the task the question
* belongs to (see {@link #ownsTicket(Task, String)}).
* The question {@code workerSession} is currently paused on via {@code fleet_ask}, if any
* (CB-582) — {@code fleet_status} uses this to show a pending question without the caller
* needing the ticket. {@code null} when the session has no open async question (including a
* session mid a <em>blocking</em> {@code fleet_ask}, which has no {@link Task} to look up — see
* {@link PendingAsk}).
*/
public PendingAsk pendingAsk(String workerSession, String callerTerminal) {
public PendingAsk pendingAsk(String workerSession) {
for (Task task : tasks.values()) {
Reply q = task.question;
if (q != null && workerSession.equals(task.target) && ownsTicket(task, callerTerminal)) {
if (q != null && workerSession.equals(task.target)) {
return new PendingAsk(task.ticket, q.text(), q.turnId());
}
}
@@ -1,6 +1,5 @@
package dev.ltms.fleet.msg;
import java.util.UUID;
import java.util.concurrent.CompletableFuture;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.atomic.AtomicLong;
@@ -61,7 +60,7 @@ public final class Rendezvous {
}
/** A worker's open mid-turn question: the worker session it belongs to and the answer future. */
private record AskWaiter(String session, CompletableFuture<String> answer, Owner owner) {
private record AskWaiter(String session, CompletableFuture<String> answer) {
}
/**
@@ -71,45 +70,11 @@ public final class Rendezvous {
public record AskTicket(String turnId, CompletableFuture<String> answer, boolean fresh) {
}
/**
* The caller whose accepted delegation opened a turn — the only caller allowed to answer it.
* A {@code null} terminal means the unnamed primary, an authenticated caller with no pane.
*/
public record Owner(String terminal) {
public static final Owner UNNAMED_PRIMARY = new Owner(null);
public static Owner of(String terminal) {
return terminal == null ? UNNAMED_PRIMARY : new Owner(terminal);
}
/**
* Whether {@code callerTerminal} matches {@code owner}. A {@code null} owner means no
* owner was ever recorded, and that state matches no caller, not even one whose own
* terminal is {@code null} — "no record" and "recorded as the unnamed primary" are
* different states.
*/
public static boolean permits(Owner owner, String callerTerminal) {
return owner != null && java.util.Objects.equals(owner.terminal(), callerTerminal);
}
}
/** A registered forward waiter together with the owner its delegation was opened under. */
private record ForwardWaiter(Owner owner, CompletableFuture<Resolution> future) {
}
private final ConcurrentHashMap<String, ForwardWaiter> waiters = new ConcurrentHashMap<>();
private final ConcurrentHashMap<String, CompletableFuture<Resolution>> waiters = new ConcurrentHashMap<>();
/** Reverse rendezvous (CB-205): worker questions awaiting the primary's answer, keyed by {@code turnId}. */
private final ConcurrentHashMap<String, AskWaiter> asks = new ConcurrentHashMap<>();
private final AtomicLong askSeq = new AtomicLong();
/**
* Minted once per {@code Rendezvous} instance and folded into every {@code turnId} (see
* {@link #openAsk(String)}). {@link #askSeq} alone restarts at zero for every instance, so
* without this a {@code turnId} minted by one instance could be minted again by another and
* resolve to an unrelated ask with no error; this nonce makes that impossible, because an id
* minted by one instance can never match the id space of another.
*/
private final String askBootNonce = UUID.randomUUID().toString().substring(0, 6);
/** Per-session index of the currently-open ask, so duplicate fleet_ask calls coalesce onto one turn. */
private final ConcurrentHashMap<String, String> openAsksBySession = new ConcurrentHashMap<>();
@@ -124,17 +89,8 @@ public final class Rendezvous {
* code a double open is impossible; this is a tripwire for the day that no longer holds.
*/
public CompletableFuture<Resolution> open(String session) {
return open(session, null);
}
/**
* Same as {@link #open(String)}, additionally recording {@code owner} as the caller whose
* delegation opened this waiter. A {@code null} owner records no owner at all — the
* fail-closed default {@link Owner#permits} refuses to everyone.
*/
public CompletableFuture<Resolution> open(String session, Owner owner) {
CompletableFuture<Resolution> waiter = new CompletableFuture<>();
ForwardWaiter existing = waiters.putIfAbsent(session, new ForwardWaiter(owner, waiter));
CompletableFuture<Resolution> existing = waiters.putIfAbsent(session, waiter);
if (existing != null) {
throw new IllegalStateException(
"rendezvous double-open for session " + session + " — a waiter is already registered");
@@ -149,13 +105,7 @@ public final class Rendezvous {
* successful {@code open} after a finished turn requires this close to have happened first).
*/
public void close(String session, CompletableFuture<Resolution> waiter) {
waiters.computeIfPresent(session, (s, w) -> w.future() == waiter ? null : w);
}
/** The owner recorded for {@code session}'s open waiter, or {@code null} if none is open. */
public Owner ownerOf(String session) {
ForwardWaiter w = waiters.get(session);
return w == null ? null : w.owner();
waiters.remove(session, waiter);
}
/** Whether a send is currently awaiting a resolution for {@code session}. */
@@ -169,8 +119,7 @@ public final class Rendezvous {
* send (see the CB-116 note above) rather than whichever send happens to be waiting when they fire.
*/
public CompletableFuture<Resolution> currentWaiter(String session) {
ForwardWaiter w = waiters.get(session);
return w == null ? null : w.future();
return waiters.get(session);
}
/**
@@ -196,9 +145,9 @@ public final class Rendezvous {
while (true) {
AskWaiter[] minted = { null };
String turnId = openAsksBySession.computeIfAbsent(session, _ -> {
String newTurnId = session + "#" + askBootNonce + "-" + askSeq.incrementAndGet();
String newTurnId = session + "#" + askSeq.incrementAndGet();
CompletableFuture<String> answer = new CompletableFuture<>();
AskWaiter waiter = new AskWaiter(session, answer, ownerOf(session));
AskWaiter waiter = new AskWaiter(session, answer);
asks.put(newTurnId, waiter);
minted[0] = waiter;
return newTurnId;
@@ -234,16 +183,6 @@ public final class Rendezvous {
return w == null ? null : w.session();
}
/**
* The owner recorded for {@code turnId} when its ask turn was freshly opened — the caller
* whose delegation {@link #answerAsk} must match. {@code null} if {@code turnId} is unknown or
* lapsed, or if the ask opened with no forward waiter owner on record.
*/
public Owner askOwner(String turnId) {
AskWaiter w = asks.get(turnId);
return w == null ? null : w.owner();
}
/**
* Resolve a worker's blocked {@code fleet_ask} with the primary's {@code answer}, unblocking it
* to resume its turn.
@@ -306,7 +245,7 @@ public final class Rendezvous {
}
private boolean complete(String session, Resolution resolution) {
ForwardWaiter waiter = waiters.get(session);
return waiter != null && waiter.future().complete(resolution);
CompletableFuture<Resolution> waiter = waiters.get(session);
return waiter != null && waiter.complete(resolution);
}
}
@@ -663,8 +663,6 @@ public final class FleetApp {
if (!allow(ctx, routeAction("POST /sessions/{id}/message"), id)) {
return;
}
Principal caller = ctx.attribute(CALLER);
String callerTerminal = caller == null ? null : caller.terminal();
JsonNode body;
try {
body = mapper.readTree(ctx.body());
@@ -690,19 +688,19 @@ public final class FleetApp {
// Answering a worker's fleet_ask (CB-205): always blocks, and derives the worker from turnId.
if (turnId != null && !turnId.isBlank()) {
writeReply(ctx, id, messages.answer(turnId, content, timeout, callerTerminal), timeout);
writeReply(ctx, id, messages.answer(turnId, content, timeout), timeout);
return;
}
if (!wait) {
// Fire-and-poll (CB-107): return a ticket immediately; the caller polls GET /tasks/{ticket}.
String ticket = messages.sendAsync(id, content, null, callerTerminal);
String ticket = messages.sendAsync(id, content);
ctx.status(202).json(Map.of("sessionId", id, "ticket", ticket, "status", "accepted"));
return;
}
try {
writeReply(ctx, id, messages.send(id, content, timeout, callerTerminal), timeout);
writeReply(ctx, id, messages.send(id, content, timeout), timeout);
} catch (HerdrException e) {
herdrError(ctx, e);
}
@@ -721,10 +719,6 @@ public final class FleetApp {
case STALE_TURN -> ctx.status(409).json(Map.of(
"sessionId", id, "error", "stale_turn",
"detail", "that question is no longer open (timed out or already answered)"));
case NOT_TURN_OWNER -> ctx.status(403).json(Map.of(
"sessionId", id, "error", "not_turn_owner",
"detail", "this turn belongs to a different delegation — only the caller that "
+ "opened it may answer it"));
case REPLIED, COMPLETED_UNREPLIED -> {
// replySource distinguishes a structured fleet_reply from the CB-106 completion
// fallback (a scrape of the worker's transcript when it finished without replying).
@@ -739,10 +733,9 @@ public final class FleetApp {
// silent fall-through. That is exactly the bug this ticket exists to fix:
// `default -> "done"` used to sit here and would have told a REST caller the
// delegation completed for TIMED_OUT_UNCONFIRMED, the one outcome where delivery
// is unknown. REPLIED, COMPLETED_UNREPLIED, QUESTION, STALE_TURN and
// NOT_TURN_OWNER can never actually reach this inner switch — the outer switch
// above always dispatches them first — but they still need an arm to keep this
// switch exhaustive.
// is unknown. REPLIED, COMPLETED_UNREPLIED, QUESTION and STALE_TURN can never
// actually reach this inner switch — the outer switch above always dispatches
// them first — but they still need an arm to keep this switch exhaustive.
"status", switch (reply.outcome()) {
case TIMED_OUT_WORKING -> "working";
case TIMED_OUT_QUEUED -> "queued";
@@ -752,7 +745,7 @@ public final class FleetApp {
case BUSY -> "busy";
case WORKER_FAILED -> "failed";
case BACKEND_EXHAUSTED -> "backend_exhausted";
case REPLIED, COMPLETED_UNREPLIED, QUESTION, STALE_TURN, NOT_TURN_OWNER -> "done"; // unreachable
case REPLIED, COMPLETED_UNREPLIED, QUESTION, STALE_TURN -> "done"; // unreachable
},
"detail", reply.outcome() == MessageService.Outcome.TIMED_OUT_UNCONFIRMED
? "no reply within " + timeout + "ms; delivery is unconfirmed — the "
@@ -881,12 +874,10 @@ public final class FleetApp {
body.put("sessionId", id);
body.put("status", messages.status(id).name().toLowerCase());
body.put("ready", deliverable.test(id));
// A worker paused mid-turn in an async fleet_ask is otherwise invisible to a status
// poll — surface the open question and how to answer it, same as fleet_poll's
// Phase.ASKING view, but only to the caller whose terminal created that delegation, or
// to a caller with no terminal at all (the unnamed primary).
Principal caller = ctx.attribute(CALLER);
MessageService.PendingAsk ask = messages.pendingAsk(id, caller == null ? null : caller.terminal());
// CB-582: a worker paused mid-turn in an async fleet_ask is otherwise invisible to a
// status poll — surface the open question and how to answer it, same as fleet_poll's
// Phase.ASKING view.
MessageService.PendingAsk ask = messages.pendingAsk(id);
if (ask != null) {
body.put("question", ask.question());
body.put("turnId", ask.turnId());
@@ -903,8 +894,7 @@ public final class FleetApp {
if (!allow(ctx, routeAction("GET /tasks/{ticket}"), null)) {
return;
}
Principal caller = ctx.attribute(CALLER);
MessageService.TaskView v = messages.poll(ctx.pathParam("ticket"), caller == null ? null : caller.terminal());
MessageService.TaskView v = messages.poll(ctx.pathParam("ticket"));
if (v == null) {
ctx.status(404).json(Map.of("error", "unknown_ticket", "detail", "no such task (or it has expired)"));
return;
@@ -26,7 +26,6 @@ import java.util.concurrent.atomic.AtomicBoolean;
import java.util.concurrent.atomic.AtomicLong;
import java.util.function.Consumer;
import java.util.function.LongSupplier;
import java.util.function.Supplier;
/**
* Authoritative in-daemon registry of the worker sessions this {@code fleetd} process spawned.
@@ -58,18 +57,6 @@ public final class SessionManager implements TurnListener {
* Populated on every spawn path, removed on {@link #release}.
*/
private final ConcurrentHashMap<String /*paneId*/, PeerHandle> handles = new ConcurrentHashMap<>();
/**
* fleetd #702: a pane mid-teardown, keyed by paneId, held from just before its registry entry
* is removed until {@link #releaseRemoved} finishes. {@link #spawnedMemberRole} consults this
* alongside the registry, so a caller resolving the pane's terminal during that window still
* sees a live member and never falls through to a tab map.
*
* <p>Depth-counted rather than a plain set: two threads can be tearing down the same pane at
* once (the CAS in {@link #releaseIfCurrent} exists for exactly that race), and with a set the
* loser's {@code finally} would unmark the pane while the winner is still mid-teardown,
* reopening the window this exists to close.
*/
private final ConcurrentHashMap<String /*paneId*/, Releasing> releasing = new ConcurrentHashMap<>();
private final MemberPresence presence;
private final SecureRandom nonceRandom = new SecureRandom();
private final AtomicLong nonceSeq = new AtomicLong();
@@ -255,10 +242,6 @@ public final class SessionManager implements TurnListener {
handle.id(), handle.terminalId(), resolvedProfile, actualRole, cwd, ownerTerminal, now, now, 0,
MemberSession.State.SPAWNING, null, null, handle.charterReceipt(), handle.agentSessionId());
registry.put(handle.id(), session);
// A presence contact that already arrived for this terminal found no registry
// entry to transition and gave up silently. Retry it now that one exists; remove
// this call and such a session stays in SPAWNING even though it is present.
reconcilePresence(handle.terminalId());
handles.put(handle.id(), handle);
log.debug("acquired session id={} terminal={} profile={} owner={}",
handle.id(), handle.terminalId(), session.profile(), session.ownerTerminal());
@@ -320,11 +303,9 @@ public final class SessionManager implements TurnListener {
* is a logged path an operator can reclaim, the cost of a deleted one is unrecoverable work.
*/
private MemberSession release(String paneId, ReleaseCause cause) {
return releaseWindow(paneId, registry.get(paneId), () -> {
MemberSession removed = registry.remove(paneId);
releaseRemoved(paneId, removed, handles.remove(paneId), cause);
return removed;
});
MemberSession removed = registry.remove(paneId);
releaseRemoved(paneId, removed, handles.remove(paneId), cause);
return removed;
}
/**
@@ -333,85 +314,16 @@ public final class SessionManager implements TurnListener {
* DONE record from stopping a worker that delivery has made BUSY.
*/
private boolean releaseIfCurrent(MemberSession expected, ReleaseCause cause) {
return releaseWindow(expected.paneId(), expected, () -> {
if (!registry.remove(expected.paneId(), expected)) {
// A lifecycle transition replaced the record between the caller's check and this
// remove. Log it: this race is by definition unobservable otherwise, and a reaper
// that silently declines to reap is the hardest kind of behaviour to diagnose
// after the fact.
log.debug("skipping reap of pane={}: its registry record changed after the idle "
+ "check (most likely a delivery made it BUSY)", expected.paneId());
return false;
}
releaseRemoved(expected.paneId(), expected, handles.remove(expected.paneId()), cause);
return true;
});
}
/**
* fleetd #702: mark {@code paneId} as mid-teardown — using {@code known}'s terminal/role when
* it is available — for the whole of {@code teardown}, which removes the registry entry and
* then runs {@link #releaseRemoved}. Shared by both registry-removal sites ({@link #release}'s
* unconditional remove and {@link #releaseIfCurrent}'s CAS remove) so neither can leave the
* other's window unmarked.
*
* <p>The mark is written before {@code teardown} runs — so it covers the removal itself, not
* only what comes after it — and cleared in a {@code finally}, so an unchecked throw out of
* {@code teardown} (including one from {@link PeerLauncher#stop}, which declares nothing) can
* never leave the pane marked for the rest of the daemon's life.
*/
private <T> T releaseWindow(String paneId, MemberSession known, Supplier<T> teardown) {
releasing.compute(paneId, (_, prior) -> Releasing.enter(prior, known));
try {
return teardown.get();
} finally {
releasing.compute(paneId, (_, prior) -> prior == null ? null : prior.leave());
if (!registry.remove(expected.paneId(), expected)) {
// A lifecycle transition replaced the record between the caller's check and this remove.
// Log it: this race is by definition unobservable otherwise, and a reaper that silently
// declines to reap is the hardest kind of behaviour to diagnose after the fact.
log.debug("skipping reap of pane={}: its registry record changed after the idle check "
+ "(most likely a delivery made it BUSY)", expected.paneId());
return false;
}
}
/**
* Depth count plus the terminal/role a mid-teardown pane belongs to, for
* {@link #spawnedMemberRole}. The terminal/role come from whichever call into
* {@link #releaseWindow} first knew them: a call that finds the registry entry already gone
* passes a {@code null} session, and must not blank out what the first call recorded.
*/
record Releasing(int depth, String terminalId, MemberRole role) {
static Releasing enter(Releasing prior, MemberSession known) {
int depth = (prior == null ? 0 : prior.depth()) + 1;
String terminalId = known != null ? known.terminalId() : prior == null ? null : prior.terminalId();
MemberRole role = known != null ? known.role() : prior == null ? null : prior.role();
return new Releasing(depth, terminalId, role);
}
Releasing leave() {
return depth <= 1 ? null : new Releasing(depth - 1, terminalId, role);
}
}
/**
* The role of the live spawned member occupying {@code terminal} — whether it is currently in
* the registry, or mid-teardown between {@link #release} removing its registry entry and
* {@link #releaseRemoved} actually stopping its pane (fleetd #702). {@code null} for a terminal
* that is neither: this method is the one reader a caller resolver consults before any tab
* map, so a live or releasing member's identity never falls back to a tab label.
*
* <p>Checks the registry directly via {@link #findByTerminal} rather than {@link #roster()},
* so this hot-path lookup (consulted on every resolve) never pays for a list copy or a stream.
*/
public MemberRole spawnedMemberRole(String terminal) {
MemberSession session = findByTerminal(terminal);
if (session != null) {
return session.role();
}
if (terminal == null) {
return null;
}
for (Releasing r : releasing.values()) {
if (terminal.equals(r.terminalId())) {
return r.role();
}
}
return null;
releaseRemoved(expected.paneId(), expected, handles.remove(expected.paneId()), cause);
return true;
}
private void releaseRemoved(String paneId, MemberSession removed, PeerHandle removedHandle,
@@ -470,12 +382,6 @@ public final class SessionManager implements TurnListener {
MemberSession resolved = resolveAgentSessionId(removed, removedHandle);
notifyReleased(new ReleaseDetail(resolved.terminalId(), resolved.worktree(),
resolved.branch(), snapshotRef, resolved.agentSessionId()));
String terminal = removed.terminalId();
if (terminal != null && !terminal.isBlank()) {
// Without this, a terminal stays marked present after its pane is gone, so a
// later send to the same id would read as deliverable instead of refused.
presence.forget(terminal);
}
}
}
// CB-581: the pane must always stop, even if the dirty check above threw. A session removed
@@ -819,10 +725,6 @@ public final class SessionManager implements TurnListener {
handle.charterReceipt(),
handle.agentSessionId());
registry.put(handle.id(), session);
// A presence contact that already arrived for this terminal found no registry entry to
// transition and gave up silently. Retry it now that one exists; remove this call and
// such a session stays in SPAWNING even though it is present.
reconcilePresence(handle.terminalId());
handles.put(handle.id(), handle);
log.debug("acquired worktree session id={} terminal={} profile={} branch={} path={}",
handle.id(), handle.terminalId(), session.profile(), session.branch(), session.worktree());
@@ -997,20 +899,6 @@ public final class SessionManager implements TurnListener {
transitionByTerminal(terminalId, MemberSession.State.SPAWNING, MemberSession.State.READY);
}
/**
* Completes a newly registered session's {@code SPAWNING -> READY} transition when {@code
* terminalId} was already marked present before this ran. A terminal never marked present is
* left in {@code SPAWNING}; it reaches {@code READY} normally through {@link #onReady} once
* its own contact arrives. Callers must run this only once the session's registry entry is
* already visible — {@link #onReady}'s transition matches against that entry, and reconciling
* before the entry exists finds nothing to transition.
*/
private void reconcilePresence(String terminalId) {
if (terminalId != null && !terminalId.isBlank() && presence.isPresent(terminalId)) {
onReady(terminalId);
}
}
/**
* Lifecycle hook: a message was delivered into the worker — it is now busy on a turn.
* The turn count is bumped and the activity timestamp is refreshed. A {@code DONE} session
@@ -15,7 +15,6 @@ class AuthzTest {
private static final Principal ARCH_DESIGN = Principal.architect("lead-designer", "term_design", 400);
private static final Principal ARCH_OTHER = Principal.architect("reviewer", "term_review", 500);
private static final Principal COLLABORATOR = Principal.collaborator("ops", "term_collab", 600);
private static final Principal OBSERVER = Principal.observer("term_observer", 700);
@Test
void anonymousIsAuthorizedForNothing() {
@@ -31,22 +30,6 @@ class AuthzTest {
assertTrue(Authz.isUnauthenticated(null));
}
/**
* {@code MessageService.answer}'s turn-ownership check treats a caller with no terminal as
* matching a turn recorded for the unnamed primary. That rule only stays safe because an
* unauthenticated caller — whose terminal is also {@code null} — never reaches {@code answer}
* at all: {@link #anonymousIsAuthorizedForNothing} already covers every action including
* {@code ANSWER}, but this test names the exact coupling so a future change to either side
* cannot drift without turning this test red.
*/
@Test
void anAnonymousCallerIsRefusedAnswerSoItCanNeverBeMistakenForTheUnnamedPrimary() {
assertFalse(Authz.permits(ANON, ANSWER, null),
"an anonymous caller, whose terminal is also null, must never reach answer() — the "
+ "turn-ownership check's null-terminal match for the unnamed primary owner "
+ "relies on this gate refusing it first");
}
@Test
void orchestrationBelongsToThePrimaryAlone() {
for (Authz.Action a : new Authz.Action[]{SPAWN, STOP, SEND, DRAIN}) {
@@ -267,49 +250,4 @@ class AuthzTest {
assertTrue(WORKER_A.isSpawnedMember());
assertTrue(ARCH_DESIGN.isSpawnedMember());
}
// ── the observer matrix ─────────────────────────────────────────────────────────────────────
@Test
void anObserverMayReadAndScrapeMetrics() {
assertTrue(Authz.permits(OBSERVER, READ, null));
assertTrue(Authz.permits(OBSERVER, METRICS, null));
}
@Test
void anObserverMayReplyAndAskOnlyAsItsOwnPane() {
assertTrue(Authz.permits(OBSERVER, REPLY, "term_observer"), "its own pane is its own");
assertTrue(Authz.permits(OBSERVER, ASK, "term_observer"));
assertFalse(Authz.permits(OBSERVER, REPLY, "term_design"),
"an observer must not reply on another pane");
assertFalse(Authz.permits(OBSERVER, REPLY, null),
"an absent target must not pass the own-session rule");
}
/**
* Every action beyond READ/METRICS/REPLY/ASK, asserted denied for an observer — including
* {@code TASK_READ}, which is the entire point of this role: an unconfigured pane must not be
* able to poll a ticket or read another session's status.
*/
@Test
void anObserverIsDeniedEverythingBeyondReadMetricsReplyAndAsk() {
for (Authz.Action a : Authz.Action.values()) {
if (a == READ || a == METRICS || a == REPLY || a == ASK) {
continue;
}
assertFalse(Authz.permits(OBSERVER, a, "term_observer", target -> true),
"an observer must not " + a + " even when the classifier accepts every target");
}
}
@Test
void anObserverIsNotCountedAsAnyOtherRole() {
assertFalse(OBSERVER.isPrimary());
assertFalse(OBSERVER.isWorker());
assertFalse(OBSERVER.isArchitect());
assertFalse(OBSERVER.isCollaborator());
assertFalse(OBSERVER.isSpawnedMember());
assertTrue(OBSERVER.isObserver());
}
}
@@ -1,25 +1,13 @@
package dev.ltms.fleet.auth;
import dev.ltms.fleet.guard.SubscriptionGuard;
import dev.ltms.fleet.herdr.AgentControl;
import dev.ltms.fleet.herdr.FakeHerdr;
import dev.ltms.fleet.herdr.PaneLocator;
import dev.ltms.fleet.herdr.WorkspaceControl;
import dev.ltms.fleet.mcp.ConnectionIdentity;
import dev.ltms.fleet.config.FleetConfig;
import dev.ltms.fleet.member.ClaudeCodeLauncher;
import dev.ltms.fleet.peer.MemberRole;
import dev.ltms.fleet.session.MemberSession;
import dev.ltms.fleet.session.SessionManager;
import dev.ltms.fleet.session.WorktreeRequest;
import dev.ltms.fleet.session.Worktrees;
import org.junit.jupiter.api.Test;
import java.util.List;
import java.util.Map;
import java.util.Optional;
import java.util.Set;
import java.util.concurrent.atomic.AtomicReference;
import static org.junit.jupiter.api.Assertions.*;
@@ -57,23 +45,17 @@ class CallerResolverTest {
return members;
}
/**
* With no roster wired up at all (the simple constructor), a loopback pane that owns a herdr
* pane but is not recognised as a live spawned member lands on the {@link Role#OBSERVER} floor
* — unforgeable and never token-gated, exactly like a worker's own identity, because it comes
* from the same connection-derived pane mapping.
*/
@Test
void aLoopbackPaneWithNoLiveRosterResolvesToObserverRegardlessOfAuthMode() {
void aLoopbackWorkerPaneResolvesToWorkerRegardlessOfAuthMode() {
Principal underTrust = new CallerResolver(workerIdentity()).resolve("127.0.0.1", 42, null);
Principal underToken = new CallerResolver(workerIdentity(), true, "s3cret")
.resolve("127.0.0.1", 42, null);
assertEquals(Role.OBSERVER, underTrust.role());
assertEquals(Role.WORKER, underTrust.role());
assertEquals("term_a", underTrust.terminal());
assertEquals(Role.OBSERVER, underToken.role(),
"the floor is unforgeable and must never be token-gated — otherwise enabling auth "
+ "would lock every unconfigured pane out of even READ");
assertEquals(Role.WORKER, underToken.role(),
"worker identity is unforgeable and must never be token-gated — otherwise enabling "
+ "auth would lock the whole fleet out of fleet_reply");
assertEquals("term_a", underToken.terminal());
}
@@ -97,20 +79,20 @@ class CallerResolverTest {
}
@Test
void otherPanesRemainAtTheFloorWhenAPinIsSet() {
void otherPanesRemainWorkersWhenAPinIsSet() {
Principal p = CallerResolver.pinnedTo(workerIdentity(), false, null, "term_someone_else")
.resolve("127.0.0.1", 42, null);
assertEquals(Role.OBSERVER, p.role());
assertEquals(Role.WORKER, p.role());
assertEquals("term_a", p.terminal());
}
/** The pin is optional config, so an absent or whitespace one must change nothing at all. */
@Test
void aBlankPinLeavesFloorResolutionUntouched() {
assertEquals(Role.OBSERVER,
void aBlankPinLeavesWorkerResolutionUntouched() {
assertEquals(Role.WORKER,
CallerResolver.pinnedTo(workerIdentity(), false, null, " ").resolve("127.0.0.1", 42, null).role());
assertEquals(Role.OBSERVER,
assertEquals(Role.WORKER,
CallerResolver.pinnedTo(workerIdentity(), false, null, null).resolve("127.0.0.1", 42, null).role());
}
@@ -231,13 +213,13 @@ class CallerResolverTest {
* mid-scan teardown into a refusal — the real match is still found and resolves as a worker.
