Compare commits

...

5 Commits

Author SHA1 Message Date
Dai Ha 279d6f5fbd fleetd #257: free must stop subtracting leadSeatCount
CI / build (pull_request) Successful in 1m43s
CI / contract (pull_request) Successful in 1m51s
fleet_list's free row subtracted leadSeatCount, but the real spawn gate
(CompositePeerLauncher#enforceMaxLoad) only ever compares live against
maxLoad and never reads leadSeatCount. So free could report 0 while a
fleet_spawn on that exact profile still succeeded, and a lead trusting
free:0 gave up on capacity the gate would still grant.

free now always equals max(0, maxLoad - live); leadSeats stays in the
row as an informational fact, never subtracted. Documented in the
fleet_list tool description and fleetd.example.yaml.
2026-09-03 16:28:32 +07:00
Dai Ha 3bad9f5785 Merge #260: restore the already-gone-worktree teardown test (#116)
CI / contract (push) Successful in 58s
CI / build (push) Successful in 2m9s
2026-09-03 16:13:15 +07:00
Dai Ha 8308c0b68f fleetd #116: recover the already-gone-worktree teardown regression test
CI / contract (pull_request) Successful in 1m16s
CI / build (pull_request) Successful in 1m23s
Ports the intent of the lost CB-576 commit c393600 (worker/cb576-01a04b-17,
never merged, package dev.ltms.bridged.*) onto main's dev.ltms.fleet.*
tree. Adds FakeWorktrees.markGone (tracks which add()'d worktree paths
still "exist", mirroring GitWorktrees.hasUncommitted's Files.exists guard
for the already-gone case) and a regression test,
releaseStillStopsPaneAndNotifiesWhenWorktreeIsGone, asserting that
SessionManager.release still fires notifyReleased, stops the pane, and
falls through to remove() when the worktree is already gone.
2026-09-03 16:11:27 +07:00
Dai Ha 3b3063eb2b #252: the route scrape counted commented-out registrations as live
CI / build (push) Successful in 1m44s
CI / contract (push) Successful in 2m21s
Follow-up to #259, found by verifying the guard rather than trusting it.

The scrape read FleetApp's raw source, so a registration disabled with `//`
still matched. Measured: commenting out `app.get("/tasks/{ticket}", ...)` left
the test GREEN, while deleting the same line was caught. Only the commented-out
shape was blind, and it is the silent direction — the inventory would keep
claiming a route the server no longer serves.

Drop whole-line comments before scraping. Only lines starting with //, * or /*
are dropped, deliberately not every // on a line: that would also cut a string
literal containing // (a URL) and could silently delete a real registration
sharing the line. The remaining gap is a trailing comment beside real code; no
registration here has that shape, and the vacuity test catches a scrape that
loses registrations wholesale.

Proof: with the fix, the comment-out mutation fails naming
"Removed ... [GET /tasks/{ticket}]"; reverted, FleetApp.java confirmed clean.
mvn clean install: 0 compile errors, 1257 tests, BUILD SUCCESS.
2026-09-03 15:59:35 +07:00
Dai Ha df9086263d Merge #259: guard the REST route inventory against drift (#252) 2026-09-03 15:56:15 +07:00
6 changed files with 225 additions and 49 deletions
+28 -21
View File
@@ -307,16 +307,22 @@ profiles:
