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

fleetd #226: reserve architect slots before launch #228

Merged
ltms merged 2 commits from worker/cb226-architect-slot-race-cd3aa8-3 into main 2026-09-02 02:54:52 +02:00
Conversation 0 Commits 2 Files Changed 5
+227 -69
agent commented 2026-09-02 02:45:27 +02:00
Member
Copy Link
Copy Source

Fixes fleetd #226.

Production change

  • Add atomic architect-slot reservations in MemberRegistry.
  • SessionManager reserves before launch, binds after launch, and releases on every RuntimeException path before a successful bind.
  • A full architect pool refuses before the launcher starts a process. The acquired fallback and WARN remain for an unexpected failed reservation bind.
  • Correct the terminalToSlot lock comment.

Tests

  • aContendedArchitectSlotRefusesBeforeTheLauncherStartsAMember drives SessionManager.acquire with a real ClaudeCodeLauncher/FakeHerdr and proves agent.start was never called. This is the proof that no member receives a wrong charter.
  • aSecondArchitectOnAnAlreadyBoundProfileIsHeldAsDevNotArchitectAndWarnsLoudly reserves successfully, simulates a racer taking the released slot so bind fails, then drives SessionManager.acquire. It asserts DEV and checks the WARN names the profile and terminal. I watched this test fail first: Tests run: 1, Failures: 1, Errors: 0; expected ARCHITECT but was DEV.
  • reservedArchitectSlotBindsTheLaunchedMember checks the successful live binding.
  • failedArchitectLaunchReleasesItsReservationForTheNextLaunch makes a real spawn fail its readiness gate, then a retry succeeds as ARCHITECT.
  • Updated the worktree capacity expectation to refuse before launch.

Charter receipt correction
I dropped the claim that CharterReceipt.role proves charter delivery. That field echoes the request. The no-agent.start contention test is the criterion 4 proof.

Reservation release
I counted two acquire branches: normal and worktree. In the normal branch, one RuntimeException catch covers launcher.spawn and later setup before return; it releases an unbound reservation. In the worktree branch, the outer RuntimeException catch releases the reservation for worktree setup, launcher.spawn, and setup after launch. A successful bind removes the reservation and creates the live terminal binding, so it must not be released. The failed-readiness-then-retry test covers the launcher exception path.

Build
mvn clean install
Tests run: 1092, Failures: 0, Errors: 0, Skipped: 0
BUILD SUCCESS

Fixes fleetd #226. Production change - Add atomic architect-slot reservations in MemberRegistry. - SessionManager reserves before launch, binds after launch, and releases on every RuntimeException path before a successful bind. - A full architect pool refuses before the launcher starts a process. The acquired fallback and WARN remain for an unexpected failed reservation bind. - Correct the terminalToSlot lock comment. Tests - aContendedArchitectSlotRefusesBeforeTheLauncherStartsAMember drives SessionManager.acquire with a real ClaudeCodeLauncher/FakeHerdr and proves agent.start was never called. This is the proof that no member receives a wrong charter. - aSecondArchitectOnAnAlreadyBoundProfileIsHeldAsDevNotArchitectAndWarnsLoudly reserves successfully, simulates a racer taking the released slot so bind fails, then drives SessionManager.acquire. It asserts DEV and checks the WARN names the profile and terminal. I watched this test fail first: Tests run: 1, Failures: 1, Errors: 0; expected ARCHITECT but was DEV. - reservedArchitectSlotBindsTheLaunchedMember checks the successful live binding. - failedArchitectLaunchReleasesItsReservationForTheNextLaunch makes a real spawn fail its readiness gate, then a retry succeeds as ARCHITECT. - Updated the worktree capacity expectation to refuse before launch. Charter receipt correction I dropped the claim that CharterReceipt.role proves charter delivery. That field echoes the request. The no-agent.start contention test is the criterion 4 proof. Reservation release I counted two acquire branches: normal and worktree. In the normal branch, one RuntimeException catch covers launcher.spawn and later setup before return; it releases an unbound reservation. In the worktree branch, the outer RuntimeException catch releases the reservation for worktree setup, launcher.spawn, and setup after launch. A successful bind removes the reservation and creates the live terminal binding, so it must not be released. The failed-readiness-then-retry test covers the launcher exception path. Build mvn clean install Tests run: 1092, Failures: 0, Errors: 0, Skipped: 0 BUILD SUCCESS
agent added 1 commit 2026-09-02 02:45:28 +02:00
#226 reserve architect slots before launch
CI / contract (pull_request) Successful in 1m19s
Details
CI / build (pull_request) Failing after 1m33s
Details
e694deace3
agent added 1 commit 2026-09-02 02:50:37 +02:00
#226 retain architect fallback coverage
CI / contract (pull_request) Successful in 1m17s
Details
CI / build (pull_request) Successful in 1m33s
Details
b40f477210
ltms merged commit cc919aa2b6 into main 2026-09-02 02:54:52 +02:00
Sign in to join this conversation.
Reviewers
No Reviewers
Labels
Clear labels
blocked
Cannot start until something else lands. The body says what.
 needs-live-proof
Merged and green, but never shown working on the running daemon. Not the same as done.
 ready-to-delegate
Scope, files and acceptance criteria are written. A worker can be briefed from the body alone.
 silent-default
A feature that compiles, passes tests, and ships turned off. Nine recurrences and counting.
No Label
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
agent kevin ltms
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: fleet/fleetd#228
Delete Branch "worker/cb226-architect-slot-race-cd3aa8-3"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?