*/
@Test
void aHerdrErrorOnANonOwningPaneStillResolvesTheRealPane() {
void aHerdrErrorOnANonOwningPaneStillResolvesTheRealWorker() {
FakeHerdr vanishedElsewhere = new FakeHerdr().processInfoFailsForPane("w2:p9", "pane_not_found");
ConnectionIdentity id = new ConnectionIdentity(new PaneLocator(vanishedElsewhere), _ -> FakeHerdr.WORKER_PID);
Principal p = new CallerResolver(id).resolve("127.0.0.1", 55555, null);
assertEquals(Role.OBSERVER, p.role());
assertEquals(Role.WORKER, p.role());
assertEquals("term_a", p.terminal());
}
@@ -281,12 +263,12 @@ class CallerResolverTest {
}
@Test
void aPaneAbsentFromTheRegistryFallsToTheObserverFloor() {
void aPaneAbsentFromTheRegistryIsStillAWorker() {
Principal p = new CallerResolver(workerIdentity(), false, null,
Map.of("term_elsewhere", "gpt-sol-5.6"))
.resolve("127.0.0.1", 42, null);
assertEquals(Role.OBSERVER, p.role());
assertEquals(Role.WORKER, p.role());
assertEquals("term_a", p.terminal());
assertNull(p.name());
}
@@ -312,16 +294,16 @@ class CallerResolverTest {
}
@Test
void anEmptyRegistryLeavesEveryPaneAtTheObserverFloor() {
void anEmptyRegistryLeavesEveryPaneAWorker() {
Map<String, String> noLeads = null;
assertEquals(Role.OBSERVER,
assertEquals(Role.WORKER,
new CallerResolver(workerIdentity(), false, null, Map.of())
.resolve("127.0.0.1", 42, null).role());
assertEquals(Role.OBSERVER,
assertEquals(Role.WORKER,
new CallerResolver(workerIdentity(), false, null, noLeads)
.resolve("127.0.0.1", 42, null).role());
// And the same for the live-registry form, whose supplier may also be absent.
assertEquals(Role.OBSERVER,
// CB-531: and the same for the live-registry form, whose supplier may also be absent.
assertEquals(Role.WORKER,
CallerResolver.withLeads(workerIdentity(), false, null, null)
.resolve("127.0.0.1", 42, null).role());
}
@@ -394,7 +376,7 @@ class CallerResolverTest {
Map<String, String> live = new java.util.HashMap<>();
CallerResolver r = CallerResolver.withLeads(workerIdentity(), false, null, () -> live);
assertEquals(Role.OBSERVER, r.resolve("127.0.0.1", 42, null).role());
assertEquals(Role.WORKER, r.resolve("127.0.0.1", 42, null).role());
live.put("term_a", "gpt-sol-5.6"); // the scanner sees a newly-labelled tab
@@ -411,7 +393,7 @@ class CallerResolverTest {
mutable.put("term_a", "sneaky");
assertEquals(Role.OBSERVER, r.resolve("127.0.0.1", 42, null).role());
assertEquals(Role.WORKER, r.resolve("127.0.0.1", 42, null).role());
}
// ── CB-548: architect slots ─────────────────────────────────────────────────────────────────
@@ -441,13 +423,13 @@ class CallerResolverTest {
}
@Test
void anUnboundPaneResolvesToTheObserverFloor() {
void anUnboundPaneStillResolvesAsAWorker() {
MemberRegistry members = new MemberRegistry(new FleetConfig.Fleet(Map.of(),
Map.of("lead-designer", new FleetConfig.Slot("sonnet")), Map.of(), Map.of(), null));
Principal p = CallerResolver.withLeadsAndMembers(workerIdentity(), false, null,
Map::of, members).resolve("127.0.0.1", 42, null);
assertEquals(Role.OBSERVER, p.role());
assertEquals(Role.WORKER, p.role());
assertNull(p.name());
}
@@ -473,11 +455,12 @@ class CallerResolverTest {
CallerResolver r = CallerResolver.withLeadsAndMembers(workerIdentity(), false, null,
Map::of, members);
assertEquals(Role.OBSERVER, r.resolve("127.0.0.1", 42, null).role());
assertEquals(Role.WORKER, r.resolve("127.0.0.1", 42, null).role());
assertTrue(members.bind("architect:lead-designer", "term_a")); // the later lifecycle binds the slot
assertEquals(Role.ARCHITECT, r.resolve("127.0.0.1", 42, null).role());
assertEquals("architect:lead-designer", r.members().get("term_a"));
}
@Test
@@ -500,13 +483,12 @@ class CallerResolverTest {
}
@Test
void aBoundNonArchitectSlotResolvesToTheObserverFloorNotArchitect() {
void aBoundNonArchitectSlotStillResolvesAsAWorker() {
Principal p = CallerResolver.withLeadsAndMembers(workerIdentity(), false, null,
Map::of, boundMembers("dev:builder", MemberRole.DEV))
.resolve("127.0.0.1", 42, null);
assertEquals(Role.OBSERVER, p.role(), "a dev binding must never grant architect rights, "
+ "and this construction path wires no roster to recognise it as the live dev it is");
assertEquals(Role.WORKER, p.role(), "a dev binding must never grant architect rights");
}
@Test
@@ -517,17 +499,17 @@ class CallerResolverTest {
assertThrows(IllegalArgumentException.class, () -> new CallerResolver(id, true, " "));
}
@Test
void aPaneOnAnyLoopbackSourceAddressIsStillAtTheFloorNotThePrimary() {
// fleetd #305: the escalation this guards against. ConnectionIdentity used to accept only
// 127.0.0.1, so a pane connecting from 127.0.0.2 resolved to no terminal, and this
// resolver's own (wider) loopback check then made it the PRIMARY — granting spawn, stop,
// send and drain. Measured on the Linux fleet host: binding a source of 127.0.0.2 succeeds
// there, so the path is real and not theoretical.
void aWorkerOnAnyLoopbackSourceAddressIsStillAWorkerNotThePrimary() {
// fleetd #305: the escalation. ConnectionIdentity used to accept only 127.0.0.1, so a
// worker connecting from 127.0.0.2 resolved to no terminal, and this resolver's own
// (wider) loopback check then made it the PRIMARY — granting spawn, stop, send and drain.
// Measured on the Linux fleet host: binding a source of 127.0.0.2 succeeds there, so the
// path is real and not theoretical.
CallerResolver r = new CallerResolver(workerIdentity(), false, null);
for (String src : new String[]{"127.0.0.1", "127.0.0.2", "127.1.2.3", "::ffff:127.0.0.2"}) {
Principal p = r.resolve(src, 55555, null);
assertEquals(Role.OBSERVER, p.role(), "the pane must stay off PRIMARY from source " + src);
assertEquals("term_a", p.terminal(), "pane terminal from source " + src);
assertEquals(Role.WORKER, p.role(), "a worker must stay a worker from source " + src);
assertEquals("term_a", p.terminal(), "worker terminal from source " + src);
}
}
@@ -560,219 +542,6 @@ class CallerResolverTest {
assertEquals("term_a", p.terminal());
}
/**
* fleetd #702: between {@link SessionManager#release} removing the registry entry and the
* pane actually stopping, a resolve for that terminal must still see the live member and
* never fall through to a tab map — wired through the real {@link SessionManager}, not a
* hand-rolled stand-in for its {@code spawnedMemberRole}.
*
* <p>Reuses the ticket's own test idea: an injected {@code hasUncommitted} resolves the
* releasing terminal from inside {@code release}'s window — a real call landing inside the
* window, so no sleep and no race.
*
* <p>The property asserted is "no tab map is consulted", never "the same role is returned".
* The mandatory control is the lead tab map: it names this exact terminal, and the same
* resolve taken <em>outside</em> the window (before release runs) must still return the lead
* role — without that control, the in-window assertion would also pass on an empty map and
* prove nothing.
*/
@Test
void aPaneMidTeardownResolvesAsItsOwnRoleConsultingNoTabMap() {
FakeHerdr herdr = new FakeHerdr().pinNextStarts(1, "term_a", "w2:p7");
FleetConfig.Profile cfg = new FleetConfig.Profile(
"ltms-local", "http://gx00.gw:8000", "coder", null, "FLEETD_WORKER_TOKEN",
List.of("ccs", "ltms-local"), "tab", "fleetd-workers",
"worker: {profile} #{n}", null, null, null);
ClaudeCodeLauncher workers = new ClaudeCodeLauncher(new AgentControl(herdr), new WorkspaceControl(herdr),
new SubscriptionGuard(Set.of("gx00.gw")), Map.of(cfg.profile(), cfg), cfg.profile(), _ -> null);
AtomicReference<Principal> duringWindow = new AtomicReference<>();
AtomicReference<CallerResolver> resolverRef = new AtomicReference<>();
Worktrees worktrees = new Worktrees() {
@Override
public String add(String repoRoot, String branch, String baseRef) {
return "/wt/" + branch.replace('/', '_');
}
@Override
public void remove(String repoRoot, String worktreePath) {
}
@Override
public void deleteBranch(String repoRoot, String branch) {
}
@Override
public boolean hasUncommitted(String worktreePath) {
// Runs from INSIDE release()'s git-status shell-out: the registry entry is
// already gone, but the pane has not stopped yet.
duringWindow.set(resolverRef.get().resolve("127.0.0.1", 42, null));
return false;
}
@Override
public void overlayParity(String repoRoot, String worktreePath, List<String> overlay) {
}
@Override
public String repoRoot(String cwd) {
return "/repo";
}
@Override
public Optional<String> snapshot(String worktreePath, String branch, String message) {
return Optional.empty();
}
@Override
public WipRefStats wipRefs(String repoRoot) {
return new WipRefStats(0, 0L);
}
@Override
public int pruneWipRefs(String repoRoot, long minAgeMillis) {
return 0;
}
@Override
public void shareWithGroup(String repoRoot, String worktreePath) {
}
};
SessionManager sessions = new SessionManager(workers, worktrees);
// The lead tab map names "term_a" before anything is ever spawned onto it — the control
// this test needs. Built up front so the SAME resolver answers every resolve() call below.
ConnectionIdentity identity = new ConnectionIdentity(new PaneLocator(herdr), _ -> FakeHerdr.WORKER_PID);
CallerResolver resolver = CallerResolver.withLeadsAndMembers(identity, false, null,
() -> Map.of("term_a", "the-lead"), null, sessions::spawnedMemberRole, Map::of);
resolverRef.set(resolver);
Principal before = resolver.resolve("127.0.0.1", 42, null);
assertEquals(Role.PRIMARY, before.role(),
"control: with no live or releasing member on this terminal, the lead tab map must "
+ "win — this is what proves the in-window assertion below is not passing on "
+ "an empty map");
assertEquals("the-lead", before.name());
MemberSession s = sessions.acquire("ltms-local", null, "/caller/proj", null,
new WorktreeRequest("fleetd-702", null));
assertEquals("term_a", s.terminalId(), "sanity: the spawn resolved to the pinned pane");
sessions.release(s.paneId());
assertNotNull(duringWindow.get(), "the dirty check must have run and captured a resolve");
assertEquals(Role.WORKER, duringWindow.get().role(),
"inside the window the pane must resolve as its own live-member role, consulting no "
+ "tab map — a lead tab naming the same terminal must not win");
assertEquals("term_a", duringWindow.get().terminal());
}
/**
* {@link SessionManager#releaseRemoved} unbinds the architect slot before the git-status
* shell-out that opens the teardown window, so a resolve landing inside that window must see
* the slot already unbound and resolve {@link Role#WORKER} — never {@link Role#ARCHITECT},
* and never by checking role equality against the live session, which would hold even if the
* unbind ran too late.
*
* <p>The control is the same resolve taken outside the window, while the slot is still bound,
* which must return {@link Role#ARCHITECT} — without it this test would also pass against a
* slot that was never bound, and prove nothing.
*/
@Test
void aReleasingArchitectIsDemotedToWorkerInsideTheTeardownWindow() {
FakeHerdr herdr = new FakeHerdr().pinNextStarts(1, "term_a", "w2:p7");
FleetConfig.Profile cfg = new FleetConfig.Profile(
"ltms-local", "http://gx00.gw:8000", "coder", null, "FLEETD_WORKER_TOKEN",
List.of("ccs", "ltms-local"), "tab", "fleetd-workers",
"worker: {profile} #{n}", null, null, null);
ClaudeCodeLauncher workers = new ClaudeCodeLauncher(new AgentControl(herdr), new WorkspaceControl(herdr),
new SubscriptionGuard(Set.of("gx00.gw")), Map.of(cfg.profile(), cfg), cfg.profile(), _ -> null);
AtomicReference<Principal> duringWindow = new AtomicReference<>();
AtomicReference<CallerResolver> resolverRef = new AtomicReference<>();
Worktrees worktrees = new Worktrees() {
@Override
public String add(String repoRoot, String branch, String baseRef) {
return "/wt/" + branch.replace('/', '_');
}
@Override
public void remove(String repoRoot, String worktreePath) {
}
@Override
public void deleteBranch(String repoRoot, String branch) {
}
@Override
public boolean hasUncommitted(String worktreePath) {
// Runs from INSIDE release()'s git-status shell-out: the architect slot is already
// unbound by this point, but the pane has not stopped yet.
duringWindow.set(resolverRef.get().resolve("127.0.0.1", 42, null));
return false;
}
@Override
public void overlayParity(String repoRoot, String worktreePath, List<String> overlay) {
}
@Override
public String repoRoot(String cwd) {
return "/repo";
}
@Override
public Optional<String> snapshot(String worktreePath, String branch, String message) {
return Optional.empty();
}
@Override
public WipRefStats wipRefs(String repoRoot) {
return new WipRefStats(0, 0L);
}
@Override
public int pruneWipRefs(String repoRoot, long minAgeMillis) {
return 0;
}
@Override
public void shareWithGroup(String repoRoot, String worktreePath) {
}
};
SessionManager sessions = new SessionManager(workers, worktrees);
MemberRegistry members = new MemberRegistry(new FleetConfig.Fleet(Map.of(),
Map.of("lead-designer", new FleetConfig.Slot("ltms-local")), Map.of(), Map.of(), null));
sessions.setMemberLifecycle(members);
ConnectionIdentity identity = new ConnectionIdentity(new PaneLocator(herdr), _ -> FakeHerdr.WORKER_PID);
CallerResolver resolver = CallerResolver.withLeadsAndMembers(identity, false, null,
Map::of, members, sessions::spawnedMemberRole, Map::of);
resolverRef.set(resolver);
MemberSession s = sessions.acquire("ltms-local", MemberRole.ARCHITECT, null, "/caller/proj", null,
new WorktreeRequest("fleetd-702d", null));
assertEquals("term_a", s.terminalId(), "sanity: the spawn resolved to the pinned pane");
assertEquals(MemberRole.ARCHITECT, s.role(), "sanity: the slot bind succeeded");
Principal before = resolver.resolve("127.0.0.1", 42, null);
assertEquals(Role.ARCHITECT, before.role(),
"control: with the slot still bound, the pane must resolve as an architect — this "
+ "is what proves the in-window assertion below is not passing against a "
+ "slot that was never bound");
assertEquals("lead-designer", before.name());
sessions.release(s.paneId());
assertNotNull(duringWindow.get(), "the dirty check must have run and captured a resolve");
assertEquals(Role.WORKER, duringWindow.get().role(),
"inside the window the architect slot is already unbound, so the result must be "
+ "WORKER — asserting role-equality with the live session here would tempt "
+ "moving the unbind earlier or later, which would be wrong either way");
assertEquals("term_a", duringWindow.get().terminal());
}
/** A spawned architect in the roster resolves ARCHITECT, carrying its bound slot's name. */
@Test
void aSpawnedArchitectInTheRosterResolvesArchitectWithItsSlotName() {
@@ -833,14 +602,14 @@ class CallerResolverTest {
assertEquals("term_a", p.terminal());
}
/** Regression: an empty collaborator registry leaves every pane at the unconfigured-pane floor. */
/** Regression: an empty collaborator registry leaves every pane exactly as before. */
@Test
void anEmptyCollaboratorRegistryLeavesEveryPaneAtTheObserverFloor() {
void anEmptyCollaboratorRegistryLeavesEveryPaneAsBefore() {
Principal p = CallerResolver.withLeadsAndMembers(workerIdentity(), false, null, Map::of,
new MemberRegistry(null), t -> null, Map::of)
.resolve("127.0.0.1", 42, null);
assertEquals(Role.OBSERVER, p.role());
assertEquals(Role.WORKER, p.role());
assertNull(p.name());
}
@@ -869,39 +638,6 @@ class CallerResolverTest {
assertFalse(r.knownLeadOrCollaborator().test("term_other"));
}
// ── fleetd #705: narrowing the unconfigured-pane floor to OBSERVER ──────────────────────────
/**
* The case this ticket exists for: a pane the resolver cannot place as a live spawned member,
* a lead, a bound architect slot, or a configured collaborator must land on the narrow
* {@link Role#OBSERVER} floor, never the {@link Role#WORKER} the old fallback granted.
*
* <p>The second assertion is the control the ticket requires: a terminal the roster DOES
* recognise as a live spawned member must still resolve its own role. Without it, this test
* would also pass if the fix accidentally turned every caller into an observer.
*/
@Test
void anUnconfiguredPaneResolvesObserverButARegisteredMemberStillResolvesItsOwnRole() {
Principal unconfigured = new CallerResolver(workerIdentity()).resolve("127.0.0.1", 42, null);
assertEquals(Role.OBSERVER, unconfigured.role(),
"a pane matching none of the configured or live-roster roles must fall to the "
+ "floor, not WORKER");
assertEquals("term_a", unconfigured.terminal());
Principal registered = CallerResolver.withLeadsAndMembers(workerIdentity(), false, null,
Map::of, new MemberRegistry(null),
t -> "term_a".equals(t) ? MemberRole.DEV : null, Map::of)
.resolve("127.0.0.1", 42, null);
assertEquals(Role.WORKER, registered.role(),
"control: a live spawned member must keep resolving its own role, never the "
+ "unconfigured-pane floor");
}
@Test
void describeNamesTheObserverByItsPane() {
assertEquals("observer:term_a", Principal.observer("term_a", 1).describe());
}
@Test
void knownLeadOrCollaboratorIsFalseForASpawnedMembersTerminal() {
// The exact scenario a collaborator's SEND must never reach: a live spawned member's own
@@ -295,10 +295,9 @@ class MemberRegistryLiveTest {
assertTrue(out.applied(), "the reload must actually take effect: " + out.summary());
Principal after = resolver.resolve("127.0.0.1", 42, null);
assertEquals(Role.OBSERVER, after.role(),
"removing the slot from config must demote the bound session on its NEXT request — "
+ "this harness wires no live roster for term_a, so the demotion lands on "
+ "the unconfigured-pane floor");
assertEquals(Role.WORKER, after.role(),
"removing the slot from config must demote the bound session to worker on its "
+ "NEXT request — this is the ticket's whole point");
assertEquals("term_a", after.terminal(), "same pane, same terminal — only the role changed");
}
@@ -850,8 +850,7 @@ class FleetConfigTest {
/**
* The hazard this guard closes: a pane-placed member lands inside the focused tab rather than
* its own, so it can land inside a lead's labelled tab and, while its pane carries no entry in
* the spawned-member roster, be read back as that lead.
* its own, so it can land inside a lead's labelled tab and be read back as that lead.
*/
@Test
void aPanePlacedProfileWithALeadTabRefusesToStart(@TempDir Path dir) throws Exception {
@@ -1088,9 +1087,8 @@ class FleetConfigTest {
}
/**
* A member tabLabel that can render as a configured collaborator tab is the same hazard as the
* lead case above — while its pane carries no entry in the spawned-member roster, a member
* labelled that way is read back as the collaborator.
* fleetd #669: a member tabLabel that can render as a configured collaborator tab is the same
* hazard as the lead case above — a member labelled that way is read back as the collaborator.
*/
@Test
void aProfileTabLabelOverrideMatchingACollaboratorTabRefusesToStart(@TempDir Path dir)
@@ -1,16 +1,10 @@
package dev.ltms.fleet.lead;
import ch.qos.logback.classic.Level;
import ch.qos.logback.classic.Logger;
import ch.qos.logback.classic.spi.ILoggingEvent;
import ch.qos.logback.core.read.ListAppender;
import dev.ltms.fleet.config.FleetConfig;
import dev.ltms.fleet.herdr.AgentControl;
import dev.ltms.fleet.herdr.FakeHerdr;
import dev.ltms.fleet.herdr.ResilientAgentLaunch;
import dev.ltms.fleet.herdr.WorkspaceControl;
import org.junit.jupiter.api.Test;
import org.slf4j.LoggerFactory;
import java.util.LinkedHashMap;
import java.util.List;
@@ -70,21 +64,6 @@ class LeadLauncherTest {
return new LeadLauncher(new AgentControl(herdr), new WorkspaceControl(herdr), cfg);
}
/** As {@link #launcher}, plus a fast no-op sleeper so a busy-retry test never real-sleeps. */
private static LeadLauncher fastLauncher(FakeHerdr herdr, FleetConfig cfg) {
return new LeadLauncher(new AgentControl(herdr), new WorkspaceControl(herdr), cfg, () -> { });
}
/** {@link #opusProfile()} with one argv element long enough to overflow the pane line limit. */
private static FleetConfig.Profile hugeArgvProfile() {
return new FleetConfig.Profile(
"opus", null, "claude-opus-5", null, "FLEETD_WORKER_TOKEN",
List.of("ccs", "x".repeat(1500)), "tab", "fleet", null,
"http://127.0.0.1:8765/mcp", null, null,
null, null, null,
Map.of("CLAUDE_CODE_AUTO_COMPACT_WINDOW", "300000"), null, null, true, null);
}
@SuppressWarnings("unchecked")
private static List<String> startedArgs(FakeHerdr herdr) {
return (List<String>) ((Map<String, Object>) herdr.lastCall("agent.start").params()).get("args");
@@ -107,11 +86,7 @@ class LeadLauncherTest {
assertEquals(1, launcher(herdr, configWith(lead("opus", "lead: opus", 1))).ensureLeads());
assertTrue(herdr.called("agent.start"), "a lead must actually be started");
// fleetd #727: the name carries a per-process nonce and a per-start sequence number — the
// same unique-naming scheme HerdrPeerLauncher uses for members — rather than the fixed
// "lead-opus" a stale registry entry could block a legitimate relaunch under.
assertTrue(startedName(herdr).matches("lead-opus-[0-9a-f]{6}-\\d+"),
"name is lead-<name>-<nonce>-<seq>: " + startedName(herdr));
assertEquals("lead-opus", startedName(herdr));
}
/** The tab is labelled with the configured `tab:` so the scanner finds the lead on the next resolve. */
@@ -461,247 +436,4 @@ class LeadLauncherTest {
assertEquals(0, launcher(herdr, cfg).ensureLeads());
assertTrue(herdr.calls.isEmpty(), "nothing declared ⇒ nothing scanned");
}
// ── fleetd #727: the same three launch protections every member spawn gets ───────────────────
/**
* A freshly created pane may not have redrawn its prompt yet, so herdr answers
* {@code agent_pane_busy}. The lead launch must wait it out rather than fail on the first miss
* — exactly the retry {@code HerdrPeerLauncher} already gives every member.
*/
@Test
void aLeadLaunchRetriesWhileTheSeedShellBoots() {
FakeHerdr herdr = new FakeHerdr().agentPaneBusyTimes(2);
assertEquals(1, fastLauncher(herdr, configWith(lead("opus", "lead: opus", 1))).ensureLeads(),
"the lead must still start once the shell is ready");
assertEquals(3, herdr.calls.stream().filter(c -> c.method().equals("agent.start")).count(),
"two busy rejections, then the successful start");
}
/**
* The busy retry is bounded, not an infinite poll. If the pane never becomes ready the launch
* must eventually give up and log the failure, not hang the daemon's reconcile loop forever —
* proven here by a budget that would still be busy on attempt
* {@value ResilientAgentLaunch#SHELL_READY_RETRIES} and a call count that stops exactly there.
*/
@Test
void aLeadLaunchGivesUpAfterTheBoundedBusyBudgetRatherThanLoopingForever() {
FakeHerdr herdr = new FakeHerdr().agentPaneBusyTimes(999);
assertEquals(0, fastLauncher(herdr, configWith(lead("opus", "lead: opus", 1))).ensureLeads(),
"a pane that never becomes ready must not be reported as a started lead");
assertEquals(ResilientAgentLaunch.SHELL_READY_RETRIES,
herdr.calls.stream().filter(c -> c.method().equals("agent.start")).count(),
"the retry budget is bounded: it stops after exactly SHELL_READY_RETRIES attempts");
}
/**
* herdr refuses a duplicate agent {@code name} outright. A stale registry entry — a crashed
* lead session, or a name the registry has not yet released — must not permanently block a
* legitimate relaunch, so each retry attempt carries a fresh per-start name.
*/
@Test
@SuppressWarnings("unchecked")
void aLeadLaunchRetriesUnderAFreshNameWhenTheOldNameIsStillTaken() {
FakeHerdr herdr = new FakeHerdr().agentNameTakenTimes(2);
assertEquals(1, launcher(herdr, configWith(lead("opus", "lead: opus", 1))).ensureLeads(),
"the lead must still start once a free name is found");
List<String> names = herdr.calls.stream()
.filter(c -> c.method().equals("agent.start"))
.map(c -> ((Map<String, Object>) c.params()).get("name").toString())
.toList();
assertEquals(3, names.size(), "2 rejected + 1 success");
assertEquals(3, Set.copyOf(names).size(), "each attempt must use a distinct name");
}
/**
* The name-collision retry is bounded too. If the name is taken on every attempt, the launch
* must give up rather than keep minting new names forever — the per-start naming scheme means
* every failed attempt was refused outright by herdr (no process exists under a name herdr
* refused), so a bounded, exhausted retry never leaves a second process running: nothing is
* running at all.
*/
@Test
void aLeadLaunchNeverEndsUpWithASecondProcessWhenTheNameStaysTaken() {
FakeHerdr herdr = new FakeHerdr().agentNameTakenTimes(999);
assertEquals(0, launcher(herdr, configWith(lead("opus", "lead: opus", 1))).ensureLeads(),
"a name that is never free must not be reported as a started lead");
assertEquals(ResilientAgentLaunch.NAME_RETRIES,
herdr.calls.stream().filter(c -> c.method().equals("agent.start")).count(),
"the retry budget is bounded: it stops after exactly NAME_RETRIES attempts, "
+ "never racing a duplicate into existence");
}
/**
* fleetd #220/#727: herdr types the launch command into the pane as one line, and a pty line
* buffer holds only 1024 bytes — past that the tail is dropped with no error at all, and the
* backend exits on a mangled argument. The lead launch must refuse an over-long command outright
* rather than let it be typed and silently truncated.
*/
@Test
void anOverlongLeadArgvIsRefusedRatherThanTypedAndTruncated() {
FakeHerdr herdr = new FakeHerdr();
Logger logger = (Logger) LoggerFactory.getLogger(LeadLauncher.class);
ListAppender<ILoggingEvent> appender = new ListAppender<>();
appender.start();
logger.addAppender(appender);
int started;
try {
started = launcher(herdr, configWith(lead("opus", "lead: opus", 1), hugeArgvProfile()))
.ensureLeads();
} finally {
logger.detachAppender(appender);
}
assertEquals(0, started, "an over-long command must never be reported as a started lead");
assertFalse(herdr.called("agent.start"),
"nothing may be started — a truncated command is worse than no spawn");
String warn = appender.list.stream()
.filter(e -> e.getLevel().equals(Level.WARN))
.map(ILoggingEvent::getFormattedMessage)
.filter(m -> m.contains("failed to launch"))
.findFirst()
.orElseThrow(() -> new AssertionError("expected a WARN naming the launch failure: "
+ appender.list));
assertTrue(warn.contains("1024"), "names the limit: " + warn);
assertTrue(warn.contains("opus"), "names the profile: " + warn);
assertTrue(warn.contains("x".repeat(60)), "names the culprit argument: " + warn);
}
// ── fleetd #726 unit 1: the single-lead relaunch seam ─────────────────────────────────────
/**
* The returned agent's {@code terminalId()}/{@code paneId()} are the ones the fake
* {@code AgentControl} actually started — not a coincidental field left over from the caller.