# and no refusal on cost; the cap on live members is the single thing standing between a
# fan-out and your monthly limit. Set it deliberately and keep it small.
#
# GOTCHA 3 (fleetd #176) — `maxLoad` counts members, never the lead itself. The lead is a live
# `claude` session on this SAME account (a lead is never moved off-subscription, whatever its
# own profile says), so it already holds one seat before any member spawns. If a lead's
# `fleet.leaders.<name>.profile` names THIS profile — or ANY OTHER `subscription: true`
# profile that shares this one's account (see THE SENTINEL, just below, next to
# `credentialId:`) — `fleet_list`'s `free` for this profile subtracts that lead's live
# seat(s) automatically; see `profile:` under THE FLEET below. If no lead entry names a
# profile sharing this account, fleetd has no way to know a lead holds a seat here, and `free`
# will overstate what a fresh `fleet_spawn` actually gets by exactly the seats the lead is
# quietly holding.
# GOTCHA 3 (fleetd #176, corrected by fleetd #257) — `maxLoad` counts members, never the lead
# itself. The lead is a live `claude` session on this SAME account (a lead is never moved
# off-subscription, whatever its own profile says), so it already holds one seat before any
# member spawns. If a lead's `fleet.leaders.<name>.profile` names THIS profile — or ANY OTHER
# `subscription: true` profile that shares this one's account (see THE SENTINEL, just below,
# next to `credentialId:`) — `fleet_list` reports that seat count under `leadSeats`; see
# `profile:` under THE FLEET below. `free` itself is NEVER reduced by `leadSeats`: `free` means
# "what the real placement gate (`CompositePeerLauncher#enforceMaxLoad`) will actually grant a
# fresh `fleet_spawn` right now", and that gate only ever compares live members against
# `maxLoad` — it has no notion of the lead's own seat. An earlier cut of this feature
# subtracted `leadSeats` from `free` on the theory it made `free` describe the true ceiling on
# the account, but no backend seat ceiling shared with the lead has ever actually been
# measured, and the subtraction just made `free` disagree with the one thing it is supposed to
# describe — the fleetd #257 fix. `maxLoad: 3` means 3 member slots, full stop; a lead sharing
# the account is a fact you can see in `leadSeats`, not a reason `free` undercounts spawns that
# will, in practice, succeed.
#
# THE SENTINEL (fleetd #176 stage 2, correcting an inert stage 1 fix): every `subscription:
# true` profile that leaves `credentialId` unset shares ONE implicit account-wide credential
@@ -536,17 +542,18 @@ fleet:
# auto-launched: it is also how fleetd learns which account this lead's own session shares. A
# `subscription: true` profile bills the operator's Claude account, and the lead itself is always
# a live `claude` session on that same account — `maxLoad` never counted that seat. If a lead
# entry here names a profile that shares a worker profile's account, `fleet_list`'s `free` for
# that worker profile subtracts the lead's live seat(s) automatically. "Shares the account" is
# decided by matching `effectiveCredentialId()`, which (fleetd #176 stage 2 — see THE SENTINEL,
# next to `credentialId:`, in THE WORKERS above) means: an explicit, matching `credentialId:` on
# both, OR — the common case, needing NO extra config — both being `subscription: true` with
# `credentialId` left unset, since those all share one implicit account-wide id. A lead on `opus`
# and workers on `sonnet` link automatically this way; they do NOT need the same profile name.
# Setting `profile:` on an already-running, recognise-only lead is safe — the daemon only launches
# the SHORTFALL below `instances`, so naming a profile here does not, by itself, start anything.
# Omit it and fleetd has no way to derive the sharing — there is no other reliable signal on the
# daemon's side — so that lead's seat goes uncounted, exactly as before this ticket.
# entry here names a profile that shares a worker profile's account, `fleet_list` reports the
# lead's live seat(s) on that worker profile under `leadSeats` — informational only, as of fleetd
# #257 it is NEVER subtracted from `free` (see GOTCHA 3, next to `maxLoad:`, in THE WORKERS above,
# for why). "Shares the account" is decided by matching `effectiveCredentialId()`, which (fleetd
# #176 stage 2 — see THE SENTINEL, next to `credentialId:`, in THE WORKERS above) means: an
# explicit, matching `credentialId:` on both, OR — the common case, needing NO extra config — both
# being `subscription: true` with `credentialId` left unset, since those all share one implicit
# account-wide id. A lead on `opus` and workers on `sonnet` link automatically this way; they do
# NOT need the same profile name. Setting `profile:` on an already-running, recognise-only lead is
# safe — the daemon only launches the SHORTFALL below `instances`, so naming a profile here does
# not, by itself, start anything. Omit it and fleetd has no way to derive the sharing — there is
# no other reliable signal on the daemon's side — so that lead's seat never appears in `leadSeats`.
#
# `tab:` (CB-579) is REQUIRED and is the only field identity depends on — the exact label of the
# tab hosting the lead, matched case-insensitively. Label the tab yourself and put that same
@@ -139,16 +139,24 @@ public final class FleetMcp {
/**
* fleetd #176: the seats a profile's own live LEAD session(s) hold on the same Claude
* subscription — the third reason (alongside {@link QuarantineSource} and {@link OutageSource})
* {@code free} can overstate what a fresh {@code fleet_spawn} would actually get.
