Skip to content

ci: add a nightly run of the slow and long tests - #144

Merged
mattmattox merged 8 commits into
mainfrom
claude/nightly-ci
Nov 16, 2026
Merged

mattmattox merged 8 commits into
mainfrom
claude/nightly-ci

Conversation

@mattmattox

Copy link
Copy Markdown
Contributor

Runs at 03:17 UTC (and on dispatch) what PR CI skips. Every job stays under an hour.

  • Integration without -short: 3 groups balanced by integration-timings-full.txt, from a full run on .68 (531s total, ~183s per group).
  • Docker with the concurrent tests: the 8 tests SKIP_CONCURRENT_TESTS skips move to run-tests.sh group 4; nightly calls docker-integration.yml (now workflow_call) with groups 1–4. PR runs are unchanged. The junit check-run is off for nightly so it doesn't mark main's commit.
  • -race -count=3 on lock, journal, inode, dir (~6 min).
  • Fuzz: targets are discovered with go test -list '^Fuzz' (11 today), so new ones need no edit; 25 min each, corpus cached per target, failing inputs uploaded.

Failures fail the run; the Summary job lists failed tests (tests/scripts/failed-tests.sh) and uploads nightly-failures.tsv, one suite<TAB>test per line — the key a later issue step would dedupe on. Nothing posts to issues; that waits for the user's OK.

This PR triggers a short variant (1 min fuzzing, no corpus save) to check the wiring.

🤖 Generated with Claude Code

mattmattox and others added 3 commits November 14, 2026 09:16
nightly.yml runs at 03:17 UTC what pull request CI skips: the integration
suite without -short in three timed groups, the Docker suite with its
concurrent tests as group 4, -race -count=3 on lock, journal, inode and
dir, and every FuzzXxx for 25 minutes with its corpus cached per target.
A failure fails the run; the Summary job lists the failed tests and
uploads them as nightly-failures. Nothing posts to issues.

docker-integration.yml becomes callable; pull requests keep groups 1-3
without the concurrent tests. ci.yml checks every workflow's GO_VERSION.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
GitHub expressions compare null and false as equal, so
inputs.publish-report != false skipped the report on pull requests and
pushes. A skip-report input that defaults to false reads null as false.
The fuzz job's TMPDIR moves to the fuzz step, after the directory
exists, since setup-go runs go env before it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An input that allocates without bound then kills its fuzz worker, which
saves the input, instead of the runner OOM-killing the job first.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
mattmattox and others added 4 commits November 14, 2026 09:56
The nightly runs TestOneFailedWriteKeepsAcknowledgedResults with
GOCLUSTERFS_EIO_SWEEP=metadata under -race (19 min on a Linux host; full
ran past 2 hours) and fails if the test did not run, since a -run that
matches nothing passes. tests/scripts/fuzz-budgets.txt gives targets
10m or 45m instead of 25m, from 2h runs that show which ones stop
finding coverage early and which still find it at 2h.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A pull request that changes the nightly checks the sweep's wiring with
the default sample instead of waiting on the 19-minute metadata sweep,
as fuzzing runs for 1m there.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@mattmattox
mattmattox merged commit d85bbe0 into main Nov 16, 2026
47 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant