[Copilot]

Remove the queue drain's batch barrier so a slow repository no longer leaves workers idle while other repositories have queued work.

The drain currently claims up to 16 queue items and processes their repositories with four workers. Workers can pick up more repositories within that batch, but the next claim waits for the entire batch to finish. Near the end of a batch, one slow repository can leave three workers idle.

For example, if a batch contains repositories A, B, C, and D, and B, C, and D finish first, repositories E, F, and G in the queue must still wait for A. This PR lets them start alongside A.

Rolling scheduling

Keep one four-worker pool for the drain. Whenever a repository job finishes, claim eligible work for the available slots without waiting for the other jobs.

Each job still processes a bounded group of items from one repository, preserving sequential processing and coalesced publication. New claims exclude repositories already running in that lane. The live and maintenance lanes each retain their existing four-repository limit.

Deadline and retry behavior

Stop admitting jobs ten minutes before the hard processing deadline, replacing the estimate based on the longest completed batch. Running jobs keep the existing deadline and lease protections, and remaining work passes to the successor drain.

If admission stops after a claim, return unstarted items to the queue without consuming retries or marking them complete. If a newer event arrived during the claim, its follow-up gets a fresh retry budget.

This changes scheduling, not dashboard evaluation or publication rules. Stable/canary code selection, delivery leases, and the promoted release remain unchanged. The expected benefit is less idle time between batches, not higher concurrency; no end-to-end speedup has been measured yet.

[opentelemetry-pr-dashboard]

Pull request dashboard status

Merged · refreshed 2026-10-28 13:09 UTC

Status above doesn't look right?

• Anything look wrong? Report it with what you expected; it helps us improve the dashboard.

[Copilot]

🟡 Changes recommended

The refill loop can start additional repository jobs after an existing job has already failed, violating the intended stop-on-failure behavior.

1 open issue

• Prevent refill submissions after a repository job fails · New

What changed in this PR

Replaces batch-based dashboard queue draining with a rolling scheduler that refills available repository slots without waiting for slower jobs.

Changes:

• Adds persistent four-repository scheduling, admission deadlines, and per-repository job handling.
• Extends the claim API with repository exclusions and limits, and locks shared execution caches.
• Updates scheduler and API tests and documents rolling execution.

File	Description
pull-request-dashboard/​README.md	Describes rolling queue processing.
.github/​scripts/​pull-request-dashboard/​test_owned_execution.py	Adapts owned-execution tests to repository jobs.
.github/​scripts/​pull-request-dashboard/​test_drain_queue.py	Covers rolling scheduling, concurrency, and failure handling.
.github/​scripts/​pull-request-dashboard/​test_dashboard_queue.mjs	Tests repository filtering and claim limits.
.github/​scripts/​pull-request-dashboard/​test_dashboard_queue_worker.mjs	Tests endpoint validation and filter forwarding.
.github/​scripts/​pull-request-dashboard/​RATIONALE.md	Updates the scheduling rationale.
.github/​scripts/​pull-request-dashboard/​netlify/​lib/​dashboard-queue.mjs	Adds repository-aware claim selection.
.github/​scripts/​pull-request-dashboard/​netlify/​functions/​dashboard-queue-worker.mjs	Validates and forwards optional claim filters.
.github/​scripts/​pull-request-dashboard/​execution_code.py	Locks shared code and checkout caches.
.github/​scripts/​pull-request-dashboard/​drain_queue.py	Implements rolling repository jobs and slot refill.
.github/​scripts/​pull-request-dashboard/​CONTEXT.md	Documents scheduling, deadlines, and checkout reuse.

_{Review effort: Balanced}

[Copilot]

🟡 Changes recommended

Unstarted claims can be incorrectly dead-lettered, an empty-claim race can delay scheduling, and an overlap test depends on thread timing.

3 open issues

• Prevent unstarted claims from being dead-lettered · New
• Retry claims when jobs finish during empty requests · New
• Keep first future pending to avoid claim overlap race · New

1 resolved since last review

• Prevent refill submissions after a repository job fails

_{Review effort: Balanced}

[Copilot]

🟡 Changes recommended

Requeueing dirty claims retains the previous generation’s retry count, which can prematurely dead-letter follow-up work.

1 open issue

• Reset retry attempts when requeueing claims with follow-ups · New

3 resolved since last review

• Prevent unstarted claims from being dead-lettered
• Keep first future pending to avoid claim overlap race
• Retry claims when jobs finish during empty requests

_{Review effort: Balanced}

[Copilot]

🟢 Approval recommended

No unresolved blocking issues remain, and regression coverage addresses scheduling races, admission cutoffs, and retry-preserving requeue behavior.

No open issues 🎉

1 resolved since last review

• Reset retry attempts when requeueing claims with follow-ups

_{Review effort: Balanced}