* {@code paneId()} echoes the exact {@code pane_id} the launch's own {@code agent.start} call
* carried (protocol 19: the agent starts into the pane it is asked to), and {@code
* terminalId()} is herdr's own generated id, which the fake always shapes as {@code
* term_new_<n>}.
*/
@Test
void relaunchReturnsTheStartedAgent() {
FakeHerdr herdr = new FakeHerdr();
dev.ltms.fleet.herdr.Agent started =
launcher(herdr, configWith(lead("opus", "lead: opus", 1))).relaunch("opus");
assertNotNull(started, "a launchable, configured lead must start");
Object startedPaneIdParam = ((Map<?, ?>) herdr.lastCall("agent.start").params()).get("pane_id");
assertEquals(startedPaneIdParam, started.paneId(),
"paneId() must be the pane the agent.start call actually targeted");
assertTrue(started.terminalId() != null && started.terminalId().startsWith("term_new_"),
"terminalId() must be herdr's own generated id: " + started.terminalId());
}
/** The new tab is labelled with the lead's configured {@code tab:}, and AFTER the start. */
@Test
void relaunchLabelsTheNewTabAfterStarting() {
FakeHerdr herdr = new FakeHerdr();
dev.ltms.fleet.herdr.Agent started =
launcher(herdr, configWith(lead("opus", "lead: opus", 1))).relaunch("opus");
assertNotNull(started);
assertEquals("lead: opus", ((Map<?, ?>) herdr.lastCall("tab.rename").params()).get("label"));
int startIndex = indexOfLastCall(herdr, "agent.start");
int renameIndex = indexOfLastCall(herdr, "tab.rename");
assertTrue(renameIndex > startIndex,
"the tab must be renamed AFTER the start succeeds, not before: start=" + startIndex
+ " rename=" + renameIndex);
}
private static int indexOfLastCall(FakeHerdr herdr, String method) {
int idx = -1;
List<FakeHerdr.Call> calls = herdr.calls;
for (int i = 0; i < calls.size(); i++) {
if (calls.get(i).method().equals(method)) {
idx = i;
}
}
return idx;
}
@Test
void relaunchOfAnUnknownLeadNameReturnsNullAndStartsNothing() {
FakeHerdr herdr = new FakeHerdr();
dev.ltms.fleet.herdr.Agent started =
launcher(herdr, configWith(lead("opus", "lead: opus", 1))).relaunch("not-declared");
assertNull(started);
assertFalse(herdr.called("agent.start"));
assertFalse(herdr.called("workspace.create"));
assertFalse(herdr.called("tab.create"));
}
@Test
void relaunchOfARecogniseOnlyLeadReturnsNullAndStartsNothing() {
FakeHerdr herdr = new FakeHerdr();
dev.ltms.fleet.herdr.Agent started =
launcher(herdr, configWith(lead(null, "lead: dead", 1))).relaunch("opus");
assertNull(started);
assertFalse(herdr.called("agent.start"));
}
@Test
void relaunchWithAnUnconfiguredProfileReturnsNullAndStartsNothing() {
FakeHerdr herdr = new FakeHerdr();
dev.ltms.fleet.herdr.Agent started =
launcher(herdr, configWith(lead("nope", "lead: opus", 1))).relaunch("opus");
assertNull(started);
assertFalse(herdr.called("agent.start"));
}
/**
* The outer retry {@link LeadLauncher#relaunch(String)} owns, separate from {@code
* ResilientAgentLaunch}'s internal {@code agent_name_taken} retry: a failed attempt must not
* be the end of the whole relaunch. Each of the first two attempts exhausts {@code
* ResilientAgentLaunch.NAME_RETRIES} name attempts (every one of them rejected), so each
* attempt's own tab is created and then closed; the third attempt's first name is free.
*/
@Test
void relaunchRetriesTheWholeAttemptAndSucceedsOnTheThird() {
FakeHerdr herdr = new FakeHerdr()
.agentNameTakenTimes(2 * ResilientAgentLaunch.NAME_RETRIES);
dev.ltms.fleet.herdr.Agent started =
fastLauncher(herdr, configWith(lead("opus", "lead: opus", 1))).relaunch("opus");
assertNotNull(started, "the third attempt's first name is free — it must succeed");
assertEquals(3, herdr.calls.stream().filter(c -> c.method().equals("tab.create")).count(),
"one tab per attempt: three attempts");
assertEquals(2, herdr.calls.stream().filter(c -> c.method().equals("tab.close")).count(),
"the two failed attempts' tabs must be closed");
}
/**
* Every attempt fails outright (a herdr error {@code ResilientAgentLaunch} does not retry at
* all) — {@link LeadLauncher#relaunch(String)} must give up after exactly {@code
* RELAUNCH_ATTEMPTS} and must not leak any of the tabs it created along the way.
*/
@Test
void relaunchGivesUpAfterExactlyRelaunchAttemptsAndLeaksNoTab() {
FakeHerdr herdr = new FakeHerdr().agentStartFailsWith("some_other_error");
dev.ltms.fleet.herdr.Agent started =
fastLauncher(herdr, configWith(lead("opus", "lead: opus", 1))).relaunch("opus");
assertNull(started, "every attempt failed — relaunch must give up, not hang or guess");
assertEquals(LeadLauncher.RELAUNCH_ATTEMPTS,
herdr.calls.stream().filter(c -> c.method().equals("agent.start")).count(),
"exactly RELAUNCH_ATTEMPTS attempts, no more, no fewer");
long tabsCreated = herdr.calls.stream().filter(c -> c.method().equals("tab.create")).count();
long tabsClosed = herdr.calls.stream().filter(c -> c.method().equals("tab.close")).count();
assertEquals(LeadLauncher.RELAUNCH_ATTEMPTS, tabsCreated);
assertEquals(tabsCreated, tabsClosed, "every tab this method created must be closed — no leaks");
}
}
@@ -1305,12 +1305,8 @@ class LeadRolloverTest {
int rolls = LeadRollover.OUTCOME_HISTORY_CAP + 50;
String[] tokens = new String[rolls];
for (int i = 0; i < rolls; i++) {
// a distinct terminal per iteration: confirm() single-flights per lead terminal, so
// reusing LEAD here would refuse every roll after the first instead of building up
// OUTCOME_HISTORY_CAP+50 simultaneously in-flight entries.
String terminal = "term_cap_" + i;
LeadRollover.PendingRollover pending = rollover.open(terminal, "context is full #" + i);
LeadRollover.RollDecision decision = rollover.confirm(terminal, pending.token(), true);
LeadRollover.PendingRollover pending = rollover.open(LEAD, "context is full #" + i);
LeadRollover.RollDecision decision = rollover.confirm(LEAD, pending.token(), true);
assertTrue(decision.accepted(), "roll #" + i + " should have been approved: " + decision.detail());
tokens[i] = pending.token();
}
@@ -1451,195 +1447,4 @@ class LeadRolloverTest {
assertTrue(status.detail().contains("HerdrException"), "the detail must name the exception "
+ "so an operator reading status() has something to act on: " + status.detail());
}
// ---- fleetd #726 unit 3: confirm() single-flights per lead terminal -----------------------
@Test
@DisplayName("[SINGLE-FLIGHT 1] a second confirm() for the SAME lead terminal is refused with "
+ "ROLL_ALREADY_RUNNING while the first roll is still in flight, continuationRunner ran "
+ "exactly once, and the claim is released once that first roll finishes")
void secondConfirmForSameTerminalIsRefusedWhileFirstRollIsStillInFlight() throws IOException {
FakeHerdr herdr = new FakeHerdr(); // default idle — the held roll WOULD complete once run
Path handover = writeHandover("handover contents");
AtomicLong clock = new AtomicLong(1_000);
HoldingRunner runner = new HoldingRunner();
LeadRollover rollover = newRolloverWithHoldingRunner(herdr, cfg(handover.toString()),
fixedClock(clock), runner);
LeadRollover.PendingRollover first = rollover.open(LEAD, "context is full");
LeadRollover.RollDecision firstDecision = rollover.confirm(LEAD, first.token(), true);
assertTrue(firstDecision.accepted(), "expected approval; got: " + firstDecision.reason()
+ " / " + firstDecision.detail());
assertEquals(1, runner.heldCount(), "sanity: the first roll is held, not run yet");
LeadRollover.PendingRollover second = rollover.open(LEAD, "a second request for the same terminal");
LeadRollover.RollDecision secondDecision = rollover.confirm(LEAD, second.token(), true);
assertFalse(secondDecision.accepted());
assertEquals(LeadRollover.RefusalReason.ROLL_ALREADY_RUNNING, secondDecision.reason());
assertTrue(secondDecision.detail().contains(LEAD), "the refusal detail must name the lead "
+ "terminal: " + secondDecision.detail());
assertTrue(secondDecision.detail().contains(first.token()), "the refusal detail must name "
+ "the token holding the claim: " + secondDecision.detail());
assertEquals(1, runner.heldCount(), "the refused confirm() must never have reached "
+ "continuationRunner — it ran exactly once, for the first roll only");
runner.runNext(); // let the first (and only) held roll finish
assertEquals(2, promptCallCount(herdr), "continuationRunner ran exactly once: /clear then "
+ "bootstrapText, for the first roll only");
// The claim must have been released once the first roll's continuation finished — a
// fresh request for the SAME terminal is now approved.
LeadRollover.PendingRollover third = rollover.open(LEAD, "retry after the first roll finished");
LeadRollover.RollDecision thirdDecision = rollover.confirm(LEAD, third.token(), true);
assertTrue(thirdDecision.accepted(), "the claim must have been released once the first "
+ "roll finished: " + thirdDecision.reason() + " / " + thirdDecision.detail());
}
@Test
@DisplayName("[SINGLE-FLIGHT 2] two DIFFERENT lead terminals confirm without either being "
+ "refused, and the continuation runs once per terminal")
void differentLeadTerminalsConfirmIndependently() throws IOException {
FakeHerdr herdr = new FakeHerdr(); // default idle — both rolls complete
Path handover = writeHandover("handover contents");
AtomicLong clock = new AtomicLong(1_000);
LeadRollover rollover = newRollover(herdr, cfg(handover.toString()), fixedClock(clock));
LeadRollover.PendingRollover a = rollover.open(LEAD, "context is full");
LeadRollover.PendingRollover b = rollover.open(OTHER_LEAD, "context is full too");
LeadRollover.RollDecision aDecision = rollover.confirm(LEAD, a.token(), true);
LeadRollover.RollDecision bDecision = rollover.confirm(OTHER_LEAD, b.token(), true);
assertTrue(aDecision.accepted(), "expected approval; got: " + aDecision.reason() + " / " + aDecision.detail());
assertTrue(bDecision.accepted(), "expected approval; got: " + bDecision.reason() + " / " + bDecision.detail());
assertEquals(4, promptCallCount(herdr), "both rolls ran to completion — /clear and "
+ "bootstrapText, once per terminal");
}
@Test
@DisplayName("[SINGLE-FLIGHT 3] a confirm() refused for OPERATOR_NOT_CONFIRMED does not take "
+ "the single-flight claim — a later valid confirm on the same terminal is still approved")
void refusedConfirmDoesNotTakeTheClaim() throws IOException {
FakeHerdr herdr = new FakeHerdr();
Path handover = writeHandover("handover contents");
AtomicLong clock = new AtomicLong(1_000);
LeadRollover rollover = newRollover(herdr, cfg(handover.toString(), true), fixedClock(clock));
LeadRollover.PendingRollover pending = rollover.open(LEAD, "context is full");
LeadRollover.RollDecision refused = rollover.confirm(LEAD, pending.token(), false);
assertFalse(refused.accepted());
assertEquals(LeadRollover.RefusalReason.OPERATOR_NOT_CONFIRMED, refused.reason());
// the same token stays pending after a gate refusal — retrying it with operator
// confirmation must succeed, proving the earlier refusal never took the claim.
LeadRollover.RollDecision approved = rollover.confirm(LEAD, pending.token(), true);
assertTrue(approved.accepted(), "a confirm() refused for an existing gate must never have "
+ "taken the single-flight claim: " + approved.reason() + " / " + approved.detail());
}
@Test
@DisplayName("[SINGLE-FLIGHT 4] a roll whose continuation throws a RuntimeException still "
+ "releases the claim — a fresh confirm() on that terminal is approved afterwards")
void throwingContinuationStillReleasesTheClaim() throws IOException {
FakeHerdr fake = new FakeHerdr(); // default idle — the turn-settle wait passes immediately
HerdrClient throwsOnClear = new HerdrClient() {
@Override
public JsonNode call(String method, Object params) throws HerdrException {
if ("agent.prompt".equals(method) && String.valueOf(params).contains("/clear")) {
throw new HerdrException("simulated herdr transport failure sending /clear");
}
return fake.call(method, params);
}
@Override
public void close() {
fake.close();
}
};
Path handover = writeHandover("handover contents");
AtomicLong clock = new AtomicLong(1_000);
LeadRollover rollover = newRollover(throwsOnClear, cfg(handover.toString()), fixedClock(clock));
LeadRollover.PendingRollover first = rollover.open(LEAD, "context is full");
LeadRollover.RollDecision decision = rollover.confirm(LEAD, first.token(), true);
assertTrue(decision.accepted(), "every synchronous gate passes; the throw happens only "
+ "inside the deferred continuation, which this test's synchronous runner has "
+ "already run to completion by the time confirm() returns");
LeadRollover.RollStatus status = rollover.status(first.token());
assertEquals(LeadRollover.RollState.FAILED, status.state(), "sanity: the continuation threw "
+ "and left a terminal FAILED outcome: " + status.detail());
LeadRollover.PendingRollover second = rollover.open(LEAD, "retry after the failure");
LeadRollover.RollDecision retryDecision = rollover.confirm(LEAD, second.token(), true);
assertTrue(retryDecision.accepted(), "the claim must have been released even though the "
+ "continuation threw — a release only on the success path would leave this lead "
+ "terminal unrollable forever after one failure: " + retryDecision.reason() + " / "
+ retryDecision.detail());
}
@Test
@DisplayName("[SINGLE-FLIGHT 5] cancel(token) after a successful confirm(token) returns false "
+ "and does not release the claim held by that roll")
void cancelAfterConfirmDoesNotReleaseTheClaim() throws IOException {
FakeHerdr herdr = new FakeHerdr();
Path handover = writeHandover("handover contents");
AtomicLong clock = new AtomicLong(1_000);
HoldingRunner runner = new HoldingRunner();
LeadRollover rollover = newRolloverWithHoldingRunner(herdr, cfg(handover.toString()),
fixedClock(clock), runner);
LeadRollover.PendingRollover pending = rollover.open(LEAD, "context is full");
LeadRollover.RollDecision decision = rollover.confirm(LEAD, pending.token(), true);
assertTrue(decision.accepted(), "expected approval; got: " + decision.reason() + " / " + decision.detail());
assertFalse(rollover.cancel(pending.token()), "confirm() already removed the token from "
+ "pending, so cancel() must find nothing for it");
LeadRollover.PendingRollover another = rollover.open(LEAD,
"a second request while the first roll is still held");
LeadRollover.RollDecision anotherDecision = rollover.confirm(LEAD, another.token(), true);
assertFalse(anotherDecision.accepted(), "cancel() on the already-confirmed token must not "
+ "have released the claim held by that roll");
assertEquals(LeadRollover.RefusalReason.ROLL_ALREADY_RUNNING, anotherDecision.reason());
}
@Test
@DisplayName("[SINGLE-FLIGHT 6] a continuationRunner that rejects the hand-off still releases "
+ "the claim and leaves a terminal FAILED outcome, not a stuck IN_PROGRESS")
void continuationRunnerThatRejectsTheHandOffStillReleasesTheClaim() throws IOException {
FakeHerdr herdr = new FakeHerdr();
Path handover = writeHandover("handover contents");
AtomicLong clock = new AtomicLong(1_000);
AgentControl agents = new AgentControl(herdr);
// A continuationRunner that rejects every hand-off, standing in for e.g. a bounded
// executor's RejectedExecutionException — the continuation never runs, so runRollover's
// own finally never gets a chance to release the claim either.
LeadRollover rollover = new LeadRollover(agents, () -> cfg(handover.toString()), _ -> null,
fixedClock(clock), () -> { },
r -> { throw new RuntimeException("simulated continuationRunner rejection"); });
LeadRollover.PendingRollover pending = rollover.open(LEAD, "context is full");
LeadRollover.RollDecision decision = rollover.confirm(LEAD, pending.token(), true);
assertTrue(decision.accepted(), "every gate passed before the hand-off itself rejected — "
+ "confirm() treats that rejection the same way it treats a throw INSIDE the "
+ "continuation: logged and surfaced only through status(), never as a refusal here: "
+ decision.reason() + " / " + decision.detail());
LeadRollover.RollStatus status = rollover.status(pending.token());
assertEquals(LeadRollover.RollState.FAILED, status.state(), "a rejected hand-off must leave "
+ "a TERMINAL outcome, never a stuck IN_PROGRESS — status() would otherwise have no "
+ "way to tell a roll that never started from one still genuinely running: got "
+ status.state() + " / " + status.detail());
assertNotEquals(LeadRollover.RollState.IN_PROGRESS, status.state());
LeadRollover.PendingRollover retry = rollover.open(LEAD, "retry after the rejected hand-off");
LeadRollover.RollDecision retryDecision = rollover.confirm(LEAD, retry.token(), true);
assertTrue(retryDecision.accepted(), "the claim must have been released even though "
+ "continuationRunner itself threw before the continuation ever ran — otherwise "
+ "this lead terminal could never be rolled again: " + retryDecision.reason() + " / "
+ retryDecision.detail());
}
}
@@ -355,10 +355,10 @@ class FleetMcpAuthzTest {
+ "the anchors have drifted, this test is not testing what it claims to");
Pattern trailingArg = Pattern.compile(
"listFleet\\([^;]*?,\\s*(coordinatorVisibleTo\\(principal\\(exchange\\)\\)|true|false)\\s*[,)]",
"listFleet\\([^;]*?,\\s*(coordinatorVisibleTo\\(principal\\(exchange\\)\\)|true|false)\\s*\\)\\s*;",
Pattern.DOTALL);
Matcher m = trailingArg.matcher(handlerBlock);
assertTrue(m.find(), "could not locate listFleet(...)'s coordinator boolean argument in the "
assertTrue(m.find(), "could not locate listFleet(...)'s trailing boolean argument in the "
+ "listHandler block -- the call shape changed, update this test's anchor: " + handlerBlock);
String trailing = m.group(1);
assertEquals("coordinatorVisibleTo(principal(exchange))", trailing,
@@ -366,197 +366,6 @@ class FleetMcpAuthzTest {
+ "calling, not pass a literal boolean -- found: " + trailing);
}
// --- fleetd #703: who may see fleet_list's collaborators array -------------------------------
/**
* fleetd #703: {@link FleetMcp#collaboratorsVisibleTo} is the whole policy decision for
* {@code fleet_list}'s {@code collaborators} array. Visible to exactly the roles that may
* {@code SEND} to a named peer -- the primary, an architect, and a collaborator itself -- never
* a worker, which holds {@code READ} but can never {@code SEND} to a collaborator, and never an
* anonymous caller.
*/
@Test
void onlyPrimaryArchitectAndCollaboratorMaySeeTheCollaboratorsArray() {
assertTrue(FleetMcp.collaboratorsVisibleTo(PRIMARY), "the primary must see the collaborators array");
assertTrue(FleetMcp.collaboratorsVisibleTo(ARCH_DESIGN), "an architect must see the collaborators array");
assertTrue(FleetMcp.collaboratorsVisibleTo(COLLABORATOR), "a collaborator must see its own peer roster");
assertFalse(FleetMcp.collaboratorsVisibleTo(WORKER_A),
"a worker holds READ but can never SEND to a collaborator, so it must not see the array");
assertFalse(FleetMcp.collaboratorsVisibleTo(ANON), "authenticated as nothing must not see it either");
}
/**
* fleetd #703, same reasoning as {@link #theFleetListHandlerActuallyConsultsCoordinatorVisibleTo}:
* the predicate above can be perfectly correct while the one production call site never asks it.
* This reads {@code FleetMcp.java}'s own source and asserts the {@code fleet_list} handler's
* {@code listFleet(...)} call both threads {@code callers.collaborators()} into the payload and
* asks {@code collaboratorsVisibleTo(principal(exchange))} for the visibility flag, rather than a
* literal boolean or an empty map.
*/
@Test
void theFleetListHandlerActuallyConsultsCollaboratorsVisibleTo() throws Exception {
String source = Files.readString(MCP_SOURCE);
int start = source.indexOf("listHandler =");
assertTrue(start >= 0, "could not find the fleet_list handler (listHandler) in " + MCP_SOURCE
+ " -- the scrape has stopped matching, fix the anchor before trusting this test");
int end = source.indexOf("stopHandler =", start);
assertTrue(end > start, "could not find the handler declared after listHandler to bound the scrape");
String handlerBlock = source.substring(start, end);
// CONTROL: the block we scraped really does contain a call to listFleet(...) -- if this
// fails, the anchors above moved and the assertions below would otherwise pass on nothing.
assertTrue(handlerBlock.contains("listFleet("),
"control failed: the scraped listHandler block contains no listFleet( call at all -- "
+ "the anchors have drifted, this test is not testing what it claims to");
assertTrue(handlerBlock.contains("callers.collaborators()"),
"the fleet_list handler must thread callers.collaborators() into listFleet(...), not an "
+ "empty or literal map -- block: " + handlerBlock);
assertTrue(handlerBlock.contains("collaboratorsVisibleTo(principal(exchange))"),
"the fleet_list handler must ask collaboratorsVisibleTo(principal(exchange)) who is "
+ "calling, not pass a literal boolean -- block: " + handlerBlock);
}
// --- who may see fleet_list's leads and members arrays ---------------------------------------
/**
* {@link FleetMcp#leadsVisibleTo} is the whole policy decision for {@code fleet_list}'s
* {@code leads} array: visible to exactly the roles that may {@code SEND} to a lead -- the
* primary, an architect, and a collaborator -- never a worker, which holds {@code READ} but
* can never {@code SEND} at all, and never an anonymous caller.
*/
@Test
void primaryArchitectAndCollaboratorMaySeeTheLeadsArray() {
assertTrue(FleetMcp.leadsVisibleTo(PRIMARY), "the primary must see the leads array");
assertTrue(FleetMcp.leadsVisibleTo(ARCH_DESIGN), "an architect must see the leads array");
assertTrue(FleetMcp.leadsVisibleTo(COLLABORATOR),
"a collaborator may SEND to a lead, so it must see the leads array to learn where");
assertFalse(FleetMcp.leadsVisibleTo(WORKER_A),
"a worker holds READ but can never SEND, so it must not see the leads array");
assertFalse(FleetMcp.leadsVisibleTo(ANON), "authenticated as nothing must not see it either");
}
/**
* As {@link #primaryArchitectAndCollaboratorMaySeeTheLeadsArray}, for the {@code members}
* array -- but a collaborator may {@code SEND} only to a lead or another collaborator, never
* to a spawned member, so it must not see this one.
*/
@Test
void onlyPrimaryAndArchitectMaySeeTheMembersArray() {
assertTrue(FleetMcp.membersVisibleTo(PRIMARY), "the primary must see the members array");
assertTrue(FleetMcp.membersVisibleTo(ARCH_DESIGN), "an architect must see the members array");
assertFalse(FleetMcp.membersVisibleTo(WORKER_A),
"a worker holds READ but can never SEND, so it must not see the members array");
assertFalse(FleetMcp.membersVisibleTo(COLLABORATOR),
"a collaborator may SEND to a lead, never to a spawned member, so it must not see the members array");
assertFalse(FleetMcp.membersVisibleTo(ANON), "authenticated as nothing must not see it either");
}
/**
* Same reasoning as {@link #theFleetListHandlerActuallyConsultsCoordinatorVisibleTo}: the
* predicate above can be perfectly correct while the one production call site never asks it.
* This reads {@code FleetMcp.java}'s own source and asserts the {@code fleet_list} handler's
* {@code listFleet(...)} call asks {@code leadsVisibleTo(principal(exchange))} for the leads
* visibility flag, rather than a literal boolean.
*/
@Test
void theFleetListHandlerActuallyConsultsLeadsVisibleTo() throws Exception {
String source = Files.readString(MCP_SOURCE);
int start = source.indexOf("listHandler =");
assertTrue(start >= 0, "could not find the fleet_list handler (listHandler) in " + MCP_SOURCE
+ " -- the scrape has stopped matching, fix the anchor before trusting this test");
int end = source.indexOf("stopHandler =", start);
assertTrue(end > start, "could not find the handler declared after listHandler to bound the scrape");
String handlerBlock = source.substring(start, end);
// CONTROL: the block we scraped really does contain a call to listFleet(...) -- if this
// fails, the anchors above moved and the assertion below would otherwise pass on nothing.
assertTrue(handlerBlock.contains("listFleet("),
"control failed: the scraped listHandler block contains no listFleet( call at all -- "
+ "the anchors have drifted, this test is not testing what it claims to");
assertTrue(handlerBlock.contains("leadsVisibleTo(principal(exchange))"),
"the fleet_list handler must ask leadsVisibleTo(principal(exchange)) who is calling, "
+ "not pass a literal boolean -- block: " + handlerBlock);
}
/** As {@link #theFleetListHandlerActuallyConsultsLeadsVisibleTo}, for {@code membersVisibleTo}. */
@Test
void theFleetListHandlerActuallyConsultsMembersVisibleTo() throws Exception {
String source = Files.readString(MCP_SOURCE);
int start = source.indexOf("listHandler =");
assertTrue(start >= 0, "could not find the fleet_list handler (listHandler) in " + MCP_SOURCE
+ " -- the scrape has stopped matching, fix the anchor before trusting this test");
int end = source.indexOf("stopHandler =", start);
assertTrue(end > start, "could not find the handler declared after listHandler to bound the scrape");
String handlerBlock = source.substring(start, end);
assertTrue(handlerBlock.contains("listFleet("),
"control failed: the scraped listHandler block contains no listFleet( call at all -- "
+ "the anchors have drifted, this test is not testing what it claims to");
assertTrue(handlerBlock.contains("membersVisibleTo(principal(exchange))"),
"the fleet_list handler must ask membersVisibleTo(principal(exchange)) who is calling, "
+ "not pass a literal boolean -- block: " + handlerBlock);
}
/**
* {@code fleet_poll{ticket}} must thread the calling connection's own terminal into
* {@link MessageService#poll(String, String)}, so a worker cannot read a ticket a different
* session created.
*/
@Test
void theFleetPollHandlerActuallyThreadsCallerTerminalIntoPoll() throws Exception {
String source = Files.readString(MCP_SOURCE);
int start = source.indexOf("pollHandler =");
assertTrue(start >= 0, "could not find the fleet_poll handler (pollHandler) in " + MCP_SOURCE
+ " -- the scrape has stopped matching, fix the anchor before trusting this test");
int end = source.indexOf("ackHandler =", start);
assertTrue(end > start, "could not find the handler declared after pollHandler to bound the scrape");
String handlerBlock = source.substring(start, end);
// CONTROL: the block we scraped really does call poll(...) -- if this fails, the anchors
// above moved and the assertions below would otherwise pass on nothing.
assertTrue(handlerBlock.contains("poll(messages,"),
"control failed: the scraped pollHandler block contains no poll(messages, ...) call "
+ "at all -- the anchors have drifted, this test is not testing what it claims to");
assertTrue(handlerBlock.contains("callerTerminal(exchange)"),
"the fleet_poll handler must thread callerTerminal(exchange) into poll(...), not omit "
+ "it or pass a literal null -- block: " + handlerBlock);
}
/**
* {@code fleet_status} must thread the calling connection's own terminal into
* {@link FleetMcp#status(MessageService, String, String)}, so a caller that did not create a
* worker's open delegation cannot read its pending question through the status handler either.