* subscription — a fact {@code fleet_list} reports alongside {@code free} via the
* {@code leadSeats} key.
*
* <p>{@code maxLoad} counts only <em>members</em>, never the lead itself. But a
* {@code subscription: true} profile bills the operator's own Claude account, and the lead is
* always a live {@code claude} session on that same account (it is never moved off-subscription
* — see {@code LeadLauncher}). So a fan-out that fills every member slot still leaves the lead's
* own seat unaccounted for, and the daemon reports a slot that was never really free. See
* {@code Fleetd.leadSeatLookup} for how the count is derived — from {@code fleet.leaders.<name>
* .profile} and each profile's {@code effectiveCredentialId()}, never a hardcoded constant.
* — see {@code LeadLauncher}). See {@code Fleetd.leadSeatLookup} for how the count is derived —
* from {@code fleet.leaders.<name>.profile} and each profile's {@code effectiveCredentialId()},
* never a hardcoded constant.
*
* <p>fleetd #257: this count is reported, never subtracted from {@code free}. An earlier cut of
* this feature subtracted it, on the theory that it made {@code free} describe the real ceiling
* on the account — but the real spawn gate ({@code CompositePeerLauncher#enforceMaxLoad}) never
* read this count at all, so the subtraction made {@code free} disagree with the one thing it is
* supposed to describe: what a fresh {@code fleet_spawn} will actually get. No backend seat
* ceiling shared with the lead has been measured either — see {@code fleetd.example.yaml}'s
* {@code maxLoad} docs. {@code free} now always equals {@code max(0, maxLoad - live)}, and
* {@code leadSeats} is reported purely as a fact the caller may act on however it likes.
*/
public record LeadSeatSource(Function<String, Integer> seatsFor) {
/** Inert source — no profile is ever reported as sharing a seat with a lead. */
@@ -1137,10 +1145,17 @@ public final class FleetMcp {
* <p>fleetd #176: {@code maxLoad} counts panes, not subscription seats — it never counted the
* lead's own seat on a {@code subscription: true} profile's account. {@link LeadSeatSource}
* reports that count (0 for a non-subscription profile, or when no live lead shares its
* credential), and it is subtracted from {@code free} the same way {@code live} already is —
* {@code maxLoad} itself is left untouched, so the row still reports the configured cap. The
* {@code leadSeats} key is added only when the count is positive, for the same
* byte-identical-when-unused reason as the quarantine/cool-off keys above.
* credential) via the {@code leadSeats} key, added only when the count is positive, for the
* same byte-identical-when-unused reason as the quarantine/cool-off keys above.
*
* <p>fleetd #257: {@code leadSeatCount} is reported, never subtracted from {@code free}.
* {@code free} means "what a fresh {@code fleet_spawn} on this profile will actually get", and
* the real gate ({@code CompositePeerLauncher#enforceMaxLoad}) only ever compares {@code live}
* against {@code maxLoad} — it has no notion of a lead's own seat. Subtracting
* {@code leadSeatCount} here made {@code free} disagree with the gate it is supposed to
* describe: it could report {@code free: 0} while a spawn on that exact profile still
* succeeded. {@code leadSeats} stays in the row as a fact the caller can act on however it
* likes, but it no longer changes what {@code free} means.
