fleet/fleetd
3
0
Fork 0
Code Issues 73 Pull Requests Actions Packages Projects Releases 1 Wiki Activity

CB-637: relabel lead-comms (was colliding CB-635) + document coordId route in the CLAUDE.md tool table

Browse Source 
The lead-to-lead wiring shipped with CB-635 in its comments, but CB-635 is
already the broker.uriEnv / unreachable-broker work. Relabel the mailbox +
fleet_send{coordId} + receive loop to CB-637 so a ticket number names one
feature. Add the cross-host peer-lead row to the primary intent->tool table
(kept byte-identical with the wiki template).
This commit is contained in:
Dai Ha
2026-08-24 17:48:51 +02:00
parent 1dbe3a03fc
commit edabccd885
5 changed files with 15 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files
CLAUDE.md
+2 -1
View File
@@ -112,7 +112,8 @@ the merge — and merging on a reviewer's word is delegating it by proxy.
| 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` |
| Message a **peer lead** | `fleet_send{sessionId: <their terminal>, content}` — `fleet_list` → `leads` reports it. Coordination only, **never** a task |
| 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 |
| Answer a peer lead that messaged you | `fleet_reply{content}` — the one case a lead replies |
| Collect a held reply | `fleet_poll{target}` · then `fleet_ack{target, msgId}` |
| Tear down a member | `fleet_stop{paneId}` |
bridged/src/main/java/dev/ltms/fleet/Fleetd.java
+6 -6
View File
@@ -84,7 +84,7 @@ public final class Fleetd {
/** CB-504: how long to wait at startup for herdr's socket before serving degraded. */
private static final long HERDR_WAIT_SECONDS = 30;
/**
* CB-635: how often the lead coordination loop looks for peer messages. A few seconds — slow
* CB-637: how often the lead coordination loop looks for peer messages. A few seconds — slow
* enough that an idle fleet is not polling a broker in a tight loop, fast enough that a peer
* lead's message is not left sitting once the local lead reaches a turn boundary. The mailbox
* pushes into the loop's held set on its own consumer thread, so this interval bounds only the
@@ -387,7 +387,7 @@ public final class Fleetd {
// unusable), bridged stays soft-state on the in-memory inbox. The AMQP inbox owns a broker
// connection, so keep the reference to close it in the ordered shutdown hook.
final ReplyInbox replyInbox = selectReplyInbox(cfg.broker(), System.getenv(), AmqpReplyInbox::open);
// CB-635: this daemon's lead-to-lead mailbox on the SHARED coordination vhost — a separate
// CB-637: this daemon's lead-to-lead mailbox on the SHARED coordination vhost — a separate
// broker from the reply inbox by design (see FleetConfig.Coordinator). Absent a coordinator:
// block this is null and every lead path below is simply not wired, which is exactly the
// behaviour before this ticket. It owns a broker connection, so keep the reference for the
@@ -524,7 +524,7 @@ public final class Fleetd {
}, quarantine),
leadMailbox);
// CB-635: the receive half. Only constructed when a lead mailbox actually opened — with no
// CB-637: the receive half. Only constructed when a lead mailbox actually opened — with no
// coordinator (or an unreachable one) there is nothing to deliver, so no scheduler is
// created and no thread runs. It reads the SAME live lead supplier the injector's
// deliverability gate does, so a lead found by the tab scan after startup is reachable
@@ -562,7 +562,7 @@ public final class Fleetd {
messages.close();
pushLoop.close();
if (heartbeat != null) heartbeat.close(); // CB-551: stop the idle-lead heartbeat scheduler
if (leadCoordLoop != null) leadCoordLoop.close(); // CB-635: stop delivering peer-lead messages
if (leadCoordLoop != null) leadCoordLoop.close(); // CB-637: stop delivering peer-lead messages
if (leadCoordSchedulerRef != null) leadCoordSchedulerRef.shutdownNow();
if (healthMonitor != null) healthMonitor.stop();
if (configWatcher != null) configWatcher.stop(); // CB-559: stop polling the config file
@@ -576,7 +576,7 @@ public final class Fleetd {
log.debug("reply inbox close: {}", e.toString());
}
}
// CB-635: the coordination connection goes with it — after the loop that reads it has
// CB-637: the coordination connection goes with it — after the loop that reads it has
// stopped, so no tick can be mid-ack against a closed channel.
if (leadMailbox != null) {
try {
@@ -631,7 +631,7 @@ public final class Fleetd {
}
/**
* CB-635: open this daemon's lead-to-lead mailbox, or return {@code null} to leave the feature
* CB-637: open this daemon's lead-to-lead mailbox, or return {@code null} to leave the feature
* off. Package-private and env-injected for the same reason as {@link #selectReplyInbox}: the
* selection is then testable without a broker or a mutable process environment.