*/
@Test
void theFleetStatusHandlerActuallyThreadsCallerTerminalIntoStatus() throws Exception {
String source = Files.readString(MCP_SOURCE);
int start = source.indexOf("statusHandler =");
assertTrue(start >= 0, "could not find the fleet_status handler (statusHandler) in " + MCP_SOURCE
+ " -- the scrape has stopped matching, fix the anchor before trusting this test");
int end = source.indexOf("pollHandler =", start);
assertTrue(end > start, "could not find the handler declared after statusHandler to bound the scrape");
String handlerBlock = source.substring(start, end);
// CONTROL: the block we scraped really does call status(...) -- if this fails, the anchors
// above moved and the assertions below would otherwise pass on nothing.
assertTrue(handlerBlock.contains("status(messages,"),
"control failed: the scraped statusHandler block contains no status(messages, ...) "
+ "call at all -- the anchors have drifted, this test is not testing what it claims to");
assertTrue(handlerBlock.contains("callerTerminal(exchange)"),
"the fleet_status handler must thread callerTerminal(exchange) into status(...), not "
+ "omit it or pass a literal null -- block: " + handlerBlock);
}
// --- which action each tool hands the gate (fleetd #272) ------------------------------------
/**
@@ -121,7 +121,6 @@ class FleetMcpLeadContextGaugeWiringTest {
FleetMcp.CapacitySource.none(), new FleetMcp.HealthCoverageSource(() -> "off"),
FleetMcp.LoopHealthSource.none(), FleetMcp.QuarantineSource.none(), FleetMcp.OutageSource.none(),
FleetMcp.LeadSeatSource.none(), contextGauge, leadConfigDirs,
Map.of(LEAD_TERMINAL, LEAD_NAME), LEAD_TERMINAL, Map.of(), false,
FleetMcp.CoordinationSource.none(), false, true, true);
Map.of(LEAD_TERMINAL, LEAD_NAME), LEAD_TERMINAL, FleetMcp.CoordinationSource.none(), false);
}
}
@@ -10,7 +10,6 @@ import dev.ltms.fleet.herdr.AgentControl;
import dev.ltms.fleet.herdr.AgentStatus;
import dev.ltms.fleet.herdr.FakeHerdr;
import dev.ltms.fleet.inject.LoopWatchdog;
import dev.ltms.fleet.lead.LeadContextGauge;
import dev.ltms.fleet.herdr.PaneLocator;
import dev.ltms.fleet.herdr.WorkspaceControl;
import dev.ltms.fleet.inject.Injector;
@@ -77,7 +76,7 @@ class FleetMcpTest {
private void assertSendRoundTrips(String target, Set<String> profiles) throws Exception {
CompletableFuture<McpSchema.CallToolResult> send = CompletableFuture.supplyAsync(
() -> FleetMcp.send(messages, target, "hi", 4000L, null, profiles, null));
() -> FleetMcp.send(messages, target, "hi", 4000L, null, profiles));
long deadline = System.currentTimeMillis() + 3000;
while (!rendezvous.isWaiting(target) && System.currentTimeMillis() < deadline) {
Thread.sleep(5);
@@ -103,7 +102,7 @@ class FleetMcpTest {
void sendThenReplyRoundTrips() throws Exception {
// fleet_send blocks; fleet_reply resolves it with the worker's structured answer.
CompletableFuture<McpSchema.CallToolResult> send = CompletableFuture.supplyAsync(
() -> FleetMcp.send(messages, "term_a", "review this", 4000L, null, Set.of(), null));
() -> FleetMcp.send(messages, "term_a", "review this", 4000L, null, Set.of()));
// Wait until the send has opened its waiter so the reply resolves it (CB-307: reply now
// queues in the inbox if no waiter is open, which would break the round-trip).
@@ -181,7 +180,7 @@ class FleetMcpTest {
String turnId = afterTurnId.substring(0, afterTurnId.indexOf('"'));
CompletableFuture<McpSchema.CallToolResult> answer = CompletableFuture.supplyAsync(
() -> FleetMcp.answer(messages, turnId, "config.yaml", 5000L, null));
() -> FleetMcp.answer(messages, turnId, "config.yaml", 5000L));
assertEquals("config.yaml", textOf(ask.get(6, TimeUnit.SECONDS)));
deadline = System.currentTimeMillis() + 3000;
@@ -279,7 +278,7 @@ class FleetMcpTest {
assertEquals("late reply", messages.drainReplies("term_a").getFirst().content());
CompletableFuture<McpSchema.CallToolResult> answer = CompletableFuture.supplyAsync(
() -> FleetMcp.answer(messages, firstTurnId, "config.yaml", 5000L, null));
() -> FleetMcp.answer(messages, firstTurnId, "config.yaml", 5000L));
assertEquals("config.yaml", textOf(ask.get(6, TimeUnit.SECONDS)));
while (!rendezvous.isWaiting("term_a") && System.currentTimeMillis() < deadline) {
Thread.sleep(5);
@@ -288,89 +287,6 @@ class FleetMcpTest {
assertEquals("done", textOf(answer.get(6, TimeUnit.SECONDS)));
}
// --- fleetd #715: fleet_send{turnId} is gated on the caller that owns the turn -------------
/**
* A blocking {@code fleet_send} from one caller opens the turn; a {@code fleet_send{turnId}}
* from a different caller is refused as an error, and the real owner's answer still succeeds.
*/
@Test
void aDifferentCallersMcpAnswerIsRefusedForABlockingSendButTheRealOwnerSucceeds() throws Exception {
CompletableFuture<McpSchema.CallToolResult> send = CompletableFuture.supplyAsync(
() -> FleetMcp.send(messages, T, "do X", 5000L, null, Set.of(), "term_owner"));
long deadline = System.currentTimeMillis() + 3000;
while (!rendezvous.isWaiting(T) && System.currentTimeMillis() < deadline) {
Thread.sleep(5);
}
assertTrue(rendezvous.isWaiting(T));
CompletableFuture<McpSchema.CallToolResult> ask = CompletableFuture.supplyAsync(
() -> FleetMcp.ask(messages, T, "which config?", 5000L));
McpSchema.CallToolResult question = send.get(5, TimeUnit.SECONDS);
assertTrue(textOf(question).contains("[question]"), textOf(question));
String questionText = textOf(question);
String afterTurnId = questionText.substring(questionText.indexOf("turnId=\"") + "turnId=\"".length());
String turnId = afterTurnId.substring(0, afterTurnId.indexOf('"'));
McpSchema.CallToolResult hijacked = FleetMcp.answer(messages, turnId, "evil.yaml", 500L, "term_attacker");
assertTrue(hijacked.isError(), "a caller that did not open this turn must get an error, not an answer");
assertFalse(ask.isDone(), "a refused answer must not resolve the worker's blocked fleet_ask");
CompletableFuture<McpSchema.CallToolResult> answer = CompletableFuture.supplyAsync(
() -> FleetMcp.answer(messages, turnId, "config.yaml", 5000L, "term_owner"));
assertEquals("config.yaml", textOf(ask.get(5, TimeUnit.SECONDS)));
deadline = System.currentTimeMillis() + 3000;
while (!rendezvous.isWaiting(T) && System.currentTimeMillis() < deadline) {
Thread.sleep(5);
}
FleetMcp.reply(messages, T, Role.WORKER, "done");
assertEquals("done", textOf(answer.get(5, TimeUnit.SECONDS)));
}
/**
* Same hijack and control through the fire-and-poll ({@code sendAsync}) path: the owner comes
* from the ticket's recorded creator terminal, not from a caller threaded through a live call.
*/
@Test
void aDifferentCallersMcpAnswerIsRefusedForAnAsyncSendButTheRealOwnerSucceeds() throws Exception {
McpSchema.CallToolResult accepted =
FleetMcp.sendAsync(messages, T, "do it", null, Set.of(), "term_owner");
String ticket = textOf(accepted).substring(textOf(accepted).indexOf("ticket=") + "ticket=".length()).trim();
long deadline = System.currentTimeMillis() + 3000;
while (!rendezvous.isWaiting(T) && System.currentTimeMillis() < deadline) {
Thread.sleep(5);
}
assertTrue(rendezvous.isWaiting(T));
CompletableFuture<McpSchema.CallToolResult> ask = CompletableFuture.supplyAsync(
() -> FleetMcp.ask(messages, T, "which config?", 5000L));
MessageService.TaskView asking = messages.poll(ticket);
deadline = System.currentTimeMillis() + 3000;
while (asking.phase() != MessageService.Phase.ASKING && System.currentTimeMillis() < deadline) {
Thread.sleep(5);
asking = messages.poll(ticket);
}
assertEquals(MessageService.Phase.ASKING, asking.phase());
String turnId = asking.turnId();
McpSchema.CallToolResult hijacked = FleetMcp.answer(messages, turnId, "evil.yaml", 500L, "term_attacker");
assertTrue(hijacked.isError(), "a caller that did not create this delegation must get an error");
assertFalse(ask.isDone(), "a refused answer must not resolve the worker's blocked fleet_ask");
assertEquals(MessageService.Phase.ASKING, messages.poll(ticket).phase(),
"a refused answer must not advance the async ticket's phase");
CompletableFuture<McpSchema.CallToolResult> answer = CompletableFuture.supplyAsync(
() -> FleetMcp.answer(messages, turnId, "config.yaml", 5000L, "term_owner"));
assertEquals("config.yaml", textOf(ask.get(5, TimeUnit.SECONDS)));
deadline = System.currentTimeMillis() + 3000;
while (!rendezvous.isWaiting(T) && System.currentTimeMillis() < deadline) {
Thread.sleep(5);
}
FleetMcp.reply(messages, T, Role.WORKER, "done");
assertEquals("done", textOf(answer.get(5, TimeUnit.SECONDS)));
}
@Test
void pollUnknownTicketIsAnError() {
McpSchema.CallToolResult res = FleetMcp.poll(messages, "task-999", null);
@@ -380,7 +296,7 @@ class FleetMcpTest {
@Test
void sendTimesOutWithAWorkingNote() {
McpSchema.CallToolResult res = FleetMcp.send(messages, "term_a", "hi", 120L, null, Set.of(), null);
McpSchema.CallToolResult res = FleetMcp.send(messages, "term_a", "hi", 120L, null, Set.of());
assertNotEquals(Boolean.TRUE, res.isError(), "a timeout is informational, not a tool error");
assertTrue(textOf(res).contains("no reply"), "got: " + textOf(res));
}
@@ -396,7 +312,7 @@ class FleetMcpTest {
void sendTimesOutWithAnUnconfirmedNoteNotARetryInvitation() throws Exception {
herdr.agentSendFailsWith("send_failed");
CompletableFuture<McpSchema.CallToolResult> send = CompletableFuture.supplyAsync(
() -> FleetMcp.send(messages, T, "hi", 150L, null, Set.of(), null));
() -> FleetMcp.send(messages, T, "hi", 150L, null, Set.of()));
long deadline = System.currentTimeMillis() + 2000;
while (!rendezvous.isWaiting(T) && System.currentTimeMillis() < deadline) {
//noinspection BusyWait
@@ -416,13 +332,13 @@ class FleetMcpTest {
@Test
void sendRejectsMissingArgs() {
assertTrue(FleetMcp.send(messages, null, "hi", null, null, Set.of(), null).isError());
assertTrue(FleetMcp.send(messages, "term_a", " ", null, null, Set.of(), null).isError());
assertTrue(FleetMcp.send(messages, null, "hi", null, null, Set.of()).isError());
assertTrue(FleetMcp.send(messages, "term_a", " ", null, null, Set.of()).isError());
}
@Test
void sendRejectsAConfiguredProfileNameBeforeAcceptingIt() {
McpSchema.CallToolResult blocking = FleetMcp.send(messages, "sol", "hi", 100L, null, Set.of("sol"), null);
McpSchema.CallToolResult blocking = FleetMcp.send(messages, "sol", "hi", 100L, null, Set.of("sol"));
McpSchema.CallToolResult async = FleetMcp.sendAsync(messages, "sol", "hi", null, Set.of("sol"));
assertTrue(blocking.isError());
@@ -441,7 +357,7 @@ class FleetMcpTest {
assertSendRoundTrips("term_live_member", profiles);
// A herdr-owned pane outside the bridge roster cannot be classified at accept time.
McpSchema.CallToolResult result = FleetMcp.send(messages, "external-pane", "hi", 10L, null, profiles, null);
McpSchema.CallToolResult result = FleetMcp.send(messages, "external-pane", "hi", 10L, null, profiles);
assertFalse(result.isError(), "an unclassified target must not be rejected at acceptance time");
}
@@ -533,7 +449,7 @@ class FleetMcpTest {
void askThenAnswerRoundTrips() throws Exception {
// The primary delegates and blocks; wait until its waiter is open before the worker asks.
CompletableFuture<McpSchema.CallToolResult> send = CompletableFuture.supplyAsync(
() -> FleetMcp.send(messages, "term_a", "do X", 5000L, null, Set.of(), null));
() -> FleetMcp.send(messages, "term_a", "do X", 5000L, null, Set.of()));
long deadline = System.currentTimeMillis() + 3000;
while (!rendezvous.isWaiting("term_a") && System.currentTimeMillis() < deadline) {
//noinspection BusyWait
@@ -555,7 +471,7 @@ class FleetMcpTest {
// The primary answers via fleet_send(turnId); this blocks again for the worker's reply.
CompletableFuture<McpSchema.CallToolResult> answer = CompletableFuture.supplyAsync(
() -> FleetMcp.answer(messages, turnId, "config.yaml", 5000L, null));
() -> FleetMcp.answer(messages, turnId, "config.yaml", 5000L));
// The worker's ask returns the answer — it resumes the same turn.
assertEquals("config.yaml", textOf(ask.get(6, TimeUnit.SECONDS)));
@@ -581,7 +497,7 @@ class FleetMcpTest {
@Test
void answerToAStaleTurnIsAnError() {
McpSchema.CallToolResult res = FleetMcp.answer(messages, "term_a#999", "too late", 500L, null);
McpSchema.CallToolResult res = FleetMcp.answer(messages, "term_a#999", "too late", 500L);
assertTrue(res.isError());
assertTrue(textOf(res).contains("no longer open"), textOf(res));
}
@@ -775,7 +691,7 @@ class FleetMcpTest {
workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")), sessions, null,
FleetMcp.CapacitySource.none(), new FleetMcp.HealthCoverageSource(() -> "off"),
FleetMcp.QuarantineSource.none(), FleetMcp.OutageSource.none(), FleetMcp.LeadSeatSource.none(),
Map.of(), "", new FleetMcp.CoordinationSource(channel, List.of()), true, true, true);
Map.of(), "", new FleetMcp.CoordinationSource(channel, List.of()), true);
String out = textOf(res);
// fleetd #361: reports both which coord-id a peer must use to reach ME, and this daemon's
@@ -804,7 +720,7 @@ class FleetMcpTest {
workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")), sessions, null,
FleetMcp.CapacitySource.none(), new FleetMcp.HealthCoverageSource(() -> "off"),
FleetMcp.QuarantineSource.none(), FleetMcp.OutageSource.none(), FleetMcp.LeadSeatSource.none(),
Map.of(), "", new FleetMcp.CoordinationSource(channel, List.of()), true, true, true);
Map.of(), "", new FleetMcp.CoordinationSource(channel, List.of()), true);
String out = textOf(res);
assertTrue(out.contains("\"mailbox\":{\"status\":\"unknown\"}"), out);
@@ -863,82 +779,19 @@ class FleetMcpTest {
SessionManager sessions = new SessionManager(workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")));
FakeLeadChannel channel = new FakeLeadChannel("mac-opus")
.withMailbox("mac-opus", LeadChannel.MailboxState.exists("mac-opus", 0, 1));
Principal worker = Principal.worker("term_a", 1);
boolean callerIsPrimary = Principal.worker("term_a", 1).isPrimary();
McpSchema.CallToolResult res = FleetMcp.listFleet(
workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")), sessions, null,
FleetMcp.CapacitySource.none(), new FleetMcp.HealthCoverageSource(() -> "off"),
FleetMcp.QuarantineSource.none(), FleetMcp.OutageSource.none(), FleetMcp.LeadSeatSource.none(),
Map.of(), "", new FleetMcp.CoordinationSource(channel, List.of()), worker.isPrimary(),
FleetMcp.leadsVisibleTo(worker), FleetMcp.membersVisibleTo(worker));
Map.of(), "", new FleetMcp.CoordinationSource(channel, List.of()), callerIsPrimary);
String out = textOf(res);
assertFalse(out.contains("\"coordinator\""), "a worker must never see the coordinator key at all: " + out);
assertFalse(out.contains("mac-opus"), "no fragment of the coordinator row may leak either: " + out);
assertFalse(out.contains("\"leads\""), "a worker must never see the leads key at all: " + out);
assertFalse(out.contains("\"members\""), "a worker must never see the members key at all: " + out);
assertTrue(out.contains("\"healthCoverage\""), "the rest of the result must still be present: " + out);
}
/**
* A worker's result must carry neither the {@code leads} nor the {@code members} key, and no
* fragment of either row leaks even though both are fully populated for this call -- a key
* check alone would pass on an implementation that still built the rows and only renamed or
* nested them.
*/
@Test
void listLeaksNoLeadOrMemberRowFragmentToAWorker() {
FakeHerdr h = new FakeHerdr();
FakeWorktrees worktrees = new FakeWorktrees().withRepoRoot("/repo").withPrefix("/wt");
SessionManager sessions = new SessionManager(workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")), worktrees);
MemberSession s = sessions.acquire("ltms-local", null, "/caller/proj", "term_owner",
new WorktreeRequest("cb-304", null));
Principal worker = Principal.worker("term_a", 1);
McpSchema.CallToolResult res = FleetMcp.listFleet(
workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")), sessions, null,
FleetMcp.CapacitySource.none(), new FleetMcp.HealthCoverageSource(() -> "off"),
FleetMcp.QuarantineSource.none(), FleetMcp.OutageSource.none(), FleetMcp.LeadSeatSource.none(),
Map.of("term_lead_x", "mac-opus"), "", FleetMcp.CoordinationSource.none(), worker.isPrimary(),
FleetMcp.leadsVisibleTo(worker), FleetMcp.membersVisibleTo(worker));
String out = textOf(res);
assertFalse(out.contains("\"leads\""), out);
assertFalse(out.contains("\"members\""), out);
assertFalse(out.contains(s.terminalId()), "no member row fragment may leak to a worker: " + out);
assertFalse(out.contains("term_owner"), "no member owner fragment may leak to a worker: " + out);
assertFalse(out.contains("mac-opus"), "no lead row fragment may leak to a worker: " + out);
assertTrue(out.contains("\"healthCoverage\""), out);
}
/**
* A collaborator may {@code SEND} to a lead, so it must see the {@code leads} array -- the
* only place {@code fleet_whoami} does not already give it a lead's address. It may never
* {@code SEND} to a spawned member, so the {@code members} key must stay absent for it, with
* no fragment of a populated member row leaking either.
*/
@Test
void listShowsLeadsButNotMembersToACollaborator() {
FakeHerdr h = new FakeHerdr();
FakeWorktrees worktrees = new FakeWorktrees().withRepoRoot("/repo").withPrefix("/wt");
SessionManager sessions = new SessionManager(workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")), worktrees);
MemberSession s = sessions.acquire("ltms-local", null, "/caller/proj", "term_owner",
new WorktreeRequest("cb-304", null));
Principal collaborator = Principal.collaborator("ops", "term_collab", 600);
McpSchema.CallToolResult res = FleetMcp.listFleet(
workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")), sessions, null,
FleetMcp.CapacitySource.none(), new FleetMcp.HealthCoverageSource(() -> "off"),
FleetMcp.QuarantineSource.none(), FleetMcp.OutageSource.none(), FleetMcp.LeadSeatSource.none(),
Map.of("term_lead_x", "mac-opus"), "", FleetMcp.CoordinationSource.none(), collaborator.isPrimary(),
FleetMcp.leadsVisibleTo(collaborator), FleetMcp.membersVisibleTo(collaborator));
String out = textOf(res);
assertTrue(out.contains("\"leads\""), "a collaborator must see the leads array: " + out);
assertTrue(out.contains("mac-opus"), "a collaborator must see the lead's name/address: " + out);
assertFalse(out.contains("\"members\""), "a collaborator must never see the members key: " + out);
assertFalse(out.contains(s.terminalId()), "no member row fragment may leak to a collaborator: " + out);
assertFalse(out.contains("term_owner"), "no member owner fragment may leak to a collaborator: " + out);
assertTrue(out.contains("\"leads\""), "the rest of the result must still be present: " + out);
assertTrue(out.contains("\"members\""), out);
assertTrue(out.contains("\"healthCoverage\""), out);
}
@@ -954,19 +807,16 @@ class FleetMcpTest {
SessionManager sessions = new SessionManager(workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")));
FakeLeadChannel channel = new FakeLeadChannel("mac-opus")
.withMailbox("mac-opus", LeadChannel.MailboxState.exists("mac-opus", 0, 1));
Principal architect = Principal.architect("lead-designer", "term_design", 400);
boolean callerIsPrimary = Principal.architect("lead-designer", "term_design", 400).isPrimary();
McpSchema.CallToolResult res = FleetMcp.listFleet(
workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")), sessions, null,
FleetMcp.CapacitySource.none(), new FleetMcp.HealthCoverageSource(() -> "off"),
FleetMcp.QuarantineSource.none(), FleetMcp.OutageSource.none(), FleetMcp.LeadSeatSource.none(),
Map.of(), "", new FleetMcp.CoordinationSource(channel, List.of()), architect.isPrimary(),
FleetMcp.leadsVisibleTo(architect), FleetMcp.membersVisibleTo(architect));
Map.of(), "", new FleetMcp.CoordinationSource(channel, List.of()), callerIsPrimary);
String out = textOf(res);
assertFalse(out.contains("\"coordinator\""), "an architect must never see the coordinator key either: " + out);
assertTrue(out.contains("\"leads\""), "an architect must still see the leads array: " + out);
assertTrue(out.contains("\"members\""), "an architect must still see the members array: " + out);
}
/**
@@ -991,7 +841,7 @@ class FleetMcpTest {
workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")), sessions, null,
FleetMcp.CapacitySource.none(), new FleetMcp.HealthCoverageSource(() -> "off"),
FleetMcp.QuarantineSource.none(), FleetMcp.OutageSource.none(), FleetMcp.LeadSeatSource.none(),
Map.of(), "", new FleetMcp.CoordinationSource(channel, List.of()), true, true, true));
Map.of(), "", new FleetMcp.CoordinationSource(channel, List.of()), true));
assertTrue(gatedAsPrimary.contains("\"coordinator\""), gatedAsPrimary);
assertTrue(gatedAsPrimary.contains("\"selfId\":\"mac-opus\""), gatedAsPrimary);
@@ -1000,8 +850,6 @@ class FleetMcpTest {
assertTrue(gatedAsPrimary.contains("\"heldDurable\""), gatedAsPrimary);
assertTrue(gatedAsPrimary.contains("\"held\""), gatedAsPrimary);
assertTrue(gatedAsPrimary.contains("\"peers\""), gatedAsPrimary);
assertTrue(gatedAsPrimary.contains("\"leads\""), "the primary must still see the leads array: " + gatedAsPrimary);
assertTrue(gatedAsPrimary.contains("\"members\""), "the primary must still see the members array: " + gatedAsPrimary);
}
@Test
@@ -1020,7 +868,7 @@ class FleetMcpTest {
workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")), sessions, null,
FleetMcp.CapacitySource.none(), new FleetMcp.HealthCoverageSource(() -> "off"),
FleetMcp.QuarantineSource.none(), FleetMcp.OutageSource.none(), FleetMcp.LeadSeatSource.none(),
Map.of(), "", new FleetMcp.CoordinationSource(channel, List.of()), true, true, true);
Map.of(), "", new FleetMcp.CoordinationSource(channel, List.of()), true);
String out = textOf(res);
assertTrue(out.contains("\"msgId\":\"m1\""), out);
@@ -1031,83 +879,6 @@ class FleetMcpTest {
assertTrue(out.contains("x".repeat(80) + "…"), "expected an 80-char preview with an ellipsis: " + out);
}
// --- fleetd #703: fleet_list's collaborators array -------------------------------------------
/** Calls the canonical {@code listFleet} overload directly, so a test can set the collaborators
* payload and its visibility independently of a real {@code Principal} / MCP exchange. */
private static McpSchema.CallToolResult listFleetWithCollaborators(FakeHerdr h,
Map<String, String> collaborators, boolean collaboratorsVisible) {
SessionManager sessions = new SessionManager(workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")));
return FleetMcp.listFleet(
workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")), sessions, null,
FleetMcp.CapacitySource.none(), new FleetMcp.HealthCoverageSource(() -> "off"),
FleetMcp.LoopHealthSource.none(), FleetMcp.QuarantineSource.none(), FleetMcp.OutageSource.none(),
FleetMcp.LeadSeatSource.none(), new LeadContextGauge(), FleetMcp.LeadConfigDirSource.none(),
Map.of(), "", collaborators, collaboratorsVisible,
FleetMcp.CoordinationSource.none(), false, true, true);
}
/**
* fleetd #703 acceptance A: a visible caller with one configured collaborator gets a
* {@code collaborators} row whose key ({@code CallerResolver.collaborators()}'s
* {@code terminal_id -> name} entry) lands as that row's {@code sessionId}, and {@code leads}/
* {@code members} are unaffected by the new key.
*/
@Test
void listReportsACollaboratorsRowKeyedByTheCollaboratorsTerminalId() {
FakeHerdr h = new FakeHerdr();
String out = textOf(listFleetWithCollaborators(h, Map.of("term_collab", "ops"), true));
assertTrue(out.contains("\"collaborators\":["), out);
assertTrue(out.contains("\"name\":\"ops\""), out);
assertTrue(out.contains("\"sessionId\":\"term_collab\""), out);
assertTrue(out.contains("\"leads\":[]"), out);
assertTrue(out.contains("\"members\":[]"), out);
}
/**
* fleetd #703 acceptance A, control half: with no collaborator configured, the key is absent
* (caller cannot see it) or an empty array (caller can), and {@code leads}/{@code members} are
* unchanged either way.
*/
@Test
void listOmitsOrEmptiesCollaboratorsWhenNoneAreConfigured() {
FakeHerdr h = new FakeHerdr();
String visibleButEmpty = textOf(listFleetWithCollaborators(h, Map.of(), true));
assertTrue(visibleButEmpty.contains("\"collaborators\":[]"), visibleButEmpty);
assertTrue(visibleButEmpty.contains("\"leads\":[]"), visibleButEmpty);
assertTrue(visibleButEmpty.contains("\"members\":[]"), visibleButEmpty);
String notVisible = textOf(listFleetWithCollaborators(h, Map.of(), false));
assertFalse(notVisible.contains("\"collaborators\""), notVisible);
assertTrue(notVisible.contains("\"leads\":[]"), notVisible);
assertTrue(notVisible.contains("\"members\":[]"), notVisible);
}
/**
* fleetd #703 acceptance B: a worker must not see the {@code collaborators} array at all, while
* an architect -- a role that also holds READ, same as a worker -- does see it. Both halves are
* asserted: a test that only checked the worker-hidden half would pass even if the feature were
* never wired up for anyone.
