Skip to content

fix: refuse name-based run IDs and fail seeds whose audits fail - #1469

Merged
chio-labs-delivery[bot] merged 2 commits into
mainfrom
fix/run-id-and-test-skips
Nov 14, 2026
Merged

chio-labs-delivery[bot] merged 2 commits into
mainfrom
fix/run-id-and-test-skips

Conversation

@kvlonge

@kvlonge kvlonge commented Nov 14, 2026

Copy link
Copy Markdown
Contributor

Why

  • The server derives version 5 (name-based) run IDs for group members, retries, sensor runs and scheduled runs. A person could send one of those IDs in a run, rerun or ad-hoc request and collide with a run the server will create later.
  • A seed whose error-severity audit failed was reported as a success. Its dependants were skipped, but the run page, the Failures list and "Rerun from failure" had no failed node to point at, and the integration event said the seed completed.

Changes

  • POST /api/control/runs, /runs/adhoc and /runs/rerun (and sqb server admin request-run/rerun-run) refuse a version 5 run_id with a 400 that says why. The server's own derived IDs don't pass through this check. The OpenAPI document, the TypeScript client and the server docs say which IDs to send.
  • A seed's audits now run inside the seed's attempt, on its own connection. A failing error-severity audit fails the seed with R006 ("audit for seed '…' failed after loading"), the same way a failing final audit fails a model. The terminal summary counts the seed and the audit, as it does for models, and the attempt is recorded as failed.
  • DuckShop gains an orchestration edge case, a seed with a failing audit.

Verification

  • New API test: a v5 run_id on all three endpoints returns 400 with the message on SQLite and PostgreSQL, and a random ID is still accepted. It fails with the check disabled.
  • New worker integration test: on a server run the seed is failed with R006 and its message, and the model reading it is skipped. It fails on main.
  • New e2e test: the seed's integration record is resource_attempt_failed with R006 when its audit fails, and completed when it passes.
  • Updated seed-audit e2e expectations; the seed-audit, worker, executor and result-parity suites pass (85 tests).
  • make check-ci passes on the final head.

…c requests

The server derives version 5 run IDs for group members, retries, sensor
and scheduled runs. A person-supplied version 5 ID could collide with one
of those, so job run, rerun and ad-hoc requests now refuse it with a 400
that says why. Internal paths keep their derived IDs.
The seed was reported as a success before its audits ran, so server runs
showed it succeeded while its dependants were skipped. The audits now run
first and a failing one turns the seed result into a failure with R006,
as a failing final audit does for a model.
@kvlonge kvlonge added the full-verify Run the full Verify matrix on this pull request label Nov 14, 2026
@chio-labs-delivery
chio-labs-delivery Bot added this pull request to the merge queue Nov 14, 2026
Merged via the queue into main with commit f83b16c Nov 14, 2026
45 of 67 checks passed
@chio-labs-delivery
chio-labs-delivery Bot deleted the fix/run-id-and-test-skips branch November 14, 2026 03:28
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

full-verify Run the full Verify matrix on this pull request

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant