Skip to content

test: verify fork-PR CI runs offline from prewarmed cache (#831) [DO NOT MERGE]#849

Open
vikrantpuppala wants to merge 3 commits into
databricks:mainfrom
vikrantpuppala:test/831-fork-ci-verify
Open

test: verify fork-PR CI runs offline from prewarmed cache (#831) [DO NOT MERGE]#849
vikrantpuppala wants to merge 3 commits into
databricks:mainfrom
vikrantpuppala:test/831-fork-ci-verify

Conversation

@vikrantpuppala

Copy link
Copy Markdown
Contributor

Verification PR — [DO NOT MERGE], safe to close/delete after checks report.

Fork PR from vikrantpuppala/databricks-sql-python with a single trivial comment edit. Confirms the fix in #845 + #848 end-to-end: fork PRs should now restore the forkvenv-* cache seeded on main and run unit / lint / type checks offline, instead of dying on the JFrog OIDC step (the original #831 symptom).

Expected: all Code Quality Checks (unit base/pyarrow/kernel, lint, types) go green via the cache-restore path with no JFrog access.

Refs #831

This pull request and its description were written by Isaac.

Verifies that a fork PR now runs unit/lint/type checks offline from the
prewarmed cache instead of dying on the JFrog OIDC step. This commit is a
no-op comment change; delete the branch after verification.

Signed-off-by: Vikrant Puppala <vikrant.puppala@databricks.com>
vikrantpuppala added a commit that referenced this pull request Jul 9, 2026
…831) (#850)

The warmer computed its cache key with hashFiles('**/poetry.lock') AFTER
setup-poetry ran `poetry lock`, which rewrites poetry.lock in place. The PR
jobs' restore-deps hashes the PRISTINE committed poetry.lock (it runs before
setup-poetry). The two hashes therefore never matched — a fork PR looked up
e.g. forkvenv-...-46e4852b... while the warmer had saved forkvenv-...-143cf261...
so every fork restore missed and fell back to the JFrog path (which forks
can't auth to), failing all jobs.

Capture the pristine lock hash in a step BEFORE setup-poetry and use that for
the cache key, matching what restore-deps computes. Found via the fork
verification PR #849 (all Unit Tests legs missed the cache).

Refs #831

Co-authored-by: Isaac

Signed-off-by: Vikrant Puppala <vikrant.puppala@databricks.com>
Empty commit to re-trigger checks after the warmer re-ran on main with
pristine-lock-hash keys (databricks#850). Expect unit/lint/type legs to restore the
cache and run offline now.

Signed-off-by: Vikrant Puppala <vikrant.puppala@databricks.com>
vikrantpuppala added a commit that referenced this pull request Jul 9, 2026
…#831) (#851)

Follow-ups from the fork verification run (#849) once the cache finally hit.
The offline restore mechanism works (10 legs green offline incl. all kernel
legs); these fix the remaining offline-path failures:

1. Python 3.9 can't be provisioned offline — the protected-runner image
   preinstalls 3.10-3.14 but not 3.9 (EOL), so setup-python tries to DOWNLOAD
   it and fails with no network on a fork. That hard-failed the 3.9 legs and,
   under fail-fast, cancelled the working 3.10-3.14 siblings. Move
   setup-python into restore-deps with continue-on-error, probe the restored
   interpreter, and export venv-usable; callers gate their offline run on it
   and skip 3.9 neutrally (3.9 fork coverage is in the merge queue) instead of
   failing. Keeps 3.9 in the matrix per maintainer preference without letting
   it sink the other legs.

2. check-types offline ran , which mypy rejects
   without --install-types ('--non-interactive is only supported with
   --install-types'). The warmer already primed the stubs into the venv, so
   run plain .

3. The warmer excluded 3.14-min but the PR matrix runs 3.14-min pyarrow (PR
   excludes only 3.12-min/3.13-min) — that leg missed the cache and fell back
   to JFrog. Align the warmer's excludes to the PR matrix exactly.

Refs #831

Co-authored-by: Isaac

Signed-off-by: Vikrant Puppala <vikrant.puppala@databricks.com>
…warmed) (databricks#831)

Signed-off-by: Vikrant Puppala <vikrant.puppala@databricks.com>
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