*/
@Test
void listOmitsCollaboratorsForAWorkerAndIncludesThemForAnArchitect() {
FakeHerdr h = new FakeHerdr();
Map<String, String> collaborators = Map.of("term_collab", "ops");
String asWorker = textOf(listFleetWithCollaborators(h, collaborators,
FleetMcp.collaboratorsVisibleTo(Principal.worker("term_w", 1))));
assertFalse(asWorker.contains("\"collaborators\""),
"a worker must not see the collaborators array: " + asWorker);
String asArchitect = textOf(listFleetWithCollaborators(h, collaborators,
FleetMcp.collaboratorsVisibleTo(Principal.architect("design", "term_arch", 2))));
assertTrue(asArchitect.contains("\"collaborators\":["),
"an architect must see the collaborators array: " + asArchitect);
assertTrue(asArchitect.contains("\"sessionId\":\"term_collab\""), asArchitect);
}
/**
* fleetd #421: {@code mailbox.pending} counts only broker-ready messages, so a blocked lead's
* normal, healthy state is {@code "pending": 0} next to a non-empty {@code held[]} — which
@@ -1129,7 +900,7 @@ class FleetMcpTest {
workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")), sessions, null,
FleetMcp.CapacitySource.none(), new FleetMcp.HealthCoverageSource(() -> "off"),
FleetMcp.QuarantineSource.none(), FleetMcp.OutageSource.none(), FleetMcp.LeadSeatSource.none(),
Map.of(), "", new FleetMcp.CoordinationSource(channel, List.of()), true, true, true);
Map.of(), "", new FleetMcp.CoordinationSource(channel, List.of()), true);
String out = textOf(res);
assertTrue(out.contains("\"pending\":0"), out);
@@ -1158,7 +929,7 @@ class FleetMcpTest {
workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")), sessions, null,
FleetMcp.CapacitySource.none(), new FleetMcp.HealthCoverageSource(() -> "off"),
FleetMcp.QuarantineSource.none(), FleetMcp.OutageSource.none(), FleetMcp.LeadSeatSource.none(),
Map.of(), "", new FleetMcp.CoordinationSource(channel, List.of()), true, true, true);
Map.of(), "", new FleetMcp.CoordinationSource(channel, List.of()), true);
String out = textOf(res);
assertTrue(out.contains("\"heldDurable\":false"),
@@ -1227,7 +998,7 @@ class FleetMcpTest {
FleetMcp.CapacitySource.none(), new FleetMcp.HealthCoverageSource(() -> "off"),
FleetMcp.QuarantineSource.none(), FleetMcp.OutageSource.none(), FleetMcp.LeadSeatSource.none(),
Map.of(), "",
new FleetMcp.CoordinationSource(channel, List.of("fleet01-lead", "fleet02-lead", "fleet03-lead")), true, true, true);
new FleetMcp.CoordinationSource(channel, List.of("fleet01-lead", "fleet02-lead", "fleet03-lead")), true);
String out = textOf(res);
assertTrue(out.contains("\"coordId\":\"fleet01-lead\",\"status\":\"exists\",\"pending\":2,\"consumers\":1"), out);
@@ -1920,8 +1691,8 @@ class FleetMcpTest {
Principal architect = Principal.architect("lead-designer", "term_design", 400);
MemberPresence presence = new MemberPresence();
FleetMcp.markTrackedCallerPresent(worker, presence);
FleetMcp.markTrackedCallerPresent(architect, presence);
FleetMcp.markSpawnedMemberPresent(worker, presence);
FleetMcp.markSpawnedMemberPresent(architect, presence);
assertTrue(presence.isPresent("term_worker"));
assertTrue(presence.isPresent("term_design"));
@@ -1932,40 +1703,25 @@ class FleetMcpTest {
Principal lead = Principal.leader("opus", "term_lead", 100);
MemberPresence presence = new MemberPresence();
FleetMcp.markTrackedCallerPresent(lead, presence);
FleetMcp.markTrackedCallerPresent(Principal.anonymous(), presence);
FleetMcp.markSpawnedMemberPresent(lead, presence);
FleetMcp.markSpawnedMemberPresent(Principal.anonymous(), presence);
assertFalse(presence.isPresent("term_lead"));
}
/**
* The item whose absence would be silent: an observer's own MCP contact must still mark
* presence, or a pane resolving to the unconfigured-pane floor would sit on the injector
* readiness gate forever once something addresses it.
*/
@Test
void anObserverContactMarksPresence() {
Principal observer = Principal.observer("term_observer", 800);
MemberPresence presence = new MemberPresence();
FleetMcp.markTrackedCallerPresent(observer, presence);
assertTrue(presence.isPresent("term_observer"));
}
@Test
void statusReportsLiveAgentStatus() {
FakeHerdr blocked = new FakeHerdr().agentStatus("blocked");
AgentControl blockedAgents = new AgentControl(blocked);
McpSchema.CallToolResult res = FleetMcp.status(
new MessageService(blockedAgents, new Injector(blockedAgents), rendezvous), "term_a", null);
new MessageService(blockedAgents, new Injector(blockedAgents), rendezvous), "term_a");
assertNotEquals(Boolean.TRUE, res.isError());
assertEquals("blocked", textOf(res));
}
/**
* A lead polling {@code fleet_status} on its normal cadence — not {@code fleet_poll} — must
* also see a worker's open async {@code fleet_ask} question, since the reverse-rendezvous
* CB-582: a lead polling {@code fleet_status} on its normal cadence — not {@code fleet_poll}
* — must also see a worker's open async {@code fleet_ask} question, since the reverse-rendezvous
* window it opened with is far shorter than that cadence.
*/
@Test
@@ -1988,7 +1744,7 @@ class FleetMcpTest {
} while (asking.phase() != MessageService.Phase.ASKING && System.currentTimeMillis() < deadline);
assertEquals(MessageService.Phase.ASKING, asking.phase());
McpSchema.CallToolResult res = FleetMcp.status(messages, T, null);
McpSchema.CallToolResult res = FleetMcp.status(messages, T);
assertNotEquals(Boolean.TRUE, res.isError());
String out = textOf(res);
assertTrue(out.startsWith("idle"), "the live status must still lead the text: " + out);
@@ -1999,64 +1755,7 @@ class FleetMcpTest {
// Clean up the still-open ask so the background thread does not linger past the test.
String turnId = asking.turnId();
CompletableFuture<MessageService.Reply> answer = CompletableFuture.supplyAsync(
() -> messages.answer(turnId, "config.yaml", 5000, null));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
deadline = System.currentTimeMillis() + 3000;
while (!rendezvous.isWaiting(T) && System.currentTimeMillis() < deadline) {
Thread.sleep(5);
}
assertTrue(rendezvous.resolve(T, "done"));
answer.get(5, TimeUnit.SECONDS);
}
/**
* {@code fleet_status}'s pending-ask block (the question, its {@code turnId} and its ticket)
* is shown only to the caller whose terminal created the delegation, or to a caller with no
* terminal at all (the unnamed primary) — a different terminal-bearing caller still sees the
* base status line, but none of the pending-ask fields.
*/
@Test
void statusGatesThePendingAskFieldsByTheDelegationsCreatorTerminal() throws Exception {
String ticket = messages.sendAsync(T, "task that asks", null, "term_creator");
long deadline = System.currentTimeMillis() + 3000;
while (!rendezvous.isWaiting(T) && System.currentTimeMillis() < deadline) {
Thread.sleep(5);
}
assertTrue(rendezvous.isWaiting(T), "sendAsync should have opened its rendezvous waiter");
CompletableFuture<MessageService.AskResult> ask =
CompletableFuture.supplyAsync(() -> messages.ask(T, "which config file?", 5000));
MessageService.TaskView asking;
deadline = System.currentTimeMillis() + 3000;
do {
asking = messages.poll(ticket);
Thread.sleep(5);
} while (asking.phase() != MessageService.Phase.ASKING && System.currentTimeMillis() < deadline);
assertEquals(MessageService.Phase.ASKING, asking.phase());
String other = textOf(FleetMcp.status(messages, T, "term_other"));
assertTrue(other.startsWith("idle"), "the base status must still be shown: " + other);
assertFalse(other.contains("which config file?"),
"a non-creating caller must not see the question text: " + other);
assertFalse(other.contains(asking.turnId()),
"a non-creating caller must not see the turnId: " + other);
assertFalse(other.contains(ticket),
"a non-creating caller must not see the ticket: " + other);
String creator = textOf(FleetMcp.status(messages, T, "term_creator"));
assertTrue(creator.contains("which config file?"), "the creator must see the question: " + creator);
assertTrue(creator.contains(asking.turnId()), "the creator must see the turnId: " + creator);
assertTrue(creator.contains(ticket), "the creator must see the ticket: " + creator);
String unnamed = textOf(FleetMcp.status(messages, T, null));
assertTrue(unnamed.contains("which config file?"),
"a caller with no terminal (the unnamed primary) must see the question: " + unnamed);
// Clean up the still-open ask so the background thread does not linger past the test.
String turnId = asking.turnId();
CompletableFuture<MessageService.Reply> answer = CompletableFuture.supplyAsync(
() -> messages.answer(turnId, "config.yaml", 5000, "term_creator"));
() -> messages.answer(turnId, "config.yaml", 5000));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
deadline = System.currentTimeMillis() + 3000;
while (!rendezvous.isWaiting(T) && System.currentTimeMillis() < deadline) {
@@ -2164,27 +1863,6 @@ class FleetMcpTest {
assertTrue(leadOut.contains("\"leader\":\"opus\""), leadOut);
}
/**
* An observer reports its own role and pane, never a {@code leader} key. Without an explicit
* branch it would reach the lead branch by elimination and look right only because the
* {@code leader} key is guarded on a non-null name — this pins the branch rather than the
* accident.
*/
@Test
void whoamiReportsAnObserverNotALead() {
FakeHerdr h = new FakeHerdr();
SessionManager sessions = sessionManager(h, "http://gx00.gw:8000", Set.of("gx00.gw"));
McpSchema.CallToolResult res = FleetMcp.whoami(
Principal.observer("term_observer", 900), sessions);
assertNotEquals(Boolean.TRUE, res.isError());
String out = textOf(res);
assertTrue(out.contains("\"role\":\"observer\""), out);
assertTrue(out.contains("\"sessionId\":\"term_observer\""), out);
assertFalse(out.contains("leader"), out);
}
/**
* CB-548: an architect SEND delegates as its own pane (recording the per-target delegation) but
* must NEVER become the legacy singleton "primary" fallback — the per-target map does not cure
@@ -1,280 +0,0 @@
package dev.ltms.fleet.msg;
import org.junit.jupiter.api.Test;
import java.io.IOException;
import java.nio.file.Files;
import java.nio.file.Path;
import java.util.ArrayList;
import java.util.List;
import java.util.regex.Matcher;
import java.util.regex.Pattern;
import java.util.stream.Stream;
import static org.junit.jupiter.api.Assertions.assertEquals;
import static org.junit.jupiter.api.Assertions.assertTrue;
/**
* Pins that no file under {@code src/main/java} calls the fail-open
* {@link MessageService#poll(String)} overload. That overload skips the ownership check in
* {@code MessageService}'s {@code ownsTicket} entirely, so a caller of it can read any session's
* ticket. Every production caller must go through {@link MessageService#poll(String, String)}
* and pass a {@code callerTerminal} explicitly, even when it is {@code null}.
*
* <p>This reads each file's own source text rather than reflecting on compiled bytecode, because
* the risk is a future one-word edit at a call site, not a missing overload.
*
* <p>The scan below finds a violation by its receiver, {@code messages.poll(}, rather than the
* bare method name, so it does not mistake {@link java.util.Queue#poll()} for a violation. That
* anchor only covers a {@code MessageService} reached through a variable or field named
* {@code messages}, so {@link #everyMessageServiceDeclarationIsNamedMessages} pins the naming
* convention the anchor depends on: a declaration under any other name would be invisible to the
* scan above, and must turn this second check red instead of passing silently.
*/
class MessageServicePollUsageTest {
private static final Path PRODUCTION_SOURCE = Path.of("src/main/java");
@Test
void noProductionFileCallsTheSingleArgumentPollOverload() throws IOException {
List<String> violations = new ArrayList<>();
List<String> twoArgSites = new ArrayList<>();
int filesScanned = scanForPollCalls(PRODUCTION_SOURCE, violations, twoArgSites);
// CONTROL: the scan actually walked files -- a wrong root would otherwise report "no
// violations found" having looked at nothing.
assertTrue(filesScanned > 0, "control failed: the scan under " + PRODUCTION_SOURCE
+ " visited zero .java files -- the path is wrong, so the absence of violations "
+ "below proves nothing");
assertTrue(violations.isEmpty(), "found a call to the fail-open MessageService.poll(String) "
+ "overload, which skips the ownership check entirely -- pass a callerTerminal "
+ "explicitly (even if null) through poll(String, String) instead: " + violations);
// CONTROL: the arity parser actually finds the two genuine two-argument call sites (the
// MCP handler in FleetMcp and the REST handler in FleetApp). If this drops, the parser
// itself is broken, not the production code -- a broken parser (or a scan root that
// reaches no real source) must fail loudly here rather than pass vacuously above.
assertEquals(2, twoArgSites.size(), "control failed: expected exactly the two known "
+ "two-argument messages.poll(...) call sites, found: " + twoArgSites);
assertTrue(twoArgSites.stream().anyMatch(s -> s.contains("FleetMcp.java")),
"control failed: did not find the FleetMcp.java messages.poll(ticket, callerTerminal) "
+ "site among: " + twoArgSites);
assertTrue(twoArgSites.stream().anyMatch(s -> s.contains("FleetApp.java")),
"control failed: did not find the FleetApp.java messages.poll(...) site among: "
+ twoArgSites);
}
/**
* The {@code messages.poll(} anchor above only sees a {@code MessageService} reached through
* a variable, field, or parameter named {@code messages}. This asserts that every such
* declaration under {@code src/main/java} uses that name, so a differently named declaration
* -- invisible to the scan above -- fails loudly here instead of letting that scan pass on a
* call site it never looked at.
*/
@Test
void everyMessageServiceDeclarationIsNamedMessages() throws IOException {
List<String> names = new ArrayList<>();
int filesScanned = scanForDeclarationNames(PRODUCTION_SOURCE, names);
// CONTROL: the scan actually walked files -- a wrong root would otherwise report "every
// declaration is named messages" having looked at nothing.
assertTrue(filesScanned > 0, "control failed: the scan under " + PRODUCTION_SOURCE
+ " visited zero .java files -- the path is wrong, so the result below proves nothing");
// CONTROL: the declaration pattern actually finds real declarations. Zero means the
// pattern is broken, not that every MessageService variable, field, or parameter vanished.
assertTrue(names.size() > 0, "control failed: found zero MessageService declarations under "
+ PRODUCTION_SOURCE + " -- the declaration pattern is broken, update it before "
+ "trusting the naming check below");
List<String> other = names.stream().filter(n -> !n.equals("messages")).distinct().toList();
assertTrue(other.isEmpty(), "found a MessageService declaration not named \"messages\": "
+ other + " -- the messages.poll( scan above only looks for that name, so a call "
+ "through a differently named variable or field is invisible to it; either rename "
+ "the declaration or widen that scan's anchor to cover it");
}
private static int scanForPollCalls(Path root, List<String> violations, List<String> twoArgSites)
throws IOException {
List<Path> files = javaFiles(root);
for (Path file : files) {
scanFileForPollCalls(file, violations, twoArgSites);
}
return files.size();
}
private static int scanForDeclarationNames(Path root, List<String> names) throws IOException {
List<Path> files = javaFiles(root);
Pattern declaration = Pattern.compile("MessageService\\s+([A-Za-z_][A-Za-z0-9_]*)");
for (Path file : files) {
if (file.getFileName().toString().equals("MessageService.java")) {
continue; // the type's own declaration, not a caller holding a reference to it
}
String stripped = stripComments(Files.readString(file));
Matcher m = declaration.matcher(stripped);
while (m.find()) {
int j = m.end();
while (j < stripped.length() && Character.isWhitespace(stripped.charAt(j))) j++;
if (j < stripped.length() && stripped.charAt(j) == '(') {
continue; // a method named like the convention, e.g. "MessageService messages()"
}
names.add(m.group(1));
}
}
return files.size();
}
private static List<Path> javaFiles(Path root) throws IOException {
try (Stream<Path> paths = Files.walk(root)) {
return paths.filter(p -> p.toString().endsWith(".java")).toList();
}
}
/**
* Replaces {@code //} and {@code /* *}{@code /} comment text with nothing, leaving code,
* string/char literals and line breaks untouched -- so a comment that merely mentions
* {@code MessageService} in prose can never be read as a declaration.
*/
private static String stripComments(String source) {
StringBuilder out = new StringBuilder(source.length());
boolean inString = false;
boolean inChar = false;
int i = 0;
while (i < source.length()) {
char c = source.charAt(i);
if (inString) {
out.append(c);
if (c == '\\' && i + 1 < source.length()) { out.append(source.charAt(i + 1)); i += 2; continue; }
if (c == '"') inString = false;
i++;
continue;
}
if (inChar) {
out.append(c);
if (c == '\\' && i + 1 < source.length()) { out.append(source.charAt(i + 1)); i += 2; continue; }
if (c == '\'') inChar = false;
i++;
continue;
}
if (c == '"') { inString = true; out.append(c); i++; continue; }
if (c == '\'') { inChar = true; out.append(c); i++; continue; }
if (c == '/' && i + 1 < source.length() && source.charAt(i + 1) == '/') {
while (i < source.length() && source.charAt(i) != '\n') i++;
continue; // leaves the newline itself for the next iteration to append
}
if (c == '/' && i + 1 < source.length() && source.charAt(i + 1) == '*') {
i += 2;
while (i < source.length() && !(source.charAt(i) == '*' && i + 1 < source.length()
&& source.charAt(i + 1) == '/')) {
if (source.charAt(i) == '\n') out.append('\n');
i++;
}
i += 2;
continue;
}
out.append(c);
i++;
}
return out.toString();
}
private static void scanFileForPollCalls(Path file, List<String> violations, List<String> twoArgSites)
throws IOException {
String source = Files.readString(file);
String needle = "messages.poll(";
int from = 0;
int idx;
while ((idx = source.indexOf(needle, from)) >= 0) {
int argsStart = idx + needle.length();
String args = extractBalancedArgs(source, argsStart, file, idx);
int closeParenIndex = argsStart + args.length();
from = closeParenIndex + 1;
if (args.isBlank()) {
continue; // MessageService has no zero-argument poll() -- nothing to classify
}
String site = file + ":" + lineOf(source, idx);
if (topLevelCommaCount(args) == 0) {
violations.add(site + " -- messages.poll(" + args.trim() + ")");
} else {
twoArgSites.add(site);
}
}
}
/**
* The text between {@code messages.poll(} and its matching close paren: balanced over nested
* calls, and never split by a paren or comma sitting inside a string or char literal.
*/
private static String extractBalancedArgs(String source, int start, Path file, int callIndex) {
int depth = 1;
boolean inString = false;
boolean inChar = false;
int i = start;
while (i < source.length()) {
char c = source.charAt(i);
if (inString) {
if (c == '\\') { i += 2; continue; }
if (c == '"') inString = false;
} else if (inChar) {
if (c == '\\') { i += 2; continue; }
if (c == '\'') inChar = false;
} else if (c == '"') {
inString = true;
} else if (c == '\'') {
inChar = true;
} else if (c == '(') {
depth++;
} else if (c == ')') {
depth--;
if (depth == 0) return source.substring(start, i);
}
i++;
}
throw new IllegalStateException(
"unbalanced parentheses scanning " + file + ":" + lineOf(source, callIndex));
}
/**
* Commas at paren/bracket/brace depth zero, skipping string and char literals -- the argument
* separators a human reader would see, not every comma character in the text.
*/
private static int topLevelCommaCount(String args) {
int depth = 0;
int commas = 0;
boolean inString = false;
boolean inChar = false;
int i = 0;
while (i < args.length()) {
char c = args.charAt(i);
if (inString) {
if (c == '\\') { i += 2; continue; }
if (c == '"') inString = false;
} else if (inChar) {
if (c == '\\') { i += 2; continue; }
if (c == '\'') inChar = false;
} else if (c == '"') {
inString = true;
} else if (c == '\'') {
inChar = true;
} else if (c == '(' || c == '[' || c == '{') {
depth++;
} else if (c == ')' || c == ']' || c == '}') {
depth--;
} else if (c == ',' && depth == 0) {
commas++;
}
i++;
}
return commas;
}
private static int lineOf(String source, int index) {
int line = 1;
for (int i = 0; i < index; i++) {
if (source.charAt(i) == '\n') line++;
}
return line;
}
}
@@ -78,7 +78,7 @@ class MessageServiceTest {
}
private CompletableFuture<MessageService.Reply> sendAsync(String content, long timeoutMillis) {
return CompletableFuture.supplyAsync(() -> messages.send(T, content, timeoutMillis, null));
return CompletableFuture.supplyAsync(() -> messages.send(T, content, timeoutMillis));
}
private void awaitWaiting() throws InterruptedException {
@@ -188,7 +188,7 @@ class MessageServiceTest {
localRendezvous, new InMemoryReplyInbox());
String brief = "Implement the requested change. ".repeat(20);
CompletableFuture<MessageService.Reply> send =
CompletableFuture.supplyAsync(() -> localMessages.send(T, brief, 5000, (String) null));
CompletableFuture.supplyAsync(() -> localMessages.send(T, brief, 5000));
long deadline = System.currentTimeMillis() + 2000;
while (!localRendezvous.isWaiting(T) && System.currentTimeMillis() < deadline) {
Thread.sleep(5);
@@ -341,7 +341,7 @@ class MessageServiceTest {
// The primary answers via fleet_send(turnId); this blocks again for the worker's reply.
CompletableFuture<MessageService.Reply> answer =
CompletableFuture.supplyAsync(() -> messages.answer(q.turnId(), "config.yaml", 5000, null));
CompletableFuture.supplyAsync(() -> messages.answer(q.turnId(), "config.yaml", 5000));
// The worker's ask returns the answer — it resumes the same turn.
MessageService.AskResult a = ask.get(5, TimeUnit.SECONDS);
@@ -377,7 +377,7 @@ class MessageServiceTest {
// The primary answers that one turnId; both asks unblock with the same answer.
CompletableFuture<MessageService.Reply> answer =
CompletableFuture.supplyAsync(() -> messages.answer(q.turnId(), "config.yaml", 5000, null));
CompletableFuture.supplyAsync(() -> messages.answer(q.turnId(), "config.yaml", 5000));
MessageService.AskResult a1 = ask1.get(5, TimeUnit.SECONDS);
MessageService.AskResult a2 = ask2.get(5, TimeUnit.SECONDS);
@@ -472,7 +472,7 @@ class MessageServiceTest {
"the duplicate's own timeout elapses first");
CompletableFuture<MessageService.Reply> answered =
CompletableFuture.supplyAsync(() -> messages.answer(turnId, "fleetd.yaml", 500, null));
CompletableFuture.supplyAsync(() -> messages.answer(turnId, "fleetd.yaml", 500));
MessageService.AskResult a = fresh.get(5, TimeUnit.SECONDS);
assertEquals(MessageService.AskOutcome.ANSWERED, a.outcome(),
@@ -507,7 +507,7 @@ class MessageServiceTest {
CompletableFuture<MessageService.Reply> lateAnswer = new CompletableFuture<>();
messages.setAskTimeoutRaceHookForTest(() ->
lateAnswer.complete(messages.answer(turnId, "too late", 500, null)));
lateAnswer.complete(messages.answer(turnId, "too late", 500)));
try {
MessageService.AskResult a = ask.get(5, TimeUnit.SECONDS);
assertEquals(MessageService.AskOutcome.TIMED_OUT, a.outcome());
@@ -539,7 +539,7 @@ class MessageServiceTest {
@Test
void answeringAnUnknownTurnIsStale() {
MessageService.Reply r = messages.answer(T + "#999", "too late", 500, null);
MessageService.Reply r = messages.answer(T + "#999", "too late", 500);
assertEquals(MessageService.Outcome.STALE_TURN, r.outcome(),
"an answer to a turn that never existed (or already lapsed) is stale, not a hang");
}
@@ -572,7 +572,7 @@ class MessageServiceTest {
messages.setAnswerAskLapseRaceHookForTest(() -> rendezvous.answerAsk(turnId, "raced in first"));
try {
MessageService.Reply r = messages.answer(turnId, "too late", 500, null);
MessageService.Reply r = messages.answer(turnId, "too late", 500);
assertEquals(MessageService.Outcome.STALE_TURN, r.outcome(),
"an ask already answered by the race must be seen as lapsed, not double-delivered");
} finally {
@@ -596,7 +596,7 @@ class MessageServiceTest {
void sendTimesOutBeforeDeliveryIsQueuedNotWorking() {
// Nothing ever delivers the message and nothing resolves the send, so the reply future
// times out with delivery still incomplete — the message is still queued for the worker.
MessageService.Reply r = messages.send(T, "never delivered", 50, null);
MessageService.Reply r = messages.send(T, "never delivered", 50);
assertEquals(MessageService.Outcome.TIMED_OUT_QUEUED, r.outcome(),
"an undelivered send that times out is still queued, not working");
assertNull(r.text());
@@ -605,7 +605,7 @@ class MessageServiceTest {
@Test
void sendTimesOutAfterDeliveryIsStillWorking() throws Exception {
CompletableFuture<MessageService.Reply> send =
CompletableFuture.supplyAsync(() -> messages.send(T, "do the task", 300, null));
CompletableFuture.supplyAsync(() -> messages.send(T, "do the task", 300));
awaitWaiting();
injector.onStatus(T, AgentStatus.IDLE); // deliver — the delivered future now completes
injector.onStatus(T, AgentStatus.WORKING); // worker starts but never replies
@@ -630,7 +630,7 @@ class MessageServiceTest {
void sendTimeoutUsesCancellationDeliveredWhenPickupWinsTheRace() {
messages.setTimeoutCancellationRaceHookForTest(() -> injector.onStatus(T, AgentStatus.IDLE));
try {
MessageService.Reply reply = messages.send(T, "race delivery", 50, null);
MessageService.Reply reply = messages.send(T, "race delivery", 50);
assertEquals(MessageService.Outcome.TIMED_OUT_WORKING, reply.outcome(),
"cancel reporting DELIVERED means the worker received the timed-out message");
@@ -652,7 +652,7 @@ class MessageServiceTest {
void sendTimesOutWithAttemptedDeliveryReportsUnconfirmedNotQueued() throws Exception {
herdr.agentSendFailsWith("send_failed");
CompletableFuture<MessageService.Reply> send =
CompletableFuture.supplyAsync(() -> messages.send(T, "brief", 150, null));
CompletableFuture.supplyAsync(() -> messages.send(T, "brief", 150));
awaitWaiting();
injector.onStatus(T, AgentStatus.IDLE); // triggers the failing delivery attempt → ATTEMPTED
@@ -679,7 +679,7 @@ class MessageServiceTest {
// The primary answers, unblocking the worker; but the worker never sends the follow-up
// fleet_reply, so the answering send rides out its short window as still-working.
