[vikramlc-cognite]

Summary

Start <task> now reports how the run actually went (failed, completed_with_errors, succeeded), and Stop <task> reports succeeded only once the task has ended.

Type of change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature that changes existing behavior)
• Refactor (no functional change)
• Documentation update
• Chore / tooling / CI

What changed

• tasks.py: thread-safe ErrorSummary on TaskContext, recording the count and first description/details per ErrorLevel.
• base.py _start_task_outcome (rule, in this order): cancelled -> canceled; any fatal error or a raising task -> failed with the first fatal error's description and details as resultMessage (truncated); else any error-level error -> completed_with_errors ("Completed with N error(s): "); else succeeded (warnings only or nothing).
• Stop <task>: reports running, waits until the task has ended, then succeeded. failed if the task is idle ("not currently running") or has not ended within STOP_TASK_ACTION_TIMEOUT_SECONDS (300 s); canceled if the Stop itself is cancelled (shutdown, user).
• _dto.py: ActionStatus.completed_with_errors. Module docstring documents the rule; CHANGELOG.
• Scheduled (non-action) runs are unaffected.

Why it changed

• Tasks report problems on their context instead of raising, so a failed run ended succeeded (DB broken-query, File with a missing source). Stop reported succeeded as soon as the signal was sent.
• Related issue: EDG-1035 (parent: EDG-1031)
• Related docs / discussion: PM decision: failed run -> Failed with the error text, non-fatal errors -> "Completed with Errors", Stop only once stopped, one rule for all extractors. Stacked on EDG-1033. Needs Odin completed_with_errors (cognitedata/odin#324).

What to focus on during review

• The mapping order (cancel wins over errors) and which levels count as run-ending. The DB extractor reports a whole failed query at level error, so until it reports it as fatal (follow-up in db-extractor) it ends completed_with_errors, not failed.
• The Stop wait: it runs on its own action thread, is registered as a running action so shutdown and cancel can wake it, and has a 300 s bound.
• One existing test was changed (it would block for 300 s under the new Stop behaviour).

Test evidence

• tests/test_unstable/test_action_outcome.py: 14 tests (fatal, truncation, error-only, ctx.exception, warning-only, none, raising task, cancel, scheduled run, Stop waits, Stop timeout, Stop idle, Stop on shutdown)
• uv run pytest tests/test_unstable: 300 passed; the same 1 failed + 48 environment errors as master
• ruff, ruff format, mypy, pre-commit run --all-files pass
• 8 of the 14 new tests fail with base.py/tasks.py reverted; the other 5 guard unchanged behaviour

Risks and unknowns

• Behaviour change for every extractor: runs that report errors no longer end succeeded.
• Until Odin deploys completed_with_errors (Add status codes to time series queue #324), a check-in carrying it may be rejected. Deploy Odin first.
• A Stop on a task that is stuck in a blocking call keeps the action running in Odin for up to 300 s.

Rollout and rollback

• Release order: Odin Add status codes to time series queue #324 first, then this. Revert the commit to roll back.

Checklist

• Self-reviewed the diff
• Tests added or updated (or N/A with reason)
• Docs updated (or N/A): module docstring, CHANGELOG
• No secrets, credentials, or PII committed
• Breaking changes called out above and communicated to affected teams: action outcome semantics changed (PM-approved); pending product communication

[gemini-code-assist]

Code Review

This pull request introduces a new 'completed_with_errors' action status for tasks and improves the reporting of task outcomes. It adds an 'ErrorSummary' class to track errors during task execution, allowing 'Start ' actions to report more accurate statuses (failed, completed_with_errors, or succeeded) based on reported errors. Additionally, it updates 'Stop ' actions to wait for the task to finish before reporting success, with a configurable timeout. There are no review comments to address.

[codecov]

Codecov Report

❌ Patch coverage is 97.10145% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 85.25%. Comparing base (e046cd1) to head (370f43a).

Files with missing lines	Patch %	Lines
cognite/extractorutils/unstable/core/base.py	96.00%	2 Missing ⚠️

Additional details and impacted files

@@                     Coverage Diff                     @@
##           EDG-1033-clean-shutdown     #591      +/-   ##
===========================================================
+ Coverage                    84.93%   85.25%   +0.31%     
===========================================================
  Files                           46       46              
  Lines                         4819     4882      +63     
===========================================================
+ Hits                          4093     4162      +69     
+ Misses                         726      720       -6     

Files with missing lines	Coverage Δ
cognite/extractorutils/unstable/core/_dto.py	98.44% <100.00%> (+0.01%)	⬆️
cognite/extractorutils/unstable/core/tasks.py	100.00% <100.00%> (ø)
cognite/extractorutils/unstable/core/base.py	89.31% <96.00%> (+0.87%)	⬆️

... and 2 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.