*/
private static Map<String, Object> capacityView(String profile, Function<String, Integer> liveCount,
Function<String, Integer> maxLoad, List<MemberSession> roster,
@@ -1155,7 +1170,12 @@ public final class FleetMcp {
.count();
Map<String, Object> row = new LinkedHashMap<>();
row.put("profile", profile); row.put("maxLoad", cap); row.put("live", live);
row.put("free", cap == null ? null : Math.max(0, cap - live - leadSeatCount));
// fleetd #257: free must report what the real spawn gate (CompositePeerLauncher#enforceMaxLoad)
// will actually grant, and that gate never reads leadSeatCount — only maxLoad and live. Not
// subtracting the lead's seat here used to make free UNDERSTATE what a fresh fleet_spawn would
// get, so a lead believing free:0 gave up on a profile the gate would still spawn onto.
// leadSeatCount is still reported via the leadSeats key below, just never subtracted from free.
row.put("free", cap == null ? null : Math.max(0, cap - live));
row.put("reclaimable", reclaimable);
if (leadSeatCount > 0) {
row.put("leadSeats", leadSeatCount);
@@ -1360,8 +1380,13 @@ public final class FleetMcp {
+ "member spawned without one shares its directory with others and never reports "
+ "an id, however long it runs (fleetd #249). An empty 'members' "
+ "means no members are spawned; it says nothing about peers. When capacity "
+ "facts are configured, a 'capacity' row per profile also reports free: 0 for "
+ "a quarantined profile's credential (see fleet_profiles), whatever its "
+ "facts are configured, a 'capacity' row per profile reports 'free' — the "
+ "slots a fresh fleet_spawn on that profile will actually be granted right "
+ "now (max(0, maxLoad - live)), the same check the spawn gate itself runs. A "
+ "'leadSeats' key, when present, reports how many of those live slots are a "
+ "lead session sharing this profile's subscription — informational only, "
+ "already NOT subtracted from 'free' (fleetd #257). It also reports free: 0 "
+ "for a quarantined profile's credential (see fleet_profiles), whatever its "
+ "maxLoad/live — with credentialId and quarantinedForSeconds naming the "
+ "quarantine, so 'free: 0, busy' can be told apart from 'free: 0, refusing "
+ "for N seconds'.",
@@ -34,6 +34,7 @@ import java.util.Map;
import java.util.Set;
import java.util.concurrent.CompletableFuture;
import java.util.concurrent.TimeUnit;
import java.util.function.Function;
import static org.junit.jupiter.api.Assertions.*;
@@ -810,13 +811,16 @@ class FleetMcpTest {
}
/**
* fleetd #176: this is the exact shape measured on the Mac fleet — {@code maxLoad:3, live:2},
* where one of the "free" three is really the lead's own seat on the same subscription. The old
* formula ({@code max(0, cap - live)}) reported {@code free:1}; the real ceiling is {@code 0}
* (two members plus the lead's own seat already fill all three).
* fleetd #257: this is the exact shape measured on the Mac fleet — {@code maxLoad:3, live:2},
* one lead sharing the subscription. The OLD formula ({@code max(0, cap - live - leadSeats)})
* reported {@code free:0} here, disagreeing with the real spawn gate (which never read
* {@code leadSeats} and would still grant one more spawn — see
* {@code freeMatchesWhatTheRealPlacementGateActuallyGrants} below for that proof against the
* actual gate). {@code free} must report {@code 1}: {@code leadSeats} is carried as a fact, not
* subtracted.
*/
@Test
void leadSeatSubtractsFromFreeTheSameWayLiveDoes() {
void leadSeatIsReportedButNeverSubtractedFromFree() {
FakeHerdr h = new FakeHerdr();
SessionManager sessions = new SessionManager(workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")));
FleetMcp.LeadSeatSource leadSeats = new FleetMcp.LeadSeatSource(
@@ -829,17 +833,17 @@ class FleetMcpTest {
assertTrue(out.contains("\"maxLoad\":3"), "maxLoad itself must be left untouched: " + out);
assertTrue(out.contains("\"live\":2"), out);
assertTrue(out.contains("\"free\":0"), "2 live + 1 lead seat fills all 3: " + out);
assertTrue(out.contains("\"leadSeats\":1"), out);
assertTrue(out.contains("\"free\":1"), "free is maxLoad - live only, never minus leadSeats: " + out);
assertTrue(out.contains("\"leadSeats\":1"), "leadSeats is still reported, just not subtracted: " + out);
}
/**
* fleetd #176: the OTHER measurement in the issue — a completely idle fleet still overstates
* {@code free} by the lead's own seat. {@code maxLoad:3, live:0} must report {@code free:2}, the
* real fan-out ceiling, not {@code 3}.