MessageService.Reply answer = messages.answer(q.turnId(), "config.yaml", 200, null);
MessageService.Reply answer = messages.answer(q.turnId(), "config.yaml", 200);
assertEquals(MessageService.Outcome.TIMED_OUT_WORKING, answer.outcome(),
"an answered worker that never replies times out as still working");
@@ -715,7 +715,7 @@ class MessageServiceTest {
assertEquals(MessageService.Outcome.QUESTION, q.outcome());
CompletableFuture<MessageService.Reply> answer =
CompletableFuture.supplyAsync(() -> messages.answer(q.turnId(), "config.yaml", 5000, null));
CompletableFuture.supplyAsync(() -> messages.answer(q.turnId(), "config.yaml", 5000));
ask.get(5, TimeUnit.SECONDS); // worker resumed with the answer
awaitWaiting(); // the answering call has (re)opened its own forward waiter
@@ -723,7 +723,7 @@ class MessageServiceTest {
MessageService.Reply done = answer.get(5, TimeUnit.SECONDS);
assertEquals(MessageService.Outcome.REPLIED, done.outcome());
MessageService.Reply probe = messages.send(T, "probe after normal reply", 300, null);
MessageService.Reply probe = messages.send(T, "probe after normal reply", 300);
assertNotEquals(MessageService.Outcome.BUSY, probe.outcome(),
"the session lock must be released after a normal REPLIED answer(), or this bounded "
+ "follow-up send would come back BUSY instead of timing out on its own work");
@@ -756,13 +756,13 @@ class MessageServiceTest {
// The primary answers, unblocking the worker; the worker never sends its follow-up
// fleet_reply, so the answering call rides out its short window as still-working.
CompletableFuture<MessageService.Reply> answer =
CompletableFuture.supplyAsync(() -> messages.answer(q.turnId(), "config.yaml", 200, null));
CompletableFuture.supplyAsync(() -> messages.answer(q.turnId(), "config.yaml", 200));
MessageService.Reply answered = answer.get(5, TimeUnit.SECONDS);
assertEquals(MessageService.Outcome.TIMED_OUT_WORKING, answered.outcome(),
"an answered worker that never replies times out as still working");
ask.get(5, TimeUnit.SECONDS); // drain: the worker resumed with the answer
MessageService.Reply probe = messages.send(T, "probe after timed-out-working", 300, null);
MessageService.Reply probe = messages.send(T, "probe after timed-out-working", 300);
assertNotEquals(MessageService.Outcome.BUSY, probe.outcome(),
"the session lock must be released after a TIMED_OUT_WORKING answer(), or this bounded "
+ "follow-up send would come back BUSY instead of timing out on its own work");
@@ -791,7 +791,7 @@ class MessageServiceTest {
new java.util.concurrent.atomic.AtomicReference<>();
Thread answerer = new Thread(() -> {
try {
messages.answer(turnId, "config.yaml", 5000, null);
messages.answer(turnId, "config.yaml", 5000);
caught.set(new AssertionError("expected answer() to throw"));
} catch (Throwable t) {
caught.set(t);
@@ -813,7 +813,7 @@ class MessageServiceTest {
ask.get(5, TimeUnit.SECONDS); // drain: the worker resumed with the answer
MessageService.Reply probe = messages.send(T, "probe after exceptional failure", 300, null);
MessageService.Reply probe = messages.send(T, "probe after exceptional failure", 300);
assertNotEquals(MessageService.Outcome.BUSY, probe.outcome(),
"the session lock must be released when answer()'s reply future fails exceptionally, "
+ "or this bounded follow-up send would come back BUSY instead of timing out on its own work");
@@ -840,7 +840,7 @@ class MessageServiceTest {
new java.util.concurrent.atomic.AtomicReference<>();
Thread answerer = new Thread(() -> {
try {
messages.answer(turnId, "config.yaml", 5000, null);
messages.answer(turnId, "config.yaml", 5000);
caught.set(new AssertionError("expected answer() to throw"));
} catch (Throwable t) {
caught.set(t);
@@ -859,121 +859,17 @@ class MessageServiceTest {
ask.get(5, TimeUnit.SECONDS); // drain: the worker resumed with the answer
MessageService.Reply probe = messages.send(T, "probe after interruption", 300, null);
MessageService.Reply probe = messages.send(T, "probe after interruption", 300);
assertNotEquals(MessageService.Outcome.BUSY, probe.outcome(),
"the session lock must be released when answer()'s wait is interrupted, or this bounded "
+ "follow-up send would come back BUSY instead of timing out on its own work");
}
/**
* Drive an async send on {@code T} to a resolved reply, polling as {@code owner} until the
* ticket reports {@link MessageService.Phase#DONE} (or the 2s deadline runs out). {@code owner}
* must be a terminal this ticket's creator check actually accepts, or this loops to the
* deadline and returns a non-{@code DONE} view.
*/
private MessageService.TaskView driveAsyncTicketToDone(String ticket, String owner) throws InterruptedException {
MessageService.TaskView view = null;
long deadline = System.currentTimeMillis() + 2000;
while (view == null || view.phase() != MessageService.Phase.DONE) {
if (System.currentTimeMillis() >= deadline) break;
view = messages.poll(ticket, owner);
//noinspection BusyWait
Thread.sleep(5);
}
return view;
}
@Test
void pollReturnsNullForAnUnknownTicket() {
assertNull(messages.poll("task-999999"), "a ticket that was never minted is unknown");
}
// --- fleetd #719: a per-boot nonce keeps one instance's ticket ids out of another's space ---
/** A second, fully independent instance — its own agents/injector/rendezvous/inbox, not shared. */
private MessageService newIndependentInstance() {
FakeHerdr otherHerdr = new FakeHerdr().readText("BUILD GREEN: 391 files");
AgentControl otherAgents = new AgentControl(otherHerdr);
Injector otherInjector = new Injector(otherAgents);
return new MessageService(otherAgents, otherInjector, new Rendezvous(), new InMemoryReplyInbox());
}
@Test
void twoInstancesMintDisjointTicketIds() {
MessageService other = newIndependentInstance();
String ticketFromThis = messages.sendAsync(T, "task on first instance", null, null);
String ticketFromOther = other.sendAsync(T, "task on second instance", null, null);
assertNotEquals(ticketFromThis, ticketFromOther,
"each instance mints its own id space, so even a first ticket from each must differ");
}
@Test
void foreignInstanceTicketDoesNotResolve() {
MessageService other = newIndependentInstance();
String ticket = messages.sendAsync(T, "task on first instance", null, null);
// `other` must reach the same sequence number, or this test passes against an empty map
// instead of against a colliding id.
other.sendAsync(T, "task on second instance", null, null);
// control: the id resolves in the instance that minted it, so a null below cannot be
// explained by broken plumbing — only by the ticket being foreign to `other`.
assertNotNull(messages.poll(ticket), "the minting instance must still resolve its own ticket");
assertNull(other.poll(ticket), "a ticket minted by a different instance must not resolve here");
}
// --- fleetd #705: a ticket's creator terminal gates who may poll it -------------------------
@Test
void pollByAnotherTerminalIsRefused() throws Exception {
String ticket = messages.sendAsync(T, "long task", null, "term_a");
awaitWaiting();
injector.onStatus(T, AgentStatus.IDLE); // deliver
injector.onStatus(T, AgentStatus.WORKING); // worker works
assertTrue(rendezvous.resolve(T, "secret async result"), "a reply resolves the async send");
MessageService.TaskView owner = driveAsyncTicketToDone(ticket, "term_a");
assertNotNull(owner, "the creator must still be able to read its own ticket");
assertEquals(MessageService.Phase.DONE, owner.phase());
MessageService.TaskView refused = messages.poll(ticket, "term_b");
assertNotNull(refused, "a different terminal gets a refusal, not silence");
assertNotEquals(MessageService.Phase.DONE, refused.phase(),
"a different terminal must never see the ticket as DONE");
assertNull(refused.reply(), "a refusal must never carry the reply text");
assertFalse(String.valueOf(refused).contains("secret async result"),
"the reply text must not appear anywhere in the refused view");
}
@Test
void unnamedPrimaryStillReadsAnyTicket() throws Exception {
String ticket = messages.sendAsync(T, "long task", null, "term_lead");
awaitWaiting();
injector.onStatus(T, AgentStatus.IDLE); // deliver
injector.onStatus(T, AgentStatus.WORKING); // worker works
assertTrue(rendezvous.resolve(T, "primary-visible result"), "a reply resolves the async send");
// callerTerminal == null is the unnamed primary (resolved by token or loopback trust, with
// no herdr pane) — it must read a ticket a terminal-bearing lead created.
MessageService.TaskView view = driveAsyncTicketToDone(ticket, null);
assertNotNull(view, "the unnamed primary must be able to read any ticket");
assertEquals(MessageService.Phase.DONE, view.phase());
assertEquals("primary-visible result", view.reply());
}
@Test
void creatorReadsItsOwnTicket() throws Exception {
String ticket = messages.sendAsync(T, "long task", null, "term_creator");
awaitWaiting();
injector.onStatus(T, AgentStatus.IDLE); // deliver
injector.onStatus(T, AgentStatus.WORKING); // worker works
assertTrue(rendezvous.resolve(T, "own result"), "a reply resolves the async send");
MessageService.TaskView view = driveAsyncTicketToDone(ticket, "term_creator");
assertNotNull(view, "the ticket's own creator must be able to read it");
assertEquals(MessageService.Phase.DONE, view.phase());
assertEquals("own result", view.reply());
}
@Test
void pollReportsACompletedTicket() throws Exception {
String ticket = messages.sendAsync(T, "long task");
@@ -1000,11 +896,11 @@ class MessageServiceTest {
@Test
void concurrentSendToSameSessionWhileFirstHoldsItIsBusy() throws Exception {
CompletableFuture<MessageService.Reply> first =
CompletableFuture.supplyAsync(() -> messages.send(T, "first", 5000, null));
CompletableFuture.supplyAsync(() -> messages.send(T, "first", 5000));
awaitWaiting(); // the first send now holds the session lock, blocked on its reply
// A second send to the SAME session cannot take the lock within its short window.
MessageService.Reply busy = messages.send(T, "second", 100, null);
MessageService.Reply busy = messages.send(T, "second", 100);
assertEquals(MessageService.Outcome.BUSY, busy.outcome(),
"a second send while another holds the session is busy, not a hang");
assertNull(busy.text());
@@ -1034,13 +930,13 @@ class MessageServiceTest {
PrimaryRegistry reg = new PrimaryRegistry(null);
// L accepts a delegation to W: the send wins the lock and queues delivery → L is recorded.
CompletableFuture<MessageService.Reply> first = CompletableFuture.supplyAsync(
() -> messages.send(T, "first", 5000, () -> reg.recordDelegation(T, LEAD_L), null));
() -> messages.send(T, "first", 5000, () -> reg.recordDelegation(T, LEAD_L)));
awaitWaiting();
assertEquals(LEAD_L, reg.nudgeTargetFor(T).orElseThrow(),
"an accepted send owns the delegation");
// A attempts W while L holds it → BUSY (lock never taken) → its hook never fires.
MessageService.Reply busy = messages.send(T, "second", 100, () -> reg.recordDelegation(T, LEAD_A), null);
MessageService.Reply busy = messages.send(T, "second", 100, () -> reg.recordDelegation(T, LEAD_A));
assertEquals(MessageService.Outcome.BUSY, busy.outcome());
assertEquals(LEAD_L, reg.nudgeTargetFor(T).orElseThrow(),
"a BUSY send must not steal the delegator ownership it never earned");
@@ -1061,7 +957,7 @@ class MessageServiceTest {
void anAcceptedSendAfterThePriorOwnerFinishesBecomesTheNewOwner() throws Exception {
PrimaryRegistry reg = new PrimaryRegistry(null);
CompletableFuture<MessageService.Reply> first = CompletableFuture.supplyAsync(
() -> messages.send(T, "first", 5000, () -> reg.recordDelegation(T, LEAD_L), null));
() -> messages.send(T, "first", 5000, () -> reg.recordDelegation(T, LEAD_L)));
awaitWaiting();
injector.onStatus(T, AgentStatus.IDLE);
injector.onStatus(T, AgentStatus.WORKING);
@@ -1072,7 +968,7 @@ class MessageServiceTest {
// L finished; A's later accepted send takes the delegation over.
CompletableFuture<MessageService.Reply> second = CompletableFuture.supplyAsync(
() -> messages.send(T, "second", 5000, () -> reg.recordDelegation(T, LEAD_A), null));
() -> messages.send(T, "second", 5000, () -> reg.recordDelegation(T, LEAD_A)));
awaitWaiting();
assertEquals(LEAD_A, reg.nudgeTargetFor(T).orElseThrow(),
"an accepted send after the owner finished becomes the new delegator");
@@ -1093,7 +989,7 @@ class MessageServiceTest {
void answeringAnAskDoesNotRewriteDelegatorOwnership() throws Exception {
PrimaryRegistry reg = new PrimaryRegistry(null);
CompletableFuture<MessageService.Reply> send = CompletableFuture.supplyAsync(
() -> messages.send(T, "do X", 5000, () -> reg.recordDelegation(T, LEAD_L), null));
() -> messages.send(T, "do X", 5000, () -> reg.recordDelegation(T, LEAD_L)));
awaitWaiting();
assertEquals(LEAD_L, reg.nudgeTargetFor(T).orElseThrow(), "L owns the delegation");
@@ -1108,7 +1004,7 @@ class MessageServiceTest {
// L answers the ask on the same turn; the answer path must not touch ownership.
CompletableFuture<MessageService.Reply> answer =
CompletableFuture.supplyAsync(() -> messages.answer(q.turnId(), "config.yaml", 5000, null));
CompletableFuture.supplyAsync(() -> messages.answer(q.turnId(), "config.yaml", 5000));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
awaitWaiting(); // the answering send reopened its forward waiter
assertEquals(LEAD_L, reg.nudgeTargetFor(T).orElseThrow(),
@@ -1128,7 +1024,7 @@ class MessageServiceTest {
void aThrowingAcceptedHookLeavesNoStaleWaiterOrQueuedOrphan() {
assertThrows(IllegalStateException.class,
() -> messages.send(T, "doomed", 500,
() -> { throw new IllegalStateException("ownership hook failed"); }, null),
() -> { throw new IllegalStateException("ownership hook failed"); }),
"a throwing ownership hook fails the send loudly");
assertFalse(rendezvous.isWaiting(T), "the failed send must not leave a stale rendezvous waiter");
@@ -1256,7 +1152,7 @@ class MessageServiceTest {
@Test
void abandonFailsASendThatIsStillWaitingOnAReleasedSession() throws Exception {
CompletableFuture<MessageService.Reply> send =
CompletableFuture.supplyAsync(() -> messages.send(T, "work", 30_000, null));
CompletableFuture.supplyAsync(() -> messages.send(T, "work", 30_000));
awaitWaiting();
assertTrue(messages.abandon(T, "session released"), "a live waiter is abandoned");
@@ -1276,7 +1172,7 @@ class MessageServiceTest {
@Test
void abandonDoesNotOverwriteAnAlreadyResolvedSend() throws Exception {
CompletableFuture<MessageService.Reply> send =
CompletableFuture.supplyAsync(() -> messages.send(T, "work", 30_000, null));
CompletableFuture.supplyAsync(() -> messages.send(T, "work", 30_000));
awaitWaiting();
assertTrue(rendezvous.resolve(T, "the real answer"));
@@ -1407,7 +1303,7 @@ class MessageServiceTest {
assertEquals(MessageService.Phase.ASKING, messages.poll(ticket).phase());
CompletableFuture<MessageService.Reply> answer = CompletableFuture.supplyAsync(
() -> messages.answer(asking.turnId(), "config.yaml", 5000, null));
() -> messages.answer(asking.turnId(), "config.yaml", 5000));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
awaitWaiting();
assertTrue(rendezvous.resolve(T, "done"));
@@ -1447,7 +1343,7 @@ class MessageServiceTest {
// same turnId must see it as lapsed rather than resolving a question nobody is waiting on.
assertEquals(MessageService.AskOutcome.TIMED_OUT, ask.get(5, TimeUnit.SECONDS).outcome());
assertEquals(MessageService.Outcome.STALE_TURN,
messages.answer(asking.turnId(), "config.yaml", 200, null).outcome());
messages.answer(asking.turnId(), "config.yaml", 200).outcome());
}
@Test
@@ -1484,7 +1380,7 @@ class MessageServiceTest {
// The primary answers, but its own bounded wait for the worker's resumed turn is short and
// expires before the worker (still genuinely working) gets back to it.
MessageService.Reply answerReply = messages.answer(asking.turnId(), "config.yaml", 150, null);
MessageService.Reply answerReply = messages.answer(asking.turnId(), "config.yaml", 150);
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
assertEquals(MessageService.Outcome.TIMED_OUT_WORKING, answerReply.outcome(),
"the primary's own bounded wait gives up before the worker finishes resuming");
@@ -1513,7 +1409,7 @@ class MessageServiceTest {
CompletableFuture.supplyAsync(() -> messages.ask(T, "which config?", 5000));
MessageService.TaskView asking = awaitTicketPhase(ticket, MessageService.Phase.ASKING);
MessageService.Reply answerReply = messages.answer(asking.turnId(), "config.yaml", 150, null);
MessageService.Reply answerReply = messages.answer(asking.turnId(), "config.yaml", 150);
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
assertEquals(MessageService.Outcome.TIMED_OUT_WORKING, answerReply.outcome());
@@ -1567,7 +1463,7 @@ class MessageServiceTest {
messages.setFinishAsyncTaskRaceHookForTest(() -> messages.forgetTurnForTest(turnId));
CompletableFuture<MessageService.Reply> answer =
CompletableFuture.supplyAsync(() -> messages.answer(turnId, "config.yaml", 5000, null));
CompletableFuture.supplyAsync(() -> messages.answer(turnId, "config.yaml", 5000));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
awaitWaiting(); // answer() opened its own forward waiter for the resumed worker turn
@@ -1673,7 +1569,7 @@ class MessageServiceTest {
String turnId = asking.turnId();
CompletableFuture<MessageService.Reply> answer =
CompletableFuture.supplyAsync(() -> messages.answer(turnId, "config.yaml", 5000, null));
CompletableFuture.supplyAsync(() -> messages.answer(turnId, "config.yaml", 5000));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer(),
"the worker's own ask() call must have already unblocked with the primary's answer "
+ "before we force the race below");
@@ -1720,7 +1616,7 @@ class MessageServiceTest {
MessageService.TaskView asking = awaitTicketPhase(ticket, MessageService.Phase.ASKING);
String turnId = asking.turnId();
MessageService.Reply answerReply = messages.answer(turnId, "config.yaml", 150, null);
MessageService.Reply answerReply = messages.answer(turnId, "config.yaml", 150);
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
assertEquals(MessageService.Outcome.TIMED_OUT_WORKING, answerReply.outcome(),
"the primary's own bounded wait must give up first, leaving turnId stamped with no "
@@ -1786,7 +1682,7 @@ class MessageServiceTest {
MessageService.TaskView asking = messages.poll(first);
CompletableFuture<MessageService.Reply> answer = CompletableFuture.supplyAsync(
() -> messages.answer(asking.turnId(), "config.yaml", 5000, null));
() -> messages.answer(asking.turnId(), "config.yaml", 5000));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
awaitWaiting();
assertTrue(rendezvous.resolve(T, "done"));
@@ -1820,7 +1716,7 @@ class MessageServiceTest {
// The primary answers it — answer() resumes the turn and blocks for what comes next.
CompletableFuture<MessageService.Reply> answer1 = CompletableFuture.supplyAsync(
() -> messages.answer(asking1.turnId(), "a1", 5000, null));
() -> messages.answer(asking1.turnId(), "a1", 5000));
assertEquals("a1", ask1.get(5, TimeUnit.SECONDS).answer());
// Still in the SAME resumed turn — before replying — the worker asks again.
@@ -1841,7 +1737,7 @@ class MessageServiceTest {
// The primary answers the second question; the worker finally sends its real fleet_reply.
CompletableFuture<MessageService.Reply> answer2 = CompletableFuture.supplyAsync(
() -> messages.answer(turnId2, "a2", 5000, null));
() -> messages.answer(turnId2, "a2", 5000));
assertEquals("a2", ask2.get(5, TimeUnit.SECONDS).answer());
awaitWaiting();
assertTrue(rendezvous.resolve(T, "done"));
@@ -1908,20 +1804,20 @@ class MessageServiceTest {
}
}
// --- fleet_status pendingAsk() -----------------------------------------------------------
// --- CB-582: fleet_status pendingAsk() ------------------------------------------------------
@Test
void pendingAskReturnsNullWhenNoQuestionIsOpen() throws Exception {
assertNull(messages.pendingAsk(T, null), "no async ticket at all -> no pending ask");
assertNull(messages.pendingAsk(T), "no async ticket at all -> no pending ask");
String ticket = messages.sendAsync(T, "long task");
awaitWaiting();
assertNull(messages.pendingAsk(T, null), "a plain pending delegation is not a question");
assertNull(messages.pendingAsk(T), "a plain pending delegation is not a question");
injectDelivery();
assertTrue(rendezvous.resolve(T, "done"));
awaitTicketPhase(ticket, MessageService.Phase.DONE);
assertNull(messages.pendingAsk(T, null), "a finished ticket carries no open question either");
assertNull(messages.pendingAsk(T), "a finished ticket carries no open question either");
}
@Test
@@ -1934,151 +1830,21 @@ class MessageServiceTest {
CompletableFuture.supplyAsync(() -> messages.ask(T, "which config file?", 5000));
MessageService.TaskView asking = awaitTicketPhase(ticket, MessageService.Phase.ASKING);
MessageService.PendingAsk pending = messages.pendingAsk(T, null);
MessageService.PendingAsk pending = messages.pendingAsk(T);
assertNotNull(pending, "fleet_status should see the open question");
assertEquals(ticket, pending.ticket());
assertEquals("which config file?", pending.question());
assertEquals(asking.turnId(), pending.turnId());
CompletableFuture<MessageService.Reply> answer = CompletableFuture.supplyAsync(
() -> messages.answer(asking.turnId(), "config.yaml", 5000, null));
() -> messages.answer(asking.turnId(), "config.yaml", 5000));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
assertNull(messages.pendingAsk(T, null), "an answered question is no longer pending");
assertNull(messages.pendingAsk(T), "an answered question is no longer pending");
awaitWaiting();
assertTrue(rendezvous.resolve(T, "done"));
assertEquals(MessageService.Outcome.REPLIED, answer.get(5, TimeUnit.SECONDS).outcome());
}
/**
* A caller's own terminal must match the terminal that created the delegation to see its
* pending question; a different terminal-bearing caller sees nothing, and a caller with no
* terminal at all (the unnamed primary) always sees it.
*/
@Test
void pendingAskGatesTheQuestionByTheDelegationsCreatorTerminal() throws Exception {
String ticket = messages.sendAsync(T, "task that asks", null, "term_creator");
awaitWaiting();
injectDelivery();
CompletableFuture<MessageService.AskResult> ask =
CompletableFuture.supplyAsync(() -> messages.ask(T, "which config file?", 5000));
MessageService.TaskView asking = awaitTicketPhase(ticket, MessageService.Phase.ASKING);
assertNull(messages.pendingAsk(T, "term_other"),
"a caller whose terminal did not create the delegation must not see the question");
MessageService.PendingAsk own = messages.pendingAsk(T, "term_creator");
assertNotNull(own, "the creating caller must see its own open question");
assertEquals("which config file?", own.question());
MessageService.PendingAsk unnamed = messages.pendingAsk(T, null);
assertNotNull(unnamed, "a caller with no terminal (the unnamed primary) must always see the question");
assertEquals("which config file?", unnamed.question());
CompletableFuture<MessageService.Reply> answer = CompletableFuture.supplyAsync(
() -> messages.answer(asking.turnId(), "config.yaml", 5000, "term_creator"));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
awaitWaiting();
assertTrue(rendezvous.resolve(T, "done"));
assertEquals(MessageService.Outcome.REPLIED, answer.get(5, TimeUnit.SECONDS).outcome());
}
/**
* A task created with no recorded creator terminal (a short {@code sendAsync} overload) must
* not hand its open question to any caller that does have a terminal — only a caller with no
* terminal at all may still see it.
*/
@Test
void pendingAskDeniesATerminalBearingCallerWhenTheTaskRecordsNoCreator() throws Exception {
String ticket = messages.sendAsync(T, "task that asks"); // no creatorTerminal recorded
awaitWaiting();
injectDelivery();
CompletableFuture<MessageService.AskResult> ask =
CompletableFuture.supplyAsync(() -> messages.ask(T, "which config file?", 5000));
MessageService.TaskView asking = awaitTicketPhase(ticket, MessageService.Phase.ASKING);
assertNull(messages.pendingAsk(T, "term_someone"),
"a terminal-bearing caller must not see a question whose task records no creator");
assertNotNull(messages.pendingAsk(T, null),
"the unnamed primary must still see it even with no recorded creator");
CompletableFuture<MessageService.Reply> answer = CompletableFuture.supplyAsync(
() -> messages.answer(asking.turnId(), "config.yaml", 5000, null));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
awaitWaiting();
assertTrue(rendezvous.resolve(T, "done"));
assertEquals(MessageService.Outcome.REPLIED, answer.get(5, TimeUnit.SECONDS).outcome());
}
// --- fleetd #715: answer() is gated on the caller that owns the turn -----------------------
/**
* A blocking {@code fleet_send} from one caller opens the turn; a different caller's answer is
* refused with no side effect on the rendezvous or the worker's blocked {@code fleet_ask} —
* only the real owner can answer it.
*/
@Test
void aDifferentCallersAnswerIsRefusedForABlockingSendDelegationButTheRealOwnerSucceeds() throws Exception {
CompletableFuture<MessageService.Reply> send = CompletableFuture.supplyAsync(
() -> messages.send(T, "do X", 5000, "term_owner"));
awaitWaiting();
injectDelivery();
CompletableFuture<MessageService.AskResult> ask =
CompletableFuture.supplyAsync(() -> messages.ask(T, "which config?", 5000));
MessageService.Reply q = send.get(5, TimeUnit.SECONDS);
assertEquals(MessageService.Outcome.QUESTION, q.outcome());
assertFalse(rendezvous.isWaiting(T), "the forward waiter closes once the question surfaces");
// The hijack: a different caller answers the SAME turnId.
MessageService.Reply hijacked = messages.answer(q.turnId(), "evil.yaml", 500, "term_attacker");
assertEquals(MessageService.Outcome.NOT_TURN_OWNER, hijacked.outcome(),
"a caller that did not open this turn must be refused, not served");
assertFalse(rendezvous.isWaiting(T), "a refused answer must not open a forward waiter");
assertFalse(ask.isDone(), "a refused answer must not resolve the worker's blocked fleet_ask");
assertEquals(T, rendezvous.askSession(q.turnId()), "a refused answer must leave the ask turn open");
// The control: the real owner answers the same turnId and the turn resumes normally.
CompletableFuture<MessageService.Reply> answer = CompletableFuture.supplyAsync(
() -> messages.answer(q.turnId(), "config.yaml", 5000, "term_owner"));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
awaitWaiting();
assertTrue(rendezvous.resolve(T, "done"));
assertEquals(MessageService.Outcome.REPLIED, answer.get(5, TimeUnit.SECONDS).outcome());
}
/**
* Same hijack and control as the blocking case, but the delegation is opened through the
* fire-and-poll path ({@code sendAsync}) — the owner comes from the ticket's recorded creator
* terminal, not a caller argument threaded through a live blocking call.
