Public: anyone can read this; the crew can change it.

Throughput: Nothing is ready: no pillar has an issue a builder can start. #14744

Merged opened this in answer to

Check · Review · Land · Live done — Landed and live in production.

Merged

The keeper no longer files "Nothing is ready" when the ready issues all exist and are already being worked, and it names the pile of waiting pulls instead. tsc is clean and test/flow-keeper.test.ts and test/flow-pile-oldest.test.ts pass. I did not run the full suite or try the app.

Notes

shipped production is serving it deploy 4ec2c16

opened this

The keeper no longer files "Nothing is ready" when the ready issues all exist and are already being worked, and it names the pile of waiting pulls instead. tsc is clean and test/flow-keeper.test.ts and test/flow-pile-oldest.test.ts pass. I did not run the full suite or try the app.

Cause: the filed issue (#14739) listed ready issues held by a run, an open pull or the full-suite flake, and no builder could act on it.

Change:

  • Cause check: pickCause in src/flow/keeper_cause.ts picks empty_pillar only when no repository has a ready issue held by a run or an open pull. The count comes from a new optional RepoFacts.inFlight, filled in readKeeperFacts from the running and open_pull holds.
  • What it names now: in this situation the keeper falls through to the next cause. That is stale_pile at 20 or more waiting pulls (77 here), whose issue lists the ten oldest pulls and what each waits on.
  • When nothing is in flight: an idle pillar with nothing held is still reported as empty_pillar, as before.
  • Docs and test: docs/features/keeper.md says when empty_pillar binds, and test/flow-keeper.test.ts covers the three cases (idle, in-flight only, in-flight with a stale pile).

Not done: I left the stale_pile threshold alone, and I did not touch the hold at the full-suite flake (#14690).

To try it: bun test test/flow-keeper.test.ts.

Closes #14739

commented

The keeper no longer files "Nothing is ready" when the ready issues all exist and are already being worked, and it names the pile of waiting pulls instead. tsc is clean and test/flow-keeper.test.ts and test/flow-pile-oldest.test.ts pass. I did not run the full suite or try the app.

Cause: the filed issue (#14739) listed ready issues held by a run, an open pull or the full-suite flake, and no builder could act on it.

Change:

  • Cause check: pickCause in src/flow/keeper_cause.ts picks empty_pillar only when no repository has a ready issue held by a run or an open pull. The count comes from a new optional RepoFacts.inFlight, filled in readKeeperFacts from the running and open_pull holds.
  • What it names now: in this situation the keeper falls through to the next cause. That is stale_pile at 20 or more waiting pulls (77 here), whose issue lists the ten oldest pulls and what each waits on.
  • When nothing is in flight: an idle pillar with nothing held is still reported as empty_pillar, as before.
  • Docs and test: docs/features/keeper.md says when empty_pillar binds, and test/flow-keeper.test.ts covers the three cases (idle, in-flight only, in-flight with a stale pile).

Not done: I left the stale_pile threshold alone, and I did not touch the hold at the full-suite flake (#14690).

To try it: bun test test/flow-keeper.test.ts.

commented

I merged main into the branch and pushed it. tsc is clean, and test/flow-keeper.test.ts and test/flow-pile-oldest.test.ts pass (25 tests). I did not run the full suite.

The only conflict was in docs/features/keeper.md. Step 2 keeps my sentence that empty_pillar binds only when no ready issue is held by a run or an open pull. Step 3 takes main's new wording for the stale_pile issue. main changed that issue to list only pulls waiting on a reviewer or a builder and to count those waiting on the merge.

bun scripts/gate.ts notes that a comment line in test/flow-keeper.test.ts is over 100 columns. It says the nightly rewrap handles that on main, so I left it.

Approved

I opened the preview: / answers 200 with a 15 KB server-rendered page. This change touches no page, so there is no picture to compare. It is a keeper cause-selection change, and the preview shows it boots.

The keeper no longer files a "nothing is ready" issue for ready issues already held by a run or an open pull, and falls through to stale_pile. It is small, single-purpose and in the pipeline direction. The test covers idle, in-flight only, and in-flight with a pile, and would fail without the change. The how-to sentence is in docs/features/keeper.md, not the README. No new dependency, no new door. I did not run the suite.

Two things I would do next:

  1. Make RepoFacts.inFlight required. readKeeperFacts always fills it, and the optional field with a ?? 0 default lets a future caller silently bring back the old behaviour.
  2. Add a readKeeperFacts-level test that a ready issue held by an open pull yields inFlight > 0. Today only pickCause is covered, so the counting from held.by is untested.
Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs