fix(meta): preserve prior quarantine when critical check errors on full run (#680)

[shobhitagnihotri69]

Resolves #680

Summary

On a full meta stage run, when a critical check encounters an infrastructure crash or error (run.result is None), the previous quarantine tags were inadvertently dropped from episodes_latest. This occurred because the quarantine carry-over block in src/hflow/app.py was only executed if Stage.META not in enabled_stages or if running a step subset (SelectedRegisteredSteps).

Per maintainer guidance in #680:

• On a full meta run, carry over only the quarantined:<name> tags whose check ran in this invocation and errored (run.result is None).
• Tags for checks no longer registered in the pipeline are dropped on full runs rather than preserved forever.
• If a check produces a passing verdict (verdict=True), the quarantine is cleared as expected.

Changes

• src/hflow/app.py:
  • Track errored_check_names = {run.check.name for run in report.checks if run.result is None}.
  • On full meta runs where checks errored, inspect QuarantineHistory and retain only quarantined:<name> tags where name in errored_check_names.
• tests/test_run_profiles.py:
  • Added test_full_meta_run_preserves_quarantine_when_critical_check_crashes: verifies the quarantined -> crashed -> still quarantined cycle, ensures downstream stages are skipped with SkippedByQuarantine, and tests quarantine clearing upon a passing verdict.
  • Added test_full_meta_run_drops_quarantine_for_unregistered_checks: verifies that unregistered check quarantine tags are not retained indefinitely on full runs.

Verification

• uv run ruff check & uv run ruff format --check passed cleanly.
• Full test suite passed: 2,467 passed, 49 skipped.

[greptile-apps]

Confidence Score: 5/5

[Medium risk] Changes quarantine preservation logic during check failures.

The PR appears safe to merge; no outstanding issue was found.

What we checked:

• Superseded check keeps quarantine: No. A full run carries an old tag only when the check is critical and has an error status.
• Non-critical check blocks work: No. The carry-over filter requires the current check to be critical, and the test confirms that its old tag clears.

Summary

Full meta runs now keep a prior quarantine only when its matching critical check errors, while passing or no-longer-active checks clear old tags. This keeps an infrastructure crash from letting downstream enrichment run under a quarantine that was already recorded.

• Partial runs keep their existing quarantine carry-over behavior.
• Tests cover crashes, passing checks, and tags for checks that are removed, superseded, or no longer critical.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Full meta run] --> B{Critical check errored?}
  B -- Yes --> C[Keep its prior quarantine tag]
  B -- No --> D[Do not carry its prior tag]
  C --> E[Save current quarantine state]
  D --> E

Loading

_{Reviews (3) · Last reviewed commit: "test(profiles): assert catalog_entry is ..."}

[kstonekuan]

LGTM, thanks for covering the retired, non-critical, and superseded cases. Merging.

[shobhitagnihotri69]

Thanks for merging this one , looking forward to contribute more and more