* fleetd #257: the OTHER measurement in the issue — a completely idle fleet with a lead sharing
* the subscription. {@code maxLoad:3, live:0} must report {@code free:3}, matching what the
* spawn gate (which has no notion of a lead's seat) would actually grant.
*/
@Test
void leadSeatLowersFreeOnAnOtherwiseIdleSubscriptionProfile() {
void leadSeatDoesNotLowerFreeOnAnOtherwiseIdleSubscriptionProfile() {
FakeHerdr h = new FakeHerdr();
SessionManager sessions = new SessionManager(workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")));
FleetMcp.LeadSeatSource leadSeats = new FleetMcp.LeadSeatSource(
@@ -851,10 +855,76 @@ class FleetMcpTest {
FleetMcp.QuarantineSource.none(), FleetMcp.OutageSource.none(), leadSeats, Map.of(), "", null));
assertTrue(out.contains("\"live\":0"), out);
assertTrue(out.contains("\"free\":2"), "an idle fleet's real ceiling is 3 minus the lead's own seat: " + out);
assertTrue(out.contains("\"free\":3"), "the real gate never subtracts the lead's seat: " + out);
assertTrue(out.contains("\"leadSeats\":1"), out);
}
/**
* fleetd #257 — the defect, driven against the REAL placement gate, not a copy of its
* arithmetic. {@code free} must equal exactly how many more spawns
* {@link CompositePeerLauncher#spawn} (routed through the same {@link SessionManager} fleet_spawn
* itself uses) will grant on this profile right now: this test reads whatever number
* {@code fleet_list}'s real {@code listFleet} call reports, then drives that many spawns through
* the REAL composite launcher and asserts every one succeeds, and the next one — one past what
* fleet_list promised — is refused. A test that instead hand-computed {@code cap - live} and
* compared it to {@code free} would pass even if both sides shared the same wrong formula (this
* repo has been bitten by exactly that before); this one only passes when fleet_list's number and
* the gate's real behaviour actually agree.
*
* <p>{@code liveCount} here is wired the same way {@code Fleetd.main} wires it in production: one
* function, read by both the {@link CompositePeerLauncher}'s {@code maxLoad} gate and
* {@code fleet_list}'s {@code CapacitySource}, off the SAME {@link SessionManager#roster()} — so
* the two paths cannot silently drift on what "live" means.
*/
@Test
void freeMatchesWhatTheRealPlacementGateActuallyGrants() {
FakeHerdr h = new FakeHerdr();
FleetConfig.Profile wcfg = new FleetConfig.Profile(
"sonnet", "http://gx00.gw:8000", "coder", null, "FLEETD_WORKER_TOKEN", null,
"tab", "fleetd-workers", "worker: {profile} #{n}", null,
null, null, null, null, null, null, null, 3);
Map<String, FleetConfig.Profile> profiles = Map.of(wcfg.profile(), wcfg);
ClaudeCodeLauncher delegate = new ClaudeCodeLauncher(
new AgentControl(h), new WorkspaceControl(h), new SubscriptionGuard(Set.of("gx00.gw")),
profiles, wcfg.profile(), k -> "FLEETD_WORKER_TOKEN".equals(k) ? "tok" : null);
java.util.concurrent.atomic.AtomicReference<SessionManager> smRef =
new java.util.concurrent.atomic.AtomicReference<>();
Function<String, Integer> liveCount = profile -> (int) smRef.get().roster().stream()
.filter(s -> profile.equals(s.profile())).count();
CompositePeerLauncher composite = new CompositePeerLauncher(
List.of(delegate), wcfg.profile(), profiles, PlacementPolicies.fixed(), liveCount);
SessionManager sm = new SessionManager(composite);
smRef.set(sm);
// Two members already live — the exact shape measured in fleetd #257 (maxLoad:3, live:2).
assertFalse(FleetMcp.spawn(sm, "sonnet").isError(), "setup: first live member must spawn cleanly");
assertFalse(FleetMcp.spawn(sm, "sonnet").isError(), "setup: second live member must spawn cleanly");
// A lead session shares this profile's subscription — leadSeats:1, same as the ticket.
