feat(scoring): claim a due job and drive the request-bound snapshot worker - #233
Closed
cursor[bot] wants to merge 38 commits into
Closed
cursor[bot] wants to merge 38 commits into
cursor[bot] wants to merge 38 commits into
Conversation
Cover InvalidCompletionEnvelope on Linux llvm-cov by constructing the variant and asserting its exact Display text and absent source, plus the exact Completion/Outbox wrapper messages.
Add real PostgreSQL cases for the crash-between-old-APIs path this composition exists to repair, and record the envelope's actual source/job/time gate so callers keep event identity stable. Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
A crashed worker can replay the same quarantined cause without inventing a score. A different cause fails closed instead of rewriting typed failure evidence. Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Quarantine and the bound integration event share one caller-owned transaction. Exact same-cause replay is idempotent, a different cause fails closed, mixed legacy dispositions reconcile, and a late outbox conflict rolls the quarantine back. Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Point the active-work slot at the completion-plus-failure successor so reviewers do not treat #69 as a separate landable head. Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
A crashed scoring worker must reuse the job-plus-result or job-plus-cause outbox identity instead of minting a second event after an accepted terminal write. Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Route fenced completion and permanent failure through the existing outbox helpers only after the envelope reuses the job-plus-result or job-plus-cause event_ref. A minted identity is rejected before write. Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Ask ScoringWorkerEngine first, bind the stable job-plus-result or job-plus-cause event identity, then commit through the existing terminal helper so a retry cannot mint a second outbox row. Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Keep job and outcome references that contain ':' from sharing one outbox event_ref after concatenation. Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Engine or planner errors stay ScoringWorkerCommitError::Planning and leave the leased job untouched. A stale fence after a stable identity check surfaces Completion or Failure without writing a second terminal event. Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
A later different result or a cause after accepted completion must fail closed and keep the original stable event_ref as the only terminal row. Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Add ScoringRequest::from_persisted so a worker that still has scoring_request_ref can rebuild the AssessmentSpec, instrument, scoring, calibration, and optional norm pin without inventing a score or reloading response events. Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Load the reconstructed scoring request, validate the caller envelope before asking a request-bound engine, and persist the immutable result snapshot in the same transaction as the fenced job and stable outbox event. A missing request, planner failure, or snapshot conflict leaves the job leased and does not invent a score. Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Complete the worker snapshot inside the unexpired lease window, reject a job that names a different scoring request before the engine runs, and add Display/source plus PostgreSQL evidence for corrupt request, planner failure, snapshot conflict, and engine quarantine without an invented score. Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Compose #203's job/request bind with a typed Retryable engine/transport outage so a live miss stays pending: no snapshot, no terminal outbox row, and no invented score. Exhausted retry budget quarantines the same way. Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
…orker A hosted worker can now claim one due queued or retry-scheduled job and run the request-bound snapshot path using only the stored pin. A later due claim after a retryable outage persists the real snapshot. A missing job row, an early retry claim, or a retry scheduled before the outage instant fails closed without inventing a score. Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
Contributor
Author
There was a problem hiding this comment.
Verdict: SOUND for a named-job claim, but do not land this in parallel with #234
Exact head 99f2e5e. This head claims one caller-named due job and drives run_scoring_worker_attempt_with_result_snapshot using the stored pin. Missing job, early retry, and InvalidRetryWindow fail closed. That slice is sound.
#234 dc14101 is the hosted landing: it selects the oldest due queued or retry-scheduled row with FOR UPDATE SKIP LOCKED, then scores only the stored pin. A buyer-facing worker cannot require the operator to name the job. Prefer #234 over this head, #217, #214, #203, #200, and #190. Leave those predecessors unmerged.
Independent non-author approval and required checks remain the merge gates on the chosen landing. This review does not approve or merge.
Sent by Cursor Automation: Fix Issues
Contributor
|
Closing this Draft as the named-job scoring-worker landing superseded by current #234. The independent Cursor review on this exact #233 lineage already classified the slice as sound but explicitly selected #234 as the hosted landing because #234 chooses the oldest due queued/retry-scheduled row with |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Succeeds #214. Prefer this head over #214, #203, #200, #190, #172, #168, #167, #155, #136, #115, and #69; do not land those predecessors separately.
Why
A buyer who finished the items still has no hosted worker that claims a due job and drives the request-bound snapshot path. #214 binds the job to the stored request and schedules retryable outages, but a queued or retry-scheduled row stays idle until a caller already holds a lease and supplies the pin. Landing #214 alone leaves that hosted hole.
What this adds
claim_and_run_scoring_worker_attempt_with_result_snapshotclaims one due queued or retry-scheduled job, scores only the stored request pin, and drivesrun_scoring_worker_attempt_with_result_snapshot.next_attempt_atstaysLeaseNotDuewith no snapshot or outbox.InvalidRetryWindow, job leased, no score.InvalidReference.What this does not add
fast-mlsirmexecutionReviewer action
Review this successor and leave #214, #203, and #200 unmerged. Independent non-author approval and required checks on this unchanged exact head remain the merge gates. Never self-approve.