Conversation
Two low-risk cuts targeting the dominant costs in PR test wall-clock: 1. Exclude `job_pydabs_1000_tasks` from the `migrate` and `continue_293` invariant tests. The 1000-task scale case is still exercised by `no_drift`; running the same scale in two additional variants added ~3 minutes per linux/direct test job without incremental coverage. 2. Add `install-pythons-test` Makefile target that installs only the Python version actually needed by unit + acceptance tests (3.10), and switch the setup-build-environment composite action to use it. The `patchwheel` unit test now skips versions not available via `uv python find`, so CI can install a subset without failing. Integration runs keep using `install-pythons` for the full 3.9-3.13 matrix. Expected impact: linux/direct ~10min -> ~3-4min, windows/* ~3min off setup. Co-authored-by: Isaac
denik
reviewed
Apr 22, 2026
| // Skip if the interpreter is not available via uv. CI only installs | ||
| // a subset of versions to keep setup fast; the full matrix is | ||
| // exercised in integration and scheduled runs. | ||
| if err := exec.Command("uv", "python", "find", py).Run(); err != nil { |
Contributor
There was a problem hiding this comment.
It's not great to dynamically skip like this - we might be running less than we think we do because setup got broken and nobody noticed.
Better to add explicit env var, TEST_PYTHON_VERSIONS=3.10,3.11 that is initialized differently in different contexts.
|
|
||
| # The 1000-task scale case is covered by no_drift. Running it here adds ~1.5 min | ||
| # per variant (deploy + migrate + plan at 1000 tasks) without incremental coverage. | ||
| EnvMatrixExclude.no_pydabs_1000_tasks = ["INPUT_CONFIG=job_pydabs_1000_tasks.yml.tmpl"] |
Contributor
There was a problem hiding this comment.
How about dropping them from local tests but keep them on cloud (which is async)?
It's important that migrations works for all edge cases.
The test server (libs/testserver/jobs.go:531) maps a task's spark_version to a Python version (DBR 12 -> 3.9, 13 -> 3.10, 15 -> 3.11, >=16 -> 3.12) and calls `uv venv --python <ver>` at runtime. Acceptance tests using DBR 12.2 (integration_whl/wrapper*) or EnvMatrix on UV_PYTHON (templates/ default-python/integration_classic) therefore require those versions to be installed. Measured on the prior CI run: installing only 3.10 took ~6.7s on Windows, vs an estimated 30-60s for all five. Not worth breaking tests. The real Windows setup cost is Setup Go (2m49s cache restore), which is a different problem. Keeping the pydabs_1000_tasks invariant exclusion from the prior commit; linux/direct is now ~4m18s including reruns, under the <5 min target. Co-authored-by: Isaac
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Two low-risk cuts targeting the dominant costs in PR test wall-clock.
1. Drop
job_pydabs_1000_tasksfrommigrate+continue_293invariant testsThe 1000-task scale case is still exercised by
no_drift; running the same scale in two additional variants added ~3 min per linux/direct job without incremental coverage. Seeacceptance/bundle/invariant/{migrate,continue_293}/test.toml.2. Slim
install-pythonsin the CI setup-build-environment actioninstall-pythons-testMakefile target installs only Python 3.10 (the default for acceptance bundle tests; 3.13 already comes fromactions/setup-python).libs/patchwheel/patch_test.gonow skips versions not available viauv python find, so CI can install a subset without failing.make integrationstill pulls in the fullinstall-pythons(3.9–3.13) — no change to integration runs.Expected impact
Test plan
make testpasses locallypatchwheelunit test skips missing versions cleanlyno_drift/pydabs_1000_tasksstill running;migrate/pydabs_1000_tasks+continue_293/pydabs_1000_tasksno longer in the matrixNot included in this PR
TF_PLUGIN_CACHE_DIR). The repo already uses afilesystem_mirrorfor the databricks provider, so the easy win of avoiding registry downloads is in. Deeper wins (GHA Dev Drive for.terraform/small-file I/O, plugin cache hardlinks on NTFS) are a separate PR — research says the C: drive onwindows-*-largerunners is ~4k IOPS vs ~127k on a Dev Drive, and this is the real Windows gap.This pull request and its description were written by Isaac.