fleetd #284: free failed-session capacity
CI / contract (pull_request) Successful in 1m11s
CI / build (pull_request) Successful in 1m19s

This commit is contained in:
Dai Ha
2026-09-04 10:55:21 +07:00
parent 4dd12083ab
commit 11cbfa79b4
5 changed files with 37 additions and 16 deletions
@@ -858,14 +858,15 @@ public final class Fleetd {
}
/**
* Count sessions that occupy a profile's spawn capacity. A {@code BACKEND_ERROR} session stays
* in the roster so {@code fleet_list} can show its failure, but its dead backend cannot use a
* seat or accept another delivery.
* Count sessions that occupy a profile's spawn capacity. A {@code BACKEND_ERROR} or
* {@code FAILED} session stays in the roster so {@code fleet_list} can show its failure, but a
* member that cannot accept another delivery does not use a seat.
*/
static int liveSessionCount(List<MemberSession> roster, String profileName) {
return (int) roster.stream()
.filter(session -> profileName.equals(session.profile()))
.filter(session -> session.state() != MemberSession.State.BACKEND_ERROR)
.filter(session -> session.state() != MemberSession.State.FAILED)
.count();
}
@@ -322,8 +322,9 @@ public record FleetConfig(
* statement that does not stop being true just because the profile was
* named directly. A negative value has no sane meaning (there is no
* "excluded" to degrade to below zero) and is refused at config load
* instead, naming the profile and the key. Live means any session the
* registry still owns (acquired and not yet released), in any state.
* instead, naming the profile and the key. Live means a session that can
* receive another delivery. The roster keeps terminal {@code BACKEND_ERROR}
* and {@code FAILED} sessions for diagnostics, but they do not use capacity.
* @param kind which peer launcher spawns this profile: {@code "claude-code"} (default —
* the {@link dev.ltms.fleet.member.ClaudeCodeLauncher}) or {@code "opencode"}.
* The {@code CompositePeerLauncher} routes {@code spawn}/reap by this value, so
@@ -1196,7 +1196,7 @@ public final class FleetMcp {
int leadSeatCount = leadSeats.seatsFor().apply(profile);
int reclaimable = (int) roster.stream().filter(s -> profile.equals(s.profile()))
.filter(s -> s.state() == MemberSession.State.READY || s.state() == MemberSession.State.DONE
|| s.state() == MemberSession.State.BACKEND_ERROR)
|| s.state() == MemberSession.State.BACKEND_ERROR || s.state() == MemberSession.State.FAILED)
.filter(s -> messages == null || (!messages.hasAcceptedDelivery(s.terminalId()) && !messages.hasInboxMessage(s.terminalId())))
.count();
Map<String, Object> row = new LinkedHashMap<>();
@@ -230,6 +230,28 @@ class FleetdBackendErrorSinkTest {
@Test
@DisplayName("a backend-error session no longer blocks the real maxLoad spawn gate")
void backendErrorSessionDoesNotBlockFreshSpawnAtMaxLoad() {
SessionManager sessions = capacityLimitedSessions();
MemberSession failed = sessions.acquire("terra", null, null, null);
assertTrue(sessions.onBackendError(failed.terminalId(), "backend exited"));
MemberSession fresh = sessions.acquire("terra", null, null, null);
assertEquals("terra", fresh.profile(), "the real maxLoad gate grants a fresh spawn after a backend error");
}
@Test
@DisplayName("a failed session no longer blocks the real maxLoad spawn gate")
void failedSessionDoesNotBlockFreshSpawnAtMaxLoad() {
SessionManager sessions = capacityLimitedSessions();
MemberSession failed = sessions.acquire("terra", null, null, null);
sessions.onTurnFailed(failed.terminalId());
MemberSession fresh = sessions.acquire("terra", null, null, null);
assertEquals("terra", fresh.profile(), "the real maxLoad gate grants a fresh spawn after a failed turn");
}
private static SessionManager capacityLimitedSessions() {
FakeHerdr herdr = new FakeHerdr();
FleetConfig.Profile profile = new FleetConfig.Profile("terra", "http://gx00.gw:8000", "coder",
null, "FLEETD_WORKER_TOKEN", List.of("claude"), "tab", "fleetd-workers",
@@ -242,11 +264,6 @@ class FleetdBackendErrorSinkTest {
PlacementPolicies.fixed(), name -> Fleetd.liveSessionCount(sessionsRef.get().roster(), name));
SessionManager sessions = new SessionManager(workers);
sessionsRef.set(sessions);
MemberSession failed = sessions.acquire("terra", null, null, null);
assertTrue(sessions.onBackendError(failed.terminalId(), "backend exited"));
MemberSession fresh = sessions.acquire("terra", null, null, null);
assertEquals("terra", fresh.profile(), "the real maxLoad gate grants a fresh spawn after a backend error");
return sessions;
}
}
@@ -605,18 +605,20 @@ class FleetMcpTest {
}
@Test
void capacityReportsABackendErrorSessionAsReclaimable() {
void capacityReportsTerminalFailureSessionsAsReclaimable() {
FakeHerdr h = new FakeHerdr();
SessionManager sessions = new SessionManager(workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")));
MemberSession session = sessions.acquire("ltms-local", null, null, null);
assertTrue(sessions.onBackendError(session.terminalId(), "backend exited"));
MemberSession backendError = sessions.acquire("ltms-local", null, null, null);
MemberSession failed = sessions.acquire("ltms-local", null, null, null);
assertTrue(sessions.onBackendError(backendError.terminalId(), "backend exited"));
sessions.onTurnFailed(failed.terminalId());
String out = textOf(FleetMcp.listFleet(workerService(h, "http://gx00.gw:8000", Set.of("gx00.gw")),
sessions, null, new FleetMcp.CapacitySource(profile -> 0, profile -> 1,
() -> Set.of("ltms-local"), () -> 0), new FleetMcp.HealthCoverageSource(() -> "off"),
FleetMcp.QuarantineSource.none(), Map.of(), ""));
assertTrue(out.contains("\"reclaimable\":1"), out);
assertTrue(out.contains("\"reclaimable\":2"), out);
}
@Test