FleetMcp.LeadSeatSource leadSeats = new FleetMcp.LeadSeatSource(p -> "sonnet".equals(p) ? 1 : 0);
String out = textOf(FleetMcp.listFleet(composite, sm, null,
new FleetMcp.CapacitySource(liveCount, p -> profiles.get(p).maxLoad(), profiles::keySet, () -> 0),
new FleetMcp.HealthCoverageSource(() -> "off"), FleetMcp.QuarantineSource.none(),
FleetMcp.OutageSource.none(), leadSeats, Map.of(), "", null));
int reportedFree = extractInt(out, "free");
for (int i = 0; i < reportedFree; i++) {
McpSchema.CallToolResult res = FleetMcp.spawn(sm, "sonnet");
assertFalse(res.isError(), "fleet_list promised free:" + reportedFree + "; spawn #" + (i + 1)
+ " of that many was refused by the real gate: " + textOf(res));
}
McpSchema.CallToolResult overflow = FleetMcp.spawn(sm, "sonnet");
assertTrue(overflow.isError(), "fleet_list reported free:" + reportedFree
+ " but the real placement gate granted at least one more spawn than that: " + textOf(overflow));
}
private static int extractInt(String json, String key) {
java.util.regex.Matcher m = java.util.regex.Pattern.compile("\"" + key + "\":(-?\\d+)").matcher(json);
assertTrue(m.find(), "no \"" + key + "\" field in: " + json);
return Integer.parseInt(m.group(1));
}
/** A profile with no lead seats reported must be byte-identical to before this ticket. */
@Test
void zeroLeadSeatsOmitsTheKeyAndLeavesFreeUnchanged() {
@@ -7,6 +7,7 @@ import java.util.Locale;
import java.util.Set;
import java.util.regex.Matcher;
import java.util.regex.Pattern;
import java.util.stream.Collectors;
import org.junit.jupiter.api.DisplayName;
import org.junit.jupiter.api.Test;
@@ -68,6 +69,32 @@ class RestRouteInventoryTest {
"GET /tasks/{ticket}"
);
private static final Pattern ROUTE_CALL =
Pattern.compile("app\\.(get|post|delete|put|patch)\\(\\s*\"([^\"]+)\"");
/**
* Drop whole-line comments before scraping. Without this the scrape reads commented-out code as
* live: a registration disabled with {@code //} still matched, so the route stayed in the
* inventory while the server no longer served it — a silent false PASS, measured on 2026-09-03
* by commenting out {@code app.get("/tasks/{ticket}", ...)} and watching this test stay green.
* Deleting the same line was caught correctly, so only the commented-out shape was blind.
*
* <p>Only lines whose first non-blank characters are {@code //}, {@code *} or {@code /*} are
* dropped — deliberately NOT every {@code //} anywhere on a line, because that would also cut a
* string literal containing {@code //} (a URL) and could silently delete a real registration
* sharing that line. The remaining gap is a trailing comment on the same line as real code; no
* registration in this file has that shape, and the vacuity test below would catch a scrape that
* lost registrations wholesale.
*/
private static String withoutCommentLines(String source) {
return source.lines()
.filter(line -> {
String t = line.stripLeading();
return !(t.startsWith("//") || t.startsWith("*") || t.startsWith("/*"));
})
.collect(Collectors.joining("\n"));
}
/**
* Every {@code app.<verb>("<path>")} call in {@link FleetApp}'s source, as {@code "VERB path"}.
* This matches inside an {@code if (...) { ... }} block just as well as a top-level statement —
@@ -75,8 +102,7 @@ class RestRouteInventoryTest {
* catches {@code GET /metrics} (registered conditionally on {@code metrics != null}).
