[ScriptedAlchemy]

Problem

context_scout_tests::lifecycle_less_boundary_settles_and_resumes_on_its_sessions_lifecycle failed 7 of 40 runs on master. Five failures were a saved edit refused as Backpressured, one was ([214], true) where ([214], false) was expected, and one run hung until the 300 s test timeout.

Root cause

There were two causes, one in production and one in the test.

1. Production. BoundedHookOrchestratorV1 fired a settled cycle's completion terminals while the cycle still held its concurrency permit. The permit was dropped only after settle_operation returned. A host that sends the session's next boundary as soon as it sees the terminal was refused as Backpressured. A lifecycle resume admitted in that window was dropped silently, which caused the hang.
2. Test. The test treated the resumed cycle as settled once its work reported running. A resumed cycle has no terminal. While it is still registered, the next saved edit supersedes it and inherits its lifecycle, as designed. That produced the ([214], true) failure.

Fix

• settle_operation now takes the cycle's permit. It drops the permit after it removes the in-flight entry and before it fires the terminals.
• The test waits for the resumed cycle to settle (active_tasks() == 0) before it sends the next boundary.

Verification

• New test settled_cycle_admits_the_sessions_next_boundary_from_its_terminal admits the next boundary from inside the terminal callback. Before the fix it fails every run with left: Backpressured, right: Enqueued. That commit comes first in the branch.
• //crates/tracedecay-daemon-service:unit_test --runs_per_test=40: both Context Scout tests passed 40/40. Two runs failed in an unrelated test, profile_host_admission_replay::tests::not_applicable_pending_record_does_not_back_off (left: 3, right: 2).
• Mutation. A resume that leaves the boundary parked fails the lifecycle-less test 20/20.
• bazel build --config=clippy //crates/tracedecay-daemon-service/... and rustfmt_check_2024 pass. The types_tests and advisory tests in //crates/tracedecay:unit_test pass (32 tests).

Reconciled with #4144

#4144 fixed the same race from the other side. It dropped the permit before settle_operation removed the in-flight entry. The merge keeps one implementation. settle_operation now drops the permit while it holds the in-flight lock, right after it removes the entry. Admission checks the in-flight maps and takes its permit under that same lock, so the two steps are atomic for every admission. With that, neither ordering's window remains:

• With fix(daemon): root-cause seven load-only daemon test failures #4144's order, a replay could see the cycle still in flight after its permit was freed.
• With this PR's earlier order, an admission could find the cycle gone while its permit was still held.

The terminals fire after the lock is released.

Verification after the merge:

• Daemon-service unit_test ran 40x. The Context Scout tests passed 40/40. One run failed in the unrelated not_applicable_pending_record_does_not_back_off.
• //crates/tracedecay:unit_test, filtered to types_tests and advisory, passed 20x. That includes fix(daemon): root-cause seven load-only daemon test failures #4144's replayed_admission_after_retryable_failure_completes_a_fresh_cycle.
• With a temporary 20 ms std::thread::sleep between the unlock and the terminals, the Context Scout tests and the replayed-admission test passed 10/10.
• Clippy and rustfmt pass.

[changeset-bot]

⚠️ No Changeset found

Latest commit: e8d3d0e

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR