Repository navigation
Webhook VM can race a lingering session after stopping the supervisor #893
Description
Problem
The webhook VM test stops agent-box@agent.service before hand-writing /home/agent/.claude/plugins/installed_plugins.json, but a lingering tmux/session child can still rewrite that file. The next agent-box-webhook status then reports that the file has no local-webhook@local-channels entry.
This made post-merge CI for df1bbc2de7cea01e336e497fe83318dea38a3546 fail twice in the webhook lane, while the exact PR tree passed minutes earlier.
Evidence
- Passing PR run: https://github.com/defangdevs/agent-box/actions/runs/39328802104
- Failing post-merge run, attempts 1 and 2: https://github.com/defangdevs/agent-box/actions/runs/39330932921
- Failure:
expected a quiet status ... installed_plugins.json has no local-webhook@local-channels entry - The passing log records the tmux scope exiting during the first status read; the failing logs leave enough time for the stale session path to overwrite the hand-written fixture.
Expected
The test should wait until all session/plugin synchronization writers are quiescent before mutating the cache fixture, or otherwise isolate the fixture from those writers.
Scope
This is independent of the Buzz ACP revert in #889 / draft #892; it was exposed by post-merge verification.
Metadata
Metadata
Assignees
Labels
Projects
- Status
Activity
defangdevs commentedon Oct 26, 2026
Recurred during email PR #929 verification in https://github.com/defangdevs/agent-box/actions/runs/40420337799/job/121912899227 . The test writes the 0.35.0 installed_plugins.json fixture after stopping the supervisor, then the very next status call reports no local-webhook entry. The subsequent run on the same feature branch passed the webhook lane: https://github.com/defangdevs/agent-box/actions/runs/40421071673/job/121915612477 . No status assertion or webhook fixture was weakened in the email work.
defangdevs commentedon Oct 27, 2026
Post-merge CI for 9013d17 (#928) hit this lane again: run 40421752823. Attempt 1 failed with the
installed_plugins.json has no local-webhook@local-channels entryassertion described above. Attempt 2 (rerun of failed jobs) failed differently, at the test script'ssudo -u agent env HOME=/home/agent agent-box-session rm hook-defangdevs-local-channels-9ab4(exit 2) in the session-limit section. That looks like the same lingering-session race rather than a regression in #928, which only touches the Azure template. #928's PR checks were green.defangdevs commentedon Oct 27, 2026
Reproduced three more times, always the 'expected a quiet status … has no local-webhook@local-channels entry' assertion in tests/webhook.nix status(): CI runs 40424988642 (PR #902, both attempts), 40427066153 (PR #902 with master merged) and 40426599144 (codex/231-email-intake). Not caused by #902's hibernation diff. Master was green at 9414ee7.
test(webhook): quiesce tmux after stopping supervisor (#893) (#932)