You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
E2E: AIM service stop is not confirmed on Strix Halo Ubuntu (serve-aim-lifecycle-parity, STOP FAILED rc=1) #1058
The serve-aim-lifecycle-parity scenario (tests/e2e-cucumber/features/model_serving.feature, @nightly, @requires-engine:aim) fails on a Strix Halo (gfx1151) Ubuntu runner with Docker available. Stopping the AIM service is not confirmed:
expected the AIM service to stop cleanly, got: aim-...: STOP FAILED (rc=1) Error: Local server stop not confirmed
Where it was seen
A workflow_dispatch of e2e-selfhosted.yml with platform=strix-ubuntu and include_nightly=true (run 43025895270). Both the release and the nightly channel legs failed at the same step. The sibling scenario serve-aim-inference (serve-26) passed on both legs, so the AIM serve itself works and Docker is reachable; only the stop is not confirmed.
Expected
rocm stop on a running AIM service reports the stop as confirmed (exit 0), so the lifecycle scenario can pass.
Notes
Not yet analysed: whether the container is actually left running, or the stop is only reported as unconfirmed.
Until this is fixed, the scenario is declared as an expected failure in tests/e2e-cucumber/expectations.toml for Linux gfx1151, so the nightly report does not show it as an unexpected regression.
Activity
tests: reference the serve-27 xfail row's tracker as rocm-cli#1058 (R…
test(e2e): drop the serve-27 xfail row, the AIM stop now passes on St…