anthony.wen/cds-ai
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fix categorized ATVM watcher host result recovery

Browse Source
This commit is contained in:
anthony.wen
2026-03-30 14:02:32 -04:00
parent 89f558bd39
commit 1405a2e879
4 changed files with 117 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
atvm/watcher-service/README.md
View File

@@ -62,9 +62,9 @@ Typical workflow:
- mark the final state
- exit
5. For categorized runs:
- detect each grouped sub-run in sequence from the parent run log
- wait for that grouped sub-run to finish
- send one Mattermost post for that grouped sub-run if it reached `COMPLETED` or `FAILED`
- detect each grouped sub-run in sequence from the parent run log
- wait for that grouped sub-run to finish
- send one Mattermost post for that grouped sub-run if it reached `COMPLETED` or `FAILED`
- keep the watcher alive while the parent categorized runner or related child Cypress process is still active
- do not treat one completed grouped sub-run as proof that the whole parent request is finished
- continue to the next grouped sub-run
@@ -129,5 +129,8 @@ This writes a cancellation marker, updates `state.json` to `CANCELLED`, and stop
- Categorized execution is treated as sequential grouped ATVM sub-runs, not as one parent run with internal phases.
- In categorized mode, the watcher writes per-subrun state under `subruns/` and posts each completed grouped run separately.
- In categorized mode, if the child build id label does not match the host/spec actually being executed, the watcher reports the grouped run using the inferred host-based group name instead of trusting the raw child build id label.
- In categorized mode, grouped XML can finish with only `check-xml-files.ts`; when that happens, the watcher must recover per-host results from the matching host reporter artifacts.
- Do not infer `PASS completed` from host artifact presence alone. Parse the per-host reporter result and preserve real `FAIL` and `RUN/pending` state when reconstructing grouped results.
- When the repo copy of the watcher changes, the controller install under `/opt/atvm-watcher-service` must be updated before expecting the new reporting behavior from live runs.
- Best-practice controller install path: `/opt/atvm-watcher-service`.
- This package is local-only right now. Nothing here is installed on the controller yet.