lsm / neokai / 28280172830
71%

DEFAULT BRANCH: dev
Ran
Jobs 0
Files 0
Run time –
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
28280172830

push

github

web-flow 
fix(github-poller): stop /pulls backlog from starving check_run polling (#2204)

* fix(github-poller): stop /pulls backlog from starving check_run polling

- Add a client-side cutoff for /pulls pagination. GitHub ignores the
  `since` query param on /pulls, so the previous code walked every open
  PR every poll. With sort=updated&direction=desc, rows are newest-first;
  stop paginating once a row is older than the endpoint watermark. This
  clears persistent processedPages.pulls backlogs instead of letting the
  low-rate-limit guard leave them stranded.
- Decouple check-run polling from the pulls backlog. Already-known PR
  heads (and newly discovered ones on the pages we do fetch) still get
  their check runs polled even when processedPages.pulls > 1.
- Add unit tests for the backlog/cutoff scenario and for a newly opened
  PR whose CI fails while /pulls is still catching up.

* test(github-poller): cover mixed-age /pulls page cutoff regression

Add a regression test for the primary real-world case where a /pulls
page spans the watermark boundary: the first rows are newer than the
endpoint watermark and the remaining rows are older. Asserts that the
newer rows are head-tracked, published, and that processedPages.pulls
resets to 1 without paginating past the cutoff.

* fix(github-poller): only apply /pulls cutoff after pulls-specific watermark exists

When a legacy cursor has tracked head SHAs but no endpointLastSeenAt.pulls,
the endpoint watermark falls back to the shared lastSeenAt. That shared
watermark may have been advanced by comments or check runs past a PR's own
updated_at, so applying the client-side cutoff would drop the /pulls row
before the head/open-state refresh below could run. Skip the cutoff until a
pulls-specific watermark has been seeded; this treats the cycle as a seed
fetch and prevents stale heads from lingering in the cursor.

Also add a regression test for this legacy-cursor case.

* fix(github-poller): skip stale cursor heads ... (continued)
Source Files on build 28280172830
Detailed source file information is not available for this build.