*
bridged/src/main/java/dev/ltms/fleet/mcp/FleetMcp.java
+5 -5
View File
@@ -100,7 +100,7 @@ public final class FleetMcp {
private final CapacitySource capacity;
private final HealthCoverageSource healthCoverage;
private final QuarantineSource quarantine;
/** CB-635: this daemon's lead-to-lead channel; {@code null} when no coordinator is configured. */
/** CB-637: this daemon's lead-to-lead channel; {@code null} when no coordinator is configured. */
private final LeadChannel leadChannel;
/** Capacity facts used by {@code fleet_list}; production must supply the placement live count. */
@@ -141,7 +141,7 @@ public final class FleetMcp {
}
/**
* As above, with this daemon's lead-to-lead channel (CB-635). {@code leadChannel} is
* As above, with this daemon's lead-to-lead channel (CB-637). {@code leadChannel} is
* {@code null} whenever no {@code coordinator:} block is configured or its broker could not be
* reached at boot — cross-daemon lead messaging is simply off, and {@code fleet_send{coordId}}
* says so rather than failing obscurely.
@@ -199,7 +199,7 @@ public final class FleetMcp {
String turnId = str(a, "turnId");
String coordId = str(a, "coordId");
if (coordId != null && !coordId.isBlank()) {
// CB-635: a peer LEAD on another daemon, addressed by coord-id over the shared
// CB-637: a peer LEAD on another daemon, addressed by coord-id over the shared
// coordination broker. Checked before the turnId branch so a call that sets both
// is rejected as the conflict it is, rather than silently taking one route.
return sendToLead(leadChannel, coordId, content, target, turnId);
@@ -614,7 +614,7 @@ public final class FleetMcp {
}
/**
* {@code fleet_send} carrying a {@code coordId} (CB-635): a message to a PEER LEAD, published to
* {@code fleet_send} carrying a {@code coordId} (CB-637): a message to a PEER LEAD, published to
* that lead's durable mailbox on the shared coordination broker. This is the only lead→lead path
* that crosses hosts — the existing pane-injection route can only reach a lead whose herdr socket
* this daemon shares.
@@ -950,7 +950,7 @@ public final class FleetMcp {
}
/**
* As above, additionally reporting this daemon's own lead coordination id (CB-635) when one is
* As above, additionally reporting this daemon's own lead coordination id (CB-637) when one is
* configured and its channel opened. There is no peer-discovery surface yet — a lead addresses a
* peer by a coord-id it was told — so this row exists to answer the one question the operator
* cannot answer any other way: what is MY coord-id, the one a peer must use to reach me. It is
bridged/src/test/java/dev/ltms/fleet/FleetdLeadMailboxSelectionTest.java
+1 -1
View File
@@ -15,7 +15,7 @@ import java.util.Map;
import static org.junit.jupiter.api.Assertions.*;
/**
* CB-635: the daemon decides whether lead-to-lead messaging is on in
* CB-637: the daemon decides whether lead-to-lead messaging is on in
* {@link Fleetd#openLeadMailbox}, not in the config record — so testing
* {@code Coordinator.isConfigured()} alone would pass even if {@code Fleetd} never honoured it.
* These drive the real selection with an injected env map and an injected opener, so no broker is
bridged/src/test/java/dev/ltms/fleet/mcp/FleetMcpLeadCoordTest.java
+1 -1
View File
@@ -8,7 +8,7 @@ import org.junit.jupiter.api.Test;
import static org.junit.jupiter.api.Assertions.*;
/**
* CB-635: the {@code fleet_send{coordId}} route — a message to a PEER LEAD on another daemon,
* CB-637: the {@code fleet_send{coordId}} route — a message to a PEER LEAD on another daemon,
* published to its durable mailbox instead of typed into a pane this daemon can reach.
*
* <p>Hermetic: {@link FakeLeadChannel} replaces the AMQP-backed {@code LeadMailbox}, so these run