*/
@Test
void aDifferentCallersAnswerIsRefusedForAnAsyncSendDelegationButTheRealOwnerSucceeds() throws Exception {
String ticket = messages.sendAsync(T, "task that asks", null, "term_owner");
awaitWaiting();
injectDelivery();
CompletableFuture<MessageService.AskResult> ask =
CompletableFuture.supplyAsync(() -> messages.ask(T, "which config file?", 5000));
MessageService.TaskView asking = awaitTicketPhase(ticket, MessageService.Phase.ASKING);
MessageService.Reply hijacked = messages.answer(asking.turnId(), "evil.yaml", 500, "term_attacker");
assertEquals(MessageService.Outcome.NOT_TURN_OWNER, hijacked.outcome(),
"a caller that did not create this delegation must be refused, not served");
assertFalse(ask.isDone(), "a refused answer must not resolve the worker's blocked fleet_ask");
assertEquals(MessageService.Phase.ASKING, messages.poll(ticket).phase(),
"a refused answer must not advance the async ticket's phase");
CompletableFuture<MessageService.Reply> answer = CompletableFuture.supplyAsync(
() -> messages.answer(asking.turnId(), "config.yaml", 5000, "term_owner"));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
awaitWaiting();
assertTrue(rendezvous.resolve(T, "done"));
assertEquals(MessageService.Outcome.REPLIED, answer.get(5, TimeUnit.SECONDS).outcome());
assertEquals("done", awaitTicketPhase(ticket, MessageService.Phase.DONE).reply());
}
// --- CB-588: async ticket terminal nudges ---------------------------------------------------
//
// MessageService.reply's rendezvous fast path is exactly what an async ticket always takes
@@ -2335,7 +2101,7 @@ class MessageServiceTest {
// Clean up the still-open ask so the background thread does not linger past the test.
CompletableFuture<MessageService.Reply> answer = CompletableFuture.supplyAsync(
() -> wiring.service().answer(asking.turnId(), "config.yaml", 5000, null));
() -> wiring.service().answer(asking.turnId(), "config.yaml", 5000));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
awaitWaiting();
assertTrue(rendezvous.resolve(T, "done"));
@@ -2361,7 +2127,7 @@ class MessageServiceTest {
.filter(c -> c.method().equals("agent.prompt")).count();
CompletableFuture<MessageService.Reply> answer = CompletableFuture.supplyAsync(
() -> wiring.service().answer(asking.turnId(), "config.yaml", 5000, null));
() -> wiring.service().answer(asking.turnId(), "config.yaml", 5000));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
awaitWaiting();
java.util.concurrent.CountDownLatch terminalReached = new java.util.concurrent.CountDownLatch(1);
@@ -2754,7 +2520,7 @@ class MessageServiceTest {
void hasQueuedDeliveryIsTrueAfterAnUndeliveredSendTimesOut() {
// Nothing ever delivers the message (never goes IDLE/BLOCKED), so the send times out with
// TIMED_OUT_QUEUED — same setup as sendTimesOutBeforeDeliveryIsQueuedNotWorking above.
MessageService.Reply r = messages.send(T, "never delivered", 50, null);
MessageService.Reply r = messages.send(T, "never delivered", 50);
assertEquals(MessageService.Outcome.TIMED_OUT_QUEUED, r.outcome());
assertTrue(messages.hasQueuedDelivery(T),
@@ -2766,7 +2532,7 @@ class MessageServiceTest {
@Test
void hasQueuedDeliveryClearsOnceTheTargetsNextDeliveryIsAccepted() throws Exception {
assertEquals(MessageService.Outcome.TIMED_OUT_QUEUED, messages.send(T, "first", 50, null).outcome());
assertEquals(MessageService.Outcome.TIMED_OUT_QUEUED, messages.send(T, "first", 50).outcome());
assertTrue(messages.hasQueuedDelivery(T));
// A fresh send accepts delivery (opens its own waiter) — the stale queued fact is cleared.
@@ -2783,7 +2549,7 @@ class MessageServiceTest {
@Test
void hasQueuedDeliveryClearsOnAbandon() {
assertEquals(MessageService.Outcome.TIMED_OUT_QUEUED, messages.send(T, "first", 50, null).outcome());
assertEquals(MessageService.Outcome.TIMED_OUT_QUEUED, messages.send(T, "first", 50).outcome());
assertTrue(messages.hasQueuedDelivery(T));
messages.abandon(T, "session released");
@@ -1,187 +0,0 @@
package dev.ltms.fleet.msg;
import org.junit.jupiter.api.Test;
import java.io.IOException;
import java.nio.file.Files;
import java.nio.file.Path;
import java.util.ArrayList;
import java.util.List;
import java.util.stream.Stream;
import static org.junit.jupiter.api.Assertions.assertTrue;
/**
* Pins that no file under {@code src/main/java} calls the fail-open
* {@link Rendezvous#open(String)} overload. That overload opens a forward waiter with no
* recorded {@link Rendezvous.Owner}, so the turn it opens can never be answered by anyone,
* not even the unnamed primary. Every production caller must go through
* {@link Rendezvous#open(String, Rendezvous.Owner)} and record an explicit owner.
*
* <p>This reads each file's own source text rather than reflecting on compiled bytecode, because
* the risk is a future one-word edit at a call site, not a missing overload.
*
* <p>The scan below finds a violation by its receiver, {@code rendezvous.open(}, classified by
* argument count. The one test method here points that exact scanner at a file known to hold
* many real one-argument calls before it ever looks at production, so a scanner that stops
* matching fails loudly on the known-positive case instead of leaving a clean production result
* looking like evidence it never produced.
*/
class RendezvousOpenUsageTest {
private static final Path PRODUCTION_SOURCE = Path.of("src/main/java");
private static final Path KNOWN_TEST_CALLER =
Path.of("src/test/java/dev/ltms/fleet/inject/CompletionResolverTest.java");
/**
* A pattern that cannot find a known one-argument {@code rendezvous.open(} call would also
* find none in production -- not because production is clean, but because the pattern does
* not match the text it is supposed to catch. That failure mode is exactly what let a
* {@code \b}-based regex read "no callers" under {@code git grep -E} when 81 real ones
* existed: {@code git grep} does not treat {@code \b} as a word boundary, so the pattern
* silently matched nothing anywhere, clean code and real calls alike. This test runs the
* known-positive check first, with the same scanning method the production check then
* depends on, so that mistake fails loudly here instead of reading as a clean result.
*/
@Test
void noProductionFileCallsTheSingleArgumentOpenOverload() throws IOException {
List<String> knownCalls = new ArrayList<>();
int knownFilesScanned = scanForOneArgOpenCalls(KNOWN_TEST_CALLER, knownCalls);
// CONTROL: the scan actually walked files -- a wrong root would otherwise report "found
// nothing" having looked at nothing.
assertTrue(knownFilesScanned > 0, "control failed: the scan under " + KNOWN_TEST_CALLER
+ " visited zero .java files -- the path is wrong, so neither result below proves "
+ "anything");
// CONTROL: the scanner actually finds real one-argument rendezvous.open( calls when
// pointed at a file known to hold many. If this is not satisfied, the matching logic
// itself is broken, and the production result below is the scanner failing silently,
// not production code actually being clean.
assertTrue(knownCalls.size() >= 60, "control failed: the scanner found only "
+ knownCalls.size() + " one-argument rendezvous.open( call(s) in " + KNOWN_TEST_CALLER
+ ", which is known to hold many -- the matching logic itself is broken: " + knownCalls);
List<String> violations = new ArrayList<>();
int filesScanned = scanForOneArgOpenCalls(PRODUCTION_SOURCE, violations);
// CONTROL: the scan actually walked files -- a wrong root would otherwise report "no
// violations found" having looked at nothing.
assertTrue(filesScanned > 0, "control failed: the scan under " + PRODUCTION_SOURCE
+ " visited zero .java files -- the path is wrong, so the absence of violations "
+ "below proves nothing");
assertTrue(violations.isEmpty(), "found a call to the fail-open Rendezvous.open(String) "
+ "overload, which opens a forward waiter with no recorded owner -- record an "
+ "explicit Rendezvous.Owner through open(String, Owner) instead: " + violations);
}
private static int scanForOneArgOpenCalls(Path root, List<String> sites) throws IOException {
List<Path> files = javaFiles(root);
for (Path file : files) {
scanFileForOpenCalls(file, sites);
}
return files.size();
}
private static List<Path> javaFiles(Path root) throws IOException {
try (Stream<Path> paths = Files.walk(root)) {
return paths.filter(p -> p.toString().endsWith(".java")).toList();
}
}
private static void scanFileForOpenCalls(Path file, List<String> sites) throws IOException {
String source = Files.readString(file);
String needle = "rendezvous.open(";
int from = 0;
int idx;
while ((idx = source.indexOf(needle, from)) >= 0) {
int argsStart = idx + needle.length();
String args = extractBalancedArgs(source, argsStart, file, idx);
int closeParenIndex = argsStart + args.length();
from = closeParenIndex + 1;
if (args.isBlank()) {
continue; // Rendezvous has no zero-argument open() -- a javadoc "rendezvous.open()" mention, not a call
}
if (topLevelCommaCount(args) == 0) {
sites.add(file + ":" + lineOf(source, idx) + " -- rendezvous.open(" + args.trim() + ")");
}
}
}
/**
* The text between {@code rendezvous.open(} and its matching close paren: balanced over
* nested calls, and never split by a paren or comma sitting inside a string or char literal.
*/
private static String extractBalancedArgs(String source, int start, Path file, int callIndex) {
int depth = 1;
boolean inString = false;
boolean inChar = false;
int i = start;
while (i < source.length()) {
char c = source.charAt(i);
if (inString) {
if (c == '\\') { i += 2; continue; }
if (c == '"') inString = false;
} else if (inChar) {
if (c == '\\') { i += 2; continue; }
if (c == '\'') inChar = false;
} else if (c == '"') {
inString = true;
} else if (c == '\'') {
inChar = true;
} else if (c == '(') {
depth++;
} else if (c == ')') {
depth--;
if (depth == 0) return source.substring(start, i);
}
i++;
}
throw new IllegalStateException(
"unbalanced parentheses scanning " + file + ":" + lineOf(source, callIndex));
}
/**
* Commas at paren/bracket/brace depth zero, skipping string and char literals -- the argument
* separators a human reader would see, not every comma character in the text.
*/
private static int topLevelCommaCount(String args) {
int depth = 0;
int commas = 0;
boolean inString = false;
boolean inChar = false;
int i = 0;
while (i < args.length()) {
char c = args.charAt(i);
if (inString) {
if (c == '\\') { i += 2; continue; }
if (c == '"') inString = false;
} else if (inChar) {
if (c == '\\') { i += 2; continue; }
if (c == '\'') inChar = false;
} else if (c == '"') {
inString = true;
} else if (c == '\'') {
inChar = true;
} else if (c == '(' || c == '[' || c == '{') {
depth++;
} else if (c == ')' || c == ']' || c == '}') {
depth--;
} else if (c == ',' && depth == 0) {
commas++;
}
i++;
}
return commas;
}
private static int lineOf(String source, int index) {
int line = 1;
for (int i = 0; i < index; i++) {
if (source.charAt(i) == '\n') line++;
}
return line;
}
}
@@ -132,127 +132,4 @@ class RendezvousTest {
assertNull(rendezvous.askSession(t.turnId()), "a closed ask is forgotten");
assertFalse(rendezvous.answerAsk(t.turnId(), "late"), "a closed ask can no longer be answered");
}
// ── fleetd #715: turn ownership ────────────────────────────────────────────────────────────
@Test
void openWithNoOwnerRecordsNoOwnerAndOpenWithAnOwnerRecordsIt() {
rendezvous.open(W);
assertNull(rendezvous.ownerOf(W), "the no-owner overload records no owner at all");
rendezvous.close(W, rendezvous.currentWaiter(W));
rendezvous.open(W, Rendezvous.Owner.of("term_lead"));
assertEquals(Rendezvous.Owner.of("term_lead"), rendezvous.ownerOf(W));
}
@Test
void ownerOfIsNullWhenNoWaiterIsOpen() {
assertNull(rendezvous.ownerOf(W), "no waiter open means no owner to report");
}
@Test
void openAskStampsTheForwardWaitersOwnerOntoTheFreshTurnOnly() {
rendezvous.open(W, Rendezvous.Owner.of("term_lead"));
Rendezvous.AskTicket fresh = rendezvous.openAsk(W);
assertTrue(fresh.fresh());
assertEquals(Rendezvous.Owner.of("term_lead"), rendezvous.askOwner(fresh.turnId()),
"a freshly-opened ask copies the forward waiter's current owner");
Rendezvous.AskTicket coalesced = rendezvous.openAsk(W);
assertFalse(coalesced.fresh());
assertEquals(fresh.turnId(), coalesced.turnId());
assertEquals(Rendezvous.Owner.of("term_lead"), rendezvous.askOwner(coalesced.turnId()),
"a coalesced duplicate ask rides the fresh owner's turn, unchanged");
}
@Test
void aSecondAskAfterTheFirstClosesStampsWhateverOwnerIsOpenAtThatLaterMoment() {
rendezvous.open(W, Rendezvous.Owner.of("term_lead"));
Rendezvous.AskTicket first = rendezvous.openAsk(W);
rendezvous.closeAsk(first.turnId());
// The forward waiter is reopened under a different owner before the second ask — mirrors
// answer() reopening with the owner it already checked, which can differ turn to turn.
rendezvous.close(W, rendezvous.currentWaiter(W));
rendezvous.open(W, Rendezvous.Owner.of("term_other"));
Rendezvous.AskTicket second = rendezvous.openAsk(W);
assertTrue(second.fresh());
assertEquals(Rendezvous.Owner.of("term_other"), rendezvous.askOwner(second.turnId()),
"a freshly-opened ask always copies whatever owner is open right now, not a stale one");
}
@Test
void askOwnerIsNullForAnUnknownOrLapsedTurn() {
assertNull(rendezvous.askOwner("no-such#1"));
Rendezvous.AskTicket t = rendezvous.openAsk(W);
rendezvous.closeAsk(t.turnId());
assertNull(rendezvous.askOwner(t.turnId()), "a closed ask no longer reports an owner");
}
// ── fleetd #729: per-boot nonce guards turnId against cross-instance reuse ────────────────
@Test
void twoInstancesMintDisjointTurnIds() {
Rendezvous other = new Rendezvous();
Rendezvous.AskTicket fromThis = rendezvous.openAsk(W);
Rendezvous.AskTicket fromOther = other.openAsk(W);
assertNotEquals(fromThis.turnId(), fromOther.turnId(),
"each instance mints its own id space, so even a first ask from each must differ");
}
@Test
void foreignInstanceTurnIdDoesNotResolve() {
Rendezvous other = new Rendezvous();
// `other` must reach the same sequence number as `rendezvous` (two asks each, the first
// closed so the second mints fresh), or this test passes against an empty map instead of
// against a colliding id.
Rendezvous.AskTicket firstFromThis = rendezvous.openAsk(W);
rendezvous.closeAsk(firstFromThis.turnId());
Rendezvous.AskTicket secondFromThis = rendezvous.openAsk(W);
Rendezvous.AskTicket firstFromOther = other.openAsk(W);
other.closeAsk(firstFromOther.turnId());
other.openAsk(W);
// control: the id resolves in the instance that minted it, so a false below cannot be
// explained by broken plumbing — only by the turnId being foreign to `other`.
assertTrue(rendezvous.answerAsk(secondFromThis.turnId(), "answer from this instance"),
"the minting instance must still resolve its own turnId");
assertFalse(other.answerAsk(secondFromThis.turnId(), "answer from other instance"),
"a turnId minted by a different instance must not resolve here");
}
@Test
void openAskStillCoalescesDuplicatesAndStillMintsDistinctIdsPerAsk() {
Rendezvous.AskTicket t1 = rendezvous.openAsk(W);
Rendezvous.AskTicket t2 = rendezvous.openAsk(W);
assertEquals(t1.turnId(), t2.turnId(),
"a second openAsk while one is open still coalesces onto the same turn");
assertFalse(t2.fresh(), "the coalesced ask is still reported as not fresh");
rendezvous.closeAsk(t1.turnId());
Rendezvous.AskTicket t3 = rendezvous.openAsk(W);
assertNotEquals(t1.turnId(), t3.turnId(), "two asks from the same session still get different turnIds");
}
@Test
void ownerPermitsIsFailClosedOnARecordAndThreeStatesAreDistinct() {
assertFalse(Rendezvous.Owner.permits(null, null),
"no owner on record refuses even a caller with no terminal");
assertFalse(Rendezvous.Owner.permits(null, "term_a"),
"no owner on record refuses a terminal-bearing caller too");
assertTrue(Rendezvous.Owner.permits(Rendezvous.Owner.UNNAMED_PRIMARY, null),
"the unnamed primary owner matches a caller with no terminal");
assertFalse(Rendezvous.Owner.permits(Rendezvous.Owner.UNNAMED_PRIMARY, "term_a"),
"the unnamed primary owner does not match a terminal-bearing caller");
assertTrue(Rendezvous.Owner.permits(Rendezvous.Owner.of("term_a"), "term_a"),
"a named owner matches the same terminal");
assertFalse(Rendezvous.Owner.permits(Rendezvous.Owner.of("term_a"), "term_b"),
"a named owner refuses a different terminal");
assertFalse(Rendezvous.Owner.permits(Rendezvous.Owner.of("term_a"), null),
"a named owner refuses the unnamed primary");
}
}
@@ -2,13 +2,11 @@ package dev.ltms.fleet.rest;
import ch.qos.logback.classic.Level;
import ch.qos.logback.classic.spi.ILoggingEvent;
import com.fasterxml.jackson.databind.JsonNode;
import com.fasterxml.jackson.databind.ObjectMapper;
import dev.ltms.fleet.auth.CallerResolver;
import dev.ltms.fleet.auth.Authz;
import dev.ltms.fleet.auth.MemberRegistry;
import dev.ltms.fleet.auth.Principal;
import dev.ltms.fleet.auth.Role;
import dev.ltms.fleet.config.FleetConfig;
import dev.ltms.fleet.guard.SubscriptionGuard;
import dev.ltms.fleet.herdr.AgentControl;
@@ -21,7 +19,6 @@ import dev.ltms.fleet.metrics.FleetMetrics;
import dev.ltms.fleet.metrics.Metrics;
import dev.ltms.fleet.msg.MessageService;
import dev.ltms.fleet.msg.Rendezvous;
import dev.ltms.fleet.peer.MemberRole;
import dev.ltms.fleet.session.FakeWorktrees;
import dev.ltms.fleet.session.SessionManager;
import dev.ltms.fleet.member.ClaudeCodeLauncher;
@@ -42,8 +39,6 @@ import java.util.List;
import java.util.Locale;
import java.util.Map;
import java.util.Set;
import java.util.concurrent.CompletableFuture;
import java.util.concurrent.TimeUnit;
import java.util.function.Predicate;
import java.util.regex.Matcher;
import java.util.regex.Pattern;
@@ -144,411 +139,6 @@ class FleetAppAuthTest {
assertThrows(IllegalArgumentException.class, () -> FleetApp.routeAction("GET /healthz"));
}
// --- GET /tasks/{ticket} must not be the no-check overload ----------------------------------
/**
* {@code GET /tasks/{ticket}} must resolve its caller the same way {@code allow(...)} does
* and thread that terminal into {@link MessageService#poll(String, String)}, not the
* no-check overload that ignores who is asking.
*/
@Test
void theTaskStatusRouteActuallyThreadsTheCallersTerminalIntoPoll() throws Exception {
String source = Files.readString(REST_SOURCE);
int start = source.indexOf("private void taskStatus(Context ctx) {");
assertTrue(start >= 0, "could not find taskStatus in " + REST_SOURCE
+ " -- the scrape has stopped matching, fix the anchor before trusting this test");
int end = source.indexOf("private static void herdrError(Context ctx, HerdrException e) {", start);
assertTrue(end > start, "could not find the method declared after taskStatus to bound the scrape");
String handlerBlock = source.substring(start, end);
// CONTROL: the block we scraped really does call messages.poll(...) -- if this fails, the
// anchors above moved and the assertions below would otherwise pass on nothing.
assertTrue(handlerBlock.contains("messages.poll("),
"control failed: the scraped taskStatus block contains no messages.poll( call at all "
+ "-- the anchors have drifted, this test is not testing what it claims to");
assertTrue(handlerBlock.contains("caller.terminal()"),
"the taskStatus route must thread the resolved caller's terminal into messages.poll(...), "
+ "not the no-check overload -- block: " + handlerBlock);
assertTrue(handlerBlock.contains("ctx.attribute(CALLER)"),
"the taskStatus route must resolve its caller the same way allow(...) does, not via a "
+ "second, separate resolution path -- block: " + handlerBlock);
}
/**
* {@code GET /sessions/{id}/status} must resolve its caller the same way {@code allow(...)}
* does and thread that terminal into {@link MessageService#pendingAsk(String, String)}, not
* the no-check overload that ignores who is asking.
*/
@Test
void theSessionStatusRouteActuallyThreadsTheCallersTerminalIntoPendingAsk() throws Exception {
String source = Files.readString(REST_SOURCE);
int start = source.indexOf("private void sessionStatus(Context ctx) {");
assertTrue(start >= 0, "could not find sessionStatus in " + REST_SOURCE
+ " -- the scrape has stopped matching, fix the anchor before trusting this test");
int end = source.indexOf("private void taskStatus(Context ctx) {", start);
assertTrue(end > start, "could not find the method declared after sessionStatus to bound the scrape");
String handlerBlock = source.substring(start, end);
// CONTROL: the block we scraped really does call messages.pendingAsk(...) -- if this fails,
// the anchors above moved and the assertions below would otherwise pass on nothing.
assertTrue(handlerBlock.contains("messages.pendingAsk("),
"control failed: the scraped sessionStatus block contains no messages.pendingAsk( "
+ "call at all -- the anchors have drifted, this test is not testing what it claims to");
assertTrue(handlerBlock.contains("caller.terminal()"),
"the sessionStatus route must thread the resolved caller's terminal into "
+ "messages.pendingAsk(...), not the no-check overload -- block: " + handlerBlock);
assertTrue(handlerBlock.contains("ctx.attribute(CALLER)"),
"the sessionStatus route must resolve its caller the same way allow(...) does, not via "
+ "a second, separate resolution path -- block: " + handlerBlock);
}
/**
* {@code GET /tasks/{ticket}} refuses a worker whose terminal did not create the ticket,
* while the creating worker and the unnamed primary both still read it. The ticket is minted
* directly on the shared {@link MessageService}, the same way {@code MessageServiceTest}
* drives {@link MessageService#poll(String, String)}, so this exercises only the REST poll
* route's own handling of the ownership already recorded on the ticket.
*/
@Test
void restPollRefusesADifferentWorkerButAllowsTheCreatorAndTheUnnamedPrimary() throws Exception {
FakeHerdr herdr = new FakeHerdr();
AgentControl agents = new AgentControl(herdr);
Injector injector = new Injector(agents);
MessageService messages = new MessageService(agents, injector, new Rendezvous());
Javalin creatorApp = startOnSharedService(messages, herdr, FakeHerdr.WORKER_PID); // -> term_a
Javalin otherWorkerApp = startOnSharedService(messages, herdr, 9001L); // -> term_shell
Javalin primaryApp = startOnSharedService(messages, herdr, 999_999L); // no pane -> primary
try {
String ticket = messages.sendAsync("term_a", "long task", null, "term_a");
HttpResponse<String> refused = send(otherWorkerApp.port(), "GET", "/tasks/" + ticket, null, null);
assertEquals(200, refused.statusCode());
assertTrue(refused.body().contains("forbidden"),
"a different worker's terminal must be refused, not shown the ticket: " + refused.body());
assertFalse(refused.body().contains("\"reply\""),
"a refusal must never carry reply text: " + refused.body());
HttpResponse<String> own = send(creatorApp.port(), "GET", "/tasks/" + ticket, null, null);
assertEquals(200, own.statusCode());
assertFalse(own.body().contains("forbidden"),
"the creating worker must read its own ticket: " + own.body());
HttpResponse<String> primary = send(primaryApp.port(), "GET", "/tasks/" + ticket, null, null);
assertEquals(200, primary.statusCode());
assertFalse(primary.body().contains("forbidden"),
"the unnamed primary must read any ticket: " + primary.body());
} finally {
creatorApp.stop();
otherWorkerApp.stop();
primaryApp.stop();
}
}
/**
* {@code POST /sessions/{id}/message} with {@code wait:false} must record the creating
* caller's own terminal on the ticket it returns, so that caller can still poll its own
* ticket over REST, while a different terminal is refused.
*/
@Test
void restSendAsyncRecordsTheCreatingCallersTerminalSoItCanStillPollItsOwnTicket() throws Exception {
FakeHerdr herdr = new FakeHerdr();
AgentControl agents = new AgentControl(herdr);
Injector injector = new Injector(agents);
MessageService messages = new MessageService(agents, injector, new Rendezvous());
Javalin leadApp = startOnSharedService(messages, herdr, FakeHerdr.WORKER_PID, Map.of("term_a", "lead-x"));
Javalin otherWorkerApp = startOnSharedService(messages, herdr, 9001L); // -> term_shell
try {
ObjectMapper mapper = new ObjectMapper();
HttpResponse<String> created = send(leadApp.port(), "POST", "/sessions/term_a/message",
"{\"content\":\"long task\",\"wait\":false}", null);
assertEquals(202, created.statusCode(), created.body());
String ticket = mapper.readTree(created.body()).path("ticket").asText(null);
assertNotNull(ticket, "the accepted response carried no ticket: " + created.body());
HttpResponse<String> own = send(leadApp.port(), "GET", "/tasks/" + ticket, null, null);
assertEquals(200, own.statusCode());
assertFalse(own.body().contains("forbidden"),
"the session that created the ticket over REST must be able to poll it: " + own.body());
HttpResponse<String> refused = send(otherWorkerApp.port(), "GET", "/tasks/" + ticket, null, null);
assertEquals(200, refused.statusCode());
assertTrue(refused.body().contains("forbidden: this ticket was created by a different session"),
"a different terminal must still be refused with the ownership detail, not some "
+ "other rejection: " + refused.body());
} finally {
leadApp.stop();
otherWorkerApp.stop();
}
}
/**
* {@code GET /sessions/{id}/status} shows a worker's pending {@code fleet_ask} question, its
* {@code turnId} and its ticket only to the caller whose terminal created that delegation, or
* to a caller with no terminal at all (the unnamed primary) — a different terminal-bearing
* caller still sees the base status line, but none of the pending-ask fields.
*/
@Test
void restStatusGatesThePendingAskFieldsByTheDelegationsCreatorTerminal() throws Exception {
FakeHerdr herdr = new FakeHerdr();
AgentControl agents = new AgentControl(herdr);
Injector injector = new Injector(agents);
Rendezvous rendezvous = new Rendezvous();
MessageService messages = new MessageService(agents, injector, rendezvous);
Javalin creatorApp = startOnSharedService(messages, herdr, FakeHerdr.WORKER_PID); // -> term_a
Javalin otherWorkerApp = startOnSharedService(messages, herdr, 9001L); // -> term_shell
Javalin primaryApp = startOnSharedService(messages, herdr, 999_999L); // no pane -> primary
try {
ObjectMapper mapper = new ObjectMapper();
String ticket = messages.sendAsync("term_target", "task that asks", null, "term_a");
long deadline = System.currentTimeMillis() + 3000;
while (!rendezvous.isWaiting("term_target") && System.currentTimeMillis() < deadline) {
Thread.sleep(5);
}
assertTrue(rendezvous.isWaiting("term_target"), "sendAsync should have opened its rendezvous waiter");
CompletableFuture<MessageService.AskResult> ask = CompletableFuture.supplyAsync(
() -> messages.ask("term_target", "which config file?", 5000));
MessageService.TaskView asking;
deadline = System.currentTimeMillis() + 3000;
do {
asking = messages.poll(ticket, null);
Thread.sleep(5);
} while (asking.phase() != MessageService.Phase.ASKING && System.currentTimeMillis() < deadline);
assertEquals(MessageService.Phase.ASKING, asking.phase());
String turnId = asking.turnId();
JsonNode other = mapper.readTree(
send(otherWorkerApp.port(), "GET", "/sessions/term_target/status", null, null).body());
assertEquals("idle", other.get("status").asText(), "the base status must still be shown");
assertFalse(other.has("question"), "a non-creating caller must not see the question: " + other);
assertFalse(other.has("turnId"), "a non-creating caller must not see the turnId: " + other);
assertFalse(other.has("ticket"), "a non-creating caller must not see the ticket: " + other);
JsonNode own = mapper.readTree(
send(creatorApp.port(), "GET", "/sessions/term_target/status", null, null).body());
assertEquals("which config file?", own.get("question").asText(), "the creator must see the question");
assertEquals(turnId, own.get("turnId").asText(), "the creator must see the turnId");
assertEquals(ticket, own.get("ticket").asText(), "the creator must see the ticket");
JsonNode primary = mapper.readTree(
send(primaryApp.port(), "GET", "/sessions/term_target/status", null, null).body());
assertEquals("which config file?", primary.get("question").asText(),
"a caller with no terminal (the unnamed primary) must see the question");
// Clean up the still-open ask so the background thread does not linger past the test.