*/
private static Set<String> routesTheServerRegisters() throws Exception {
String source = Files.readString(REST_SOURCE);
Matcher m = Pattern.compile("app\\.(get|post|delete|put|patch)\\(\\s*\"([^\"]+)\"").matcher(source);
Matcher m = ROUTE_CALL.matcher(withoutCommentLines(Files.readString(REST_SOURCE)));
Set<String> found = new LinkedHashSet<>();
while (m.find()) {
found.add(m.group(1).toUpperCase(Locale.ROOT) + " " + m.group(2));
@@ -45,6 +45,9 @@ public final class FakeWorktrees implements Worktrees {
private final List<String> overlayShareOrder = new CopyOnWriteArrayList<>();
private final Set<String> existingPaths = ConcurrentHashMap.newKeySet();
private final Set<String> trackedPaths = ConcurrentHashMap.newKeySet();
/** Worktree paths that currently exist, mirroring GitWorktrees' {@code Files.exists} check for
* the already-gone case (CB-576 review, fleetd #116). */
private final Set<String> worktreePaths = ConcurrentHashMap.newKeySet();
private final AtomicLong snapshotSeq = new AtomicLong();
private volatile RuntimeException addFailure;
private volatile RuntimeException snapshotFailure;
@@ -115,7 +118,15 @@ public final class FakeWorktrees implements Worktrees {
}
// The branch already carries a unique nonce, so the derived path is distinct per acquire
// without an extra counter — keep it a pure function of the branch the test can predict.
return prefix + "/" + branch.replace('/', '_');
String path = prefix + "/" + branch.replace('/', '_');
worktreePaths.add(path);
return path;
}
/** Model an operator / {@code git worktree prune} removing the worktree before release. */
public FakeWorktrees markGone(String worktreePath) {
worktreePaths.remove(worktreePath);
return this;
}
@Override
@@ -125,6 +136,11 @@ public final class FakeWorktrees implements Worktrees {
@Override
public boolean hasUncommitted(String worktreePath) {
// A path that was never added, or was marked gone, is reported clean, mirroring
// GitWorktrees' already-gone guard — never an error, so teardown still completes.
if (!worktreePaths.contains(worktreePath)) {
return false;
}
return dirty;
}
@@ -258,6 +258,38 @@ class WorktreeSessionManagerTest {
}
}
/**
* CB-576 review (fleetd #116). A worktree that is already gone (operator cleanup,
* {@code git worktree prune}, an earlier half-completed release) must not break teardown.
* {@code hasUncommitted} reports the missing path clean (mirroring {@code GitWorktrees}), so
* {@code release} still runs {@code notifyReleased} (the CB-516 fast-fail for a blocked
* {@code fleet_send} caller) and {@code launcher.stop} (so the pane is not orphaned), and falls
* through to the already-gone-tolerant {@code remove}. Since CB-581 this all happens because the
* notify-and-stop work sits in {@code release}'s {@code finally}/post-try block rather than a
* checked branch — this test pins that shape by construction.
*/
@Test
void releaseStillStopsPaneAndNotifiesWhenWorktreeIsGone() {
FakeHerdr herdr = new FakeHerdr();
FakeWorktrees worktrees = new FakeWorktrees().withRepoRoot("/repo").withPrefix("/wt");
SessionManager sessions = new SessionManager(workerService(herdr), worktrees);
List<SessionManager.ReleaseDetail> released = new java.util.concurrent.CopyOnWriteArrayList<>();
sessions.onRelease(released::add);
MemberSession s = sessions.acquire("ltms-local", null, "/caller/proj", null,
new WorktreeRequest("cb-576g", null));
worktrees.markGone(s.worktree());
sessions.release(s.paneId());
assertEquals(1, released.size(),
"notifyReleased must still fire when the worktree is already gone (CB-516)");
assertEquals(s.terminalId(), released.getFirst().terminalId());
assertTrue(herdr.called("pane.close"),
"the pane must still be stopped when the worktree is already gone");
assertEquals(1, worktrees.removeCalls().size(),
"release still calls the already-gone-tolerant remove");
}
@Test
void drainAllPreservesWorktreeOfIdleSession() {
FakeHerdr herdr = new FakeHerdr();