CompletableFuture<MessageService.Reply> answer = CompletableFuture.supplyAsync(
() -> messages.answer(turnId, "config.yaml", 5000, "term_a"));
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
deadline = System.currentTimeMillis() + 3000;
while (!rendezvous.isWaiting("term_target") && System.currentTimeMillis() < deadline) {
Thread.sleep(5);
}
assertTrue(rendezvous.resolve("term_target", "done"));
answer.get(5, TimeUnit.SECONDS);
} finally {
creatorApp.stop();
otherWorkerApp.stop();
primaryApp.stop();
}
}
/**
* {@code POST /sessions/{id}/message} with a {@code turnId} refuses a caller whose terminal
* did not open the turn, even when that caller otherwise holds ANSWER rights, and leaves the
* turn open for the real owner to resolve. Covers the REST adapter's ANSWER gate for an
* async-send ({@code wait:false}) delegation.
*/
@Test
void restAnswerIsRefusedForADifferentCallerOnAnAsyncSendDelegationButTheRealOwnerSucceeds() throws Exception {
FakeHerdr herdr = new FakeHerdr();
AgentControl agents = new AgentControl(herdr);
Injector injector = new Injector(agents);
Rendezvous rendezvous = new Rendezvous();
MessageService messages = new MessageService(agents, injector, rendezvous);
Javalin ownerApp = startOnSharedService(messages, herdr, FakeHerdr.WORKER_PID, Map.of("term_a", "lead-owner"));
Javalin attackerApp = startOnSharedService(messages, herdr, 9001L, Map.of("term_shell", "lead-attacker"));
try {
ObjectMapper mapper = new ObjectMapper();
HttpResponse<String> created = send(ownerApp.port(), "POST", "/sessions/term_target/message",
"{\"content\":\"long task\",\"wait\":false}", null);
assertEquals(202, created.statusCode(), created.body());
String ticket = mapper.readTree(created.body()).path("ticket").asText(null);
assertNotNull(ticket, "the accepted response carried no ticket: " + created.body());
long deadline = System.currentTimeMillis() + 3000;
while (!rendezvous.isWaiting("term_target") && System.currentTimeMillis() < deadline) {
Thread.sleep(5);
}
assertTrue(rendezvous.isWaiting("term_target"), "the async send should have opened its rendezvous waiter");
CompletableFuture<MessageService.AskResult> ask = CompletableFuture.supplyAsync(
() -> messages.ask("term_target", "which config file?", 5000));
MessageService.TaskView asking;
deadline = System.currentTimeMillis() + 3000;
do {
asking = messages.poll(ticket, null);
Thread.sleep(5);
} while (asking.phase() != MessageService.Phase.ASKING && System.currentTimeMillis() < deadline);
assertEquals(MessageService.Phase.ASKING, asking.phase());
String turnId = asking.turnId();
HttpResponse<String> hijacked = send(attackerApp.port(), "POST", "/sessions/term_target/message",
"{\"content\":\"hijack\",\"turnId\":\"" + turnId + "\"}", null);
assertEquals(403, hijacked.statusCode(), hijacked.body());
assertTrue(hijacked.body().contains("not_turn_owner"),
"a different caller's answer must be refused as not_turn_owner: " + hijacked.body());
assertEquals("term_target", rendezvous.askSession(turnId),
"a refused answer must leave the ask turn open");
CompletableFuture<HttpResponse<String>> owned = CompletableFuture.supplyAsync(() -> {
try {
return send(ownerApp.port(), "POST", "/sessions/term_target/message",
"{\"content\":\"config.yaml\",\"turnId\":\"" + turnId + "\"}", null);
} catch (Exception e) {
throw new RuntimeException(e);
}
});
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
deadline = System.currentTimeMillis() + 3000;
while (!rendezvous.isWaiting("term_target") && System.currentTimeMillis() < deadline) {
Thread.sleep(5);
}
assertTrue(rendezvous.resolve("term_target", "done"));
HttpResponse<String> ownedResponse = owned.get(5, TimeUnit.SECONDS);
assertEquals(200, ownedResponse.statusCode(), ownedResponse.body());
assertEquals("done", mapper.readTree(ownedResponse.body()).path("reply").asText(null),
"the real owner's answer must resolve the worker's turn");
} finally {
ownerApp.stop();
attackerApp.stop();
}
}
/**
* As above, but the delegation is a blocking send ({@code wait:true}) instead of a ticket —
* the owning caller's own HTTP request is the one that surfaces the worker's question and
* later carries the real answer. Covers the REST adapter's ANSWER gate for a blocking-send
* delegation.
*/
@Test
void restAnswerIsRefusedForADifferentCallerOnABlockingSendDelegationButTheRealOwnerSucceeds() throws Exception {
FakeHerdr herdr = new FakeHerdr();
AgentControl agents = new AgentControl(herdr);
Injector injector = new Injector(agents);
Rendezvous rendezvous = new Rendezvous();
MessageService messages = new MessageService(agents, injector, rendezvous);
Javalin ownerApp = startOnSharedService(messages, herdr, FakeHerdr.WORKER_PID, Map.of("term_a", "lead-owner"));
Javalin attackerApp = startOnSharedService(messages, herdr, 9001L, Map.of("term_shell", "lead-attacker"));
try {
ObjectMapper mapper = new ObjectMapper();
CompletableFuture<HttpResponse<String>> blocking = CompletableFuture.supplyAsync(() -> {
try {
return send(ownerApp.port(), "POST", "/sessions/term_target/message",
"{\"content\":\"long task\",\"wait\":true,\"timeoutMs\":5000}", null);
} catch (Exception e) {
throw new RuntimeException(e);
}
});
long deadline = System.currentTimeMillis() + 3000;
while (!rendezvous.isWaiting("term_target") && System.currentTimeMillis() < deadline) {
Thread.sleep(5);
}
assertTrue(rendezvous.isWaiting("term_target"), "the blocking send should have opened its rendezvous waiter");
CompletableFuture<MessageService.AskResult> ask = CompletableFuture.supplyAsync(
() -> messages.ask("term_target", "which config file?", 5000));
HttpResponse<String> questionResponse = blocking.get(5, TimeUnit.SECONDS);
assertEquals(202, questionResponse.statusCode(), questionResponse.body());
JsonNode question = mapper.readTree(questionResponse.body());
assertEquals("question", question.get("status").asText());
String turnId = question.get("turnId").asText();
HttpResponse<String> hijacked = send(attackerApp.port(), "POST", "/sessions/term_target/message",
"{\"content\":\"hijack\",\"turnId\":\"" + turnId + "\"}", null);
assertEquals(403, hijacked.statusCode(), hijacked.body());
assertTrue(hijacked.body().contains("not_turn_owner"),
"a different caller's answer must be refused as not_turn_owner: " + hijacked.body());
assertEquals("term_target", rendezvous.askSession(turnId),
"a refused answer must leave the ask turn open");
CompletableFuture<HttpResponse<String>> owned = CompletableFuture.supplyAsync(() -> {
try {
return send(ownerApp.port(), "POST", "/sessions/term_target/message",
"{\"content\":\"config.yaml\",\"turnId\":\"" + turnId + "\"}", null);
} catch (Exception e) {
throw new RuntimeException(e);
}
});
assertEquals("config.yaml", ask.get(5, TimeUnit.SECONDS).answer());
deadline = System.currentTimeMillis() + 3000;
while (!rendezvous.isWaiting("term_target") && System.currentTimeMillis() < deadline) {
Thread.sleep(5);
}
assertTrue(rendezvous.resolve("term_target", "done"));
HttpResponse<String> ownedResponse = owned.get(5, TimeUnit.SECONDS);
assertEquals(200, ownedResponse.statusCode(), ownedResponse.body());
assertEquals("done", mapper.readTree(ownedResponse.body()).path("reply").asText(null),
"the real owner's answer must resolve the worker's turn");
} finally {
ownerApp.stop();
attackerApp.stop();
}
}
/**
* As {@link #start}, but shares {@code messages} and {@code herdr} across several app
* instances bound to different pids, each returned as its own started {@link Javalin} rather
* than through the shared {@code app} field, so several differently-resolved callers can
* poll the same ticket.
*/
private Javalin startOnSharedService(MessageService messages, FakeHerdr herdr, long pid) {
return startOnSharedService(messages, herdr, pid, Map.of());
}
/**
* As {@link #startOnSharedService(MessageService, FakeHerdr, long)}, but {@code leadTerminals}
* resolves the given pid's terminal to a named lead (a caller with SEND permission) instead of
* a plain worker, for a test that needs a terminal-bearing caller able to create a ticket.
*
* <p>Every connecting pane not already claimed by {@code leadTerminals} is wired into the live
* roster as a spawned worker, so a caller's resolved role matches what its own test expects:
* a {@link Role#WORKER}, never the unconfigured-pane {@link Role#OBSERVER} floor a roster-less
* resolver would otherwise fall to.
*/
private Javalin startOnSharedService(MessageService messages, FakeHerdr herdr, long pid,
Map<String, String> leadTerminals) {
FleetConfig.Profile wcfg = new FleetConfig.Profile(
"ltms-local", "http://gx00.gw:8000", "coder", null, "FLEETD_WORKER_TOKEN", null,
"tab", "fleetd-workers", "worker: {profile} #{n}", null, null, null);
AgentControl agents = new AgentControl(herdr);
ClaudeCodeLauncher workers = new ClaudeCodeLauncher(
agents, new WorkspaceControl(herdr), new SubscriptionGuard(Set.of("gx00.gw")),
Map.of(wcfg.profile(), wcfg), wcfg.profile(),
k -> "FLEETD_WORKER_TOKEN".equals(k) ? "tok-abc" : null);
SessionManager sessions = new SessionManager(workers, new FakeWorktrees());
ConnectionIdentity identity = new ConnectionIdentity(new PaneLocator(herdr), _ -> pid);
CallerResolver callers = CallerResolver.withLeadsAndMembers(identity, false, null,
() -> leadTerminals, new MemberRegistry(null),
t -> leadTerminals.containsKey(t) ? null : MemberRole.DEV, Map::of);
Metrics appMetrics = FleetMetrics.create(sessions, new dev.ltms.fleet.msg.InMemoryReplyInbox());
return new FleetApp(herdr, workers, sessions, messages, sessions.asPresence(), null,
callers, appMetrics).build().start("127.0.0.1", 0);
}
/**
* fleetd #669 Unit A: {@code POST /sessions/{id}/message} is two call shapes behind one route,
* mirroring {@code fleet_send}'s MCP-side split into {@link Authz.Action#SEND} and {@link
@@ -1,92 +0,0 @@
package dev.ltms.fleet.session;
import dev.ltms.fleet.inject.MemberPresence;
import dev.ltms.fleet.peer.Capability;
import dev.ltms.fleet.peer.PeerHandle;
import dev.ltms.fleet.peer.PeerLauncher;
import dev.ltms.fleet.peer.SpawnRequest;
import dev.ltms.fleet.placement.PlacementDecision;
import java.util.List;
import java.util.Set;
/**
* {@link PeerLauncher} decorator that marks presence for a spawned terminal before returning its
* handle to the caller — the contact-then-register ordering fleetd #722 covers, where the
* terminal's MCP contact lands before {@link SessionManager#acquire} runs its own
* {@code registry.put}. The presence view is set after construction, via {@link #presence},
* because it is owned by the {@link SessionManager} this launcher is passed into.
*/
final class PresenceRacingLauncher implements PeerLauncher {
private final PeerLauncher delegate;
volatile MemberPresence presence;
PresenceRacingLauncher(PeerLauncher delegate) {
this.delegate = delegate;
}
@Override
public PeerHandle spawn(SpawnRequest req) {
PeerHandle handle = delegate.spawn(req);
presence.markPresent(handle.terminalId());
return handle;
}
@Override
public PeerHandle spawn(SpawnRequest req, PlacementDecision decision) {
PeerHandle handle = delegate.spawn(req, decision);
presence.markPresent(handle.terminalId());
return handle;
}
@Override
public Set<Capability> capabilities() {
return delegate.capabilities();
}
@Override
public Set<Capability> capabilitiesFor(String profileName) {
return delegate.capabilitiesFor(profileName);
}
@Override
public Set<String> profiles() {
return delegate.profiles();
}
@Override
public String defaultProfile() {
return delegate.defaultProfile();
}
@Override
public String effectiveCwd(SpawnRequest req) {
return delegate.effectiveCwd(req);
}
@Override
public List<String> parityOverlay(String profileName) {
return delegate.parityOverlay(profileName);
}
@Override
public List<?> list() {
return delegate.list();
}
@Override
public int reapOrphanWorkers() {
return delegate.reapOrphanWorkers();
}
@Override
public void stop(String id) {
delegate.stop(id);
}
@Override
public boolean clearContext(String id) {
return delegate.clearContext(id);
}
}
@@ -422,53 +422,6 @@ class SessionManagerTest {
"turn completion moves BUSY → DONE");
}
// --- fleetd #722: registration and presence must reach READY whichever lands first --------
@Test
void registerThenContactReachesReadyForPlainSpawn() {
FakeHerdr herdr = new FakeHerdr();
SessionManager sessions = sessionManager(herdr);
MemberSession session = sessions.acquire("ltms-local", null, "/caller", "term_primary");
sessions.asPresence().markPresent(session.terminalId());
assertEquals(MemberSession.State.READY, sessions.get(session.paneId()).orElseThrow().state(),
"a presence contact that arrives after registration reaches READY");
}
@Test
void contactThenRegisterStillReachesReadyForPlainSpawn() {
// The racing launcher marks presence for the spawned terminal from inside spawn() —
// before SessionManager.acquire's own registry.put runs — modeling an MCP contact that
// lands in that window.
FakeHerdr herdr = new FakeHerdr();
FleetConfig.Profile cfg = new FleetConfig.Profile(
"ltms-local", "http://gx00.gw:8000", "coder", null, "FLEETD_WORKER_TOKEN",
List.of("ccs", "ltms-local"), "tab", "fleetd-workers",
"worker: {profile} #{n}", null, null, null);
ClaudeCodeLauncher workers = new ClaudeCodeLauncher(new AgentControl(herdr), new WorkspaceControl(herdr),
new SubscriptionGuard(Set.of("gx00.gw")), Map.of(cfg.profile(), cfg), cfg.profile(), _ -> null);
PresenceRacingLauncher race = new PresenceRacingLauncher(workers);
SessionManager sessions = new SessionManager(race);
race.presence = sessions.asPresence();
MemberSession session = sessions.acquire("ltms-local", null, "/caller", "term_primary");
assertEquals(MemberSession.State.READY, sessions.get(session.paneId()).orElseThrow().state(),
"a presence contact that lands before registry.put must still reach READY");
}
@Test
void aTerminalNeverMarkedPresentStaysSpawningAfterRegistration() {
FakeHerdr herdr = new FakeHerdr();
SessionManager sessions = sessionManager(herdr);
MemberSession session = sessions.acquire("ltms-local", null, "/caller", "term_primary");
assertEquals(MemberSession.State.SPAWNING, sessions.get(session.paneId()).orElseThrow().state(),
"registration alone must not advance a terminal that was never marked present");
}
@Test
void releaseTearsDownWorkerAndRemovesFromRosterAndIsIdempotent() {
FakeHerdr herdr = new FakeHerdr();
@@ -1452,105 +1405,6 @@ class SessionManagerTest {
+ "dirty check threw");
}
// --- fleetd #736: a release must forget the member's presence entry, not just its registry
// row ---------------------------------------------------------------------------------------
@Test
void releaseByPaneIdForgetsThePresenceEntry() {
FakeHerdr herdr = new FakeHerdr();
SessionManager sessions = sessionManager(herdr);
MemberSession session = sessions.acquire("ltms-local", null, "/caller", "term_primary");
String terminal = session.terminalId();
sessions.asPresence().markPresent(terminal);
assertTrue(sessions.asPresence().isPresent(terminal), "present before the release");
sessions.release(session.paneId());
assertFalse(sessions.asPresence().isPresent(terminal),
"release must forget the terminal's presence, not just remove its registry row");
}
@Test
void reapIdleForgetsThePresenceEntryToo() {
long[] clock = {0};
FakeHerdr herdr = new FakeHerdr();
SessionManager sessions = sessionManager(herdr, () -> clock[0]);
MemberSession session = sessions.acquire("ltms-local", null, "/caller", "term_primary");
String terminal = session.terminalId();
sessions.asPresence().markPresent(terminal);
assertTrue(sessions.asPresence().isPresent(terminal), "present before the reap");
clock[0] = 11;
assertEquals(1, sessions.reapIdle(10), "READY session past TTL is reaped");
assertFalse(sessions.asPresence().isPresent(terminal),
"the idle-reap release path (releaseIfCurrent) goes through the same teardown "
+ "funnel as an explicit release, so it must forget presence too");
}
@Test
void shutdownDrainAlsoForgetsThePresenceEntry() {
FakeHerdr herdr = new FakeHerdr();
SessionManager sessions = sessionManager(herdr);
MemberSession session = sessions.acquire("ltms-local", null, "/caller", "term_primary");
String terminal = session.terminalId();
sessions.asPresence().markPresent(terminal);
assertTrue(sessions.asPresence().isPresent(terminal), "present before the drain");
sessions.drainAll(TimeUnit.MILLISECONDS.toNanos(100));
assertFalse(sessions.asPresence().isPresent(terminal),
"a shutdown drain still ends the member's process, so presence must be cleared "
+ "exactly as it is for any other release cause");
}
@Test
void releaseOfAnUnknownPaneIdDoesNotThrow() {
FakeHerdr herdr = new FakeHerdr();
SessionManager sessions = sessionManager(herdr);
assertDoesNotThrow(() -> sessions.release("no-such-pane"),
"releasing a pane id that was never registered must be a no-op, not a throw");
}
@Test
void releaseStillForgetsPresenceWhenDirtyCheckThrows() {
FakeHerdr herdr = new FakeHerdr();
RecordingWorktrees worktrees = new RecordingWorktrees();
SessionManager sessions = sessionManager(herdr, worktrees);
MemberSession s = sessions.acquire("ltms-local", null, "/caller/proj", null,
new WorktreeRequest("fleetd-736", null));
String terminal = s.terminalId();
sessions.asPresence().markPresent(terminal);
worktrees.failHasUncommittedWith(new WorktreeException("git status exited 128"));
assertDoesNotThrow(() -> sessions.release(s.paneId()),
"a throwing dirty check must not abort the release");
assertFalse(sessions.asPresence().isPresent(terminal),
"presence must be forgotten even when the dirty check throws, which pins the "
+ "forget call to the finally block that runs no matter what happened above");
}
@Test
void releaseLeavesADifferentStillLiveMembersPresenceUntouched() {
FakeHerdr herdr = new FakeHerdr();
SessionManager sessions = sessionManager(herdr);
MemberSession released = sessions.acquire("ltms-local", null, "/caller/a", "ownerA");
MemberSession stillLive = sessions.acquire("ltms-local", null, "/caller/b", "ownerB");
sessions.asPresence().markPresent(released.terminalId());
sessions.asPresence().markPresent(stillLive.terminalId());
assertTrue(sessions.asPresence().isPresent(stillLive.terminalId()),
"present before the release of the other member");
sessions.release(released.paneId());
assertFalse(sessions.asPresence().isPresent(released.terminalId()),
"the released terminal is forgotten");
assertTrue(sessions.asPresence().isPresent(stillLive.terminalId()),
"a still-live member's presence must survive an unrelated release");
}
// --- fleetd #316: the dirty check must be re-taken after the worker is stopped, not trusted
// stale from before it ------------------------------------------------------------------------
@@ -2582,160 +2436,4 @@ class SessionManagerTest {
return "threw:" + e.getClass().getName() + ":" + e.getMessage();
}
}
// --- fleetd #702: spawnedMemberRole must still answer for a pane mid-teardown ----------------
/**
* A {@link Worktrees} test double whose {@code hasUncommitted} runs an injected hook before
* answering. This is what lets a test resolve a releasing pane's terminal from inside the
* window {@link SessionManager#release} opens between removing the registry entry and
* actually stopping the pane — the hook runs synchronously on the release call's own thread,
* at the exact point {@code release} shells out to {@code git status}, so there is no sleep
* and no race to land in it.
*/
private static final class HookedWorktrees implements Worktrees {
private Runnable hook;
private boolean dirty = false;
HookedWorktrees onHasUncommitted(Runnable hook) {
this.hook = hook;
return this;
}
@Override
public String add(String repoRoot, String branch, String baseRef) {
return "/wt/" + branch.replace('/', '_');
}
@Override
public void remove(String repoRoot, String worktreePath) {
}
@Override
public void deleteBranch(String repoRoot, String branch) {
}
@Override
public boolean hasUncommitted(String worktreePath) {
if (hook != null) {
hook.run();
}
return dirty;
}
@Override
public void overlayParity(String repoRoot, String worktreePath, List<String> overlay) {
}
@Override
public String repoRoot(String cwd) {
return "/repo";
}
@Override
public java.util.Optional<String> snapshot(String worktreePath, String branch, String message) {
return java.util.Optional.empty();
}
@Override
public WipRefStats wipRefs(String repoRoot) {
return new WipRefStats(0, 0L);
}
@Override
public int pruneWipRefs(String repoRoot, long minAgeMillis) {
return 0;
}
@Override
public void shareWithGroup(String repoRoot, String worktreePath) {
}
}
@Test
void spawnedMemberRoleResolvesTheLiveRegisteredRole() {
FakeHerdr herdr = new FakeHerdr();
SessionManager sessions = sessionManager(herdr);
MemberSession s = sessions.acquire("ltms-local", null, "/caller", null);
assertEquals(s.role(), sessions.spawnedMemberRole(s.terminalId()),
"a registered session resolves to its own role");
assertNull(sessions.spawnedMemberRole("term_unknown"),
"a terminal with no session at all resolves to null");
}
@Test
void spawnedMemberRoleIsNullOnceReleaseFullyCompletes() {
FakeHerdr herdr = new FakeHerdr();
SessionManager sessions = sessionManager(herdr, new HookedWorktrees());
MemberSession s = sessions.acquire("ltms-local", null, "/caller/proj", null,
new WorktreeRequest("fleetd-702a", null));
sessions.release(s.paneId());
assertNull(sessions.spawnedMemberRole(s.terminalId()),
"once release has fully finished, the terminal is neither registered nor releasing");
}
@Test
void spawnedMemberRoleStillAnswersBetweenTheRegistryRemovalAndThePaneStop() {
FakeHerdr herdr = new FakeHerdr();
java.util.concurrent.atomic.AtomicReference<MemberRole> duringWindow = new java.util.concurrent.atomic.AtomicReference<>();
HookedWorktrees worktrees = new HookedWorktrees();
SessionManager sessions = sessionManager(herdr, worktrees);
MemberSession s = sessions.acquire("ltms-local", null, "/caller/proj", null,
new WorktreeRequest("fleetd-702b", null));
worktrees.onHasUncommitted(() -> {
// This runs from INSIDE release()'s git-status shell-out: the registry entry is
// already gone, but the pane has not stopped yet — a real call landing inside the
// exact window fleetd #702 reports, so no sleep and no race is needed to reach it.
assertTrue(sessions.get(s.paneId()).isEmpty(),
"sanity: the registry entry is already gone at this point");
duringWindow.set(sessions.spawnedMemberRole(s.terminalId()));
});
sessions.release(s.paneId());
assertEquals(s.role(), duringWindow.get(),
"spawnedMemberRole must still answer the live role while the pane is mid-teardown, "
+ "not only while the session is still in the registry");
assertNull(sessions.spawnedMemberRole(s.terminalId()),
"and once release has fully finished, the window is closed too");
}
/**
* {@code Releasing.enter} and {@code Releasing.leave} are called directly here, never through
* {@link SessionManager#release} or {@link SessionManager#spawnedMemberRole}, so a test naming
* one of them exercises only that one — a regression in the other can never hide behind it.
*/
@Test
void releasingLeaveStepsDownADepthGreaterThanOneInsteadOfRemovingIt() {
SessionManager.Releasing depthTwo = new SessionManager.Releasing(2, "term_a", MemberRole.DEV);
SessionManager.Releasing afterLeave = depthTwo.leave();
assertNotNull(afterLeave,
"depth 2 means another release of the SAME pane is still mid-teardown; leave() must "
+ "step the depth down, never remove the marker outright — removing it here "
+ "is what a plain Set would do, and would reopen the window the still-in-"
+ "flight release is relying on staying closed");
assertEquals(1, afterLeave.depth());
assertEquals("term_a", afterLeave.terminalId());
assertEquals(MemberRole.DEV, afterLeave.role());
}
@Test
void releasingEnterPreservesThePriorTerminalWhenTheOverlappingCallHasNoSessionOfItsOwn() {
SessionManager.Releasing prior = new SessionManager.Releasing(1, "term_a", MemberRole.DEV);
SessionManager.Releasing afterEnter = SessionManager.Releasing.enter(prior, null);
assertEquals(2, afterEnter.depth(), "depth still increments whether or not this enter knows its session");
assertEquals("term_a", afterEnter.terminalId(),
"an overlapping release that finds the registry entry already gone has no session of "
+ "its own to pass as known, and must not blank out the terminal the first "
+ "call already recorded — that terminal is what spawnedMemberRole matches "
+ "against");
assertEquals(MemberRole.DEV, afterEnter.role());
}
}
@@ -159,40 +159,6 @@ class WorktreeSessionManagerTest {
assertEquals(expectedPath, s.cwd(), "session cwd is the worktree path");
}
// --- fleetd #722: registration and presence must reach READY whichever lands first --------
@Test
void registerThenContactReachesReadyForWorktreeSpawn() {
FakeHerdr herdr = new FakeHerdr();
FakeWorktrees worktrees = new FakeWorktrees().withRepoRoot("/repo").withPrefix("/wt");
SessionManager sessions = new SessionManager(workerService(herdr), worktrees);
MemberSession session = sessions.acquire("ltms-local", null, "/caller/proj", "term_primary",
new WorktreeRequest("cb-722", null));
sessions.asPresence().markPresent(session.terminalId());
assertEquals(MemberSession.State.READY, sessions.get(session.paneId()).orElseThrow().state(),
"a presence contact that arrives after worktree registration reaches READY");
}
@Test
void contactThenRegisterStillReachesReadyForWorktreeSpawn() {
// The racing launcher marks presence for the spawned terminal from inside spawn() —
// before SessionManager.acquireWithWorktree's own registry.put runs — modeling an MCP
// contact that lands in that window.
FakeHerdr herdr = new FakeHerdr();
FakeWorktrees worktrees = new FakeWorktrees().withRepoRoot("/repo").withPrefix("/wt");
PresenceRacingLauncher race = new PresenceRacingLauncher(workerService(herdr));
SessionManager sessions = new SessionManager(race, worktrees);
race.presence = sessions.asPresence();
MemberSession session = sessions.acquire("ltms-local", null, "/caller/proj", "term_primary",
new WorktreeRequest("cb-722", null));
assertEquals(MemberSession.State.READY, sessions.get(session.paneId()).orElseThrow().state(),
"a presence contact that lands before worktree registration must still reach READY");
}
@Test
void worktreeArchitectAcquireAlsoBindsItsSlot() {
FakeHerdr herdr = new FakeHerdr();