docs(train): drop --dist loadfile from the shallow suite's README - #6215
docs(train): drop --dist loadfile from the shallow suite's README#6215jam-jee wants to merge 1 commit into
Conversation
The flag was removed from createCIShallowIntegBuildSpec in SageMakerMLFPySDKInfraCDK: it pins one file's tests to one xdist worker, and each test here holds a concurrency slot until its training job reaches a terminal state (~75s), so the 17-test RLVR file alone took 19m45s against the project's 30-minute timeout. Job names are unique per invocation rather than per test function (see unique_name in harness.py), so tests are free to spread across workers, which is what the wall-clock estimate in harness.py assumes. Documentation only -- the invocation itself lives in the CDK package. --- X-AI-Prompt: The fast-integ-tests CodeBuild job is still failing on PRs now that the CDK project is deployed -- diagnose and fix it X-AI-Tool: claude-code
|
Claude finished @jam-jee's task in 38s —— View job PR Review: docs(train): drop
|
Codecov Report✅ All modified and coverable lines are covered by tests. ❌ Your project check has failed because the head coverage (43.72%) is below the target coverage (65.00%). You can increase the head coverage or adjust the target coverage.
Additional details and impacted files@@ Coverage Diff @@
## master #6215 +/- ##
===========================================
- Coverage 70.79% 43.72% -27.07%
===========================================
Files 553 387 -166
Lines 67947 52399 -15548
===========================================
- Hits 48101 22910 -25191
- Misses 19846 29489 +9643 🚀 New features to boost your workflow:
|
Issue
The shallow suite's README documents
--dist loadfileas part of how the suite isinvoked. That flag has been removed from the CodeBuild invocation, so the README now
names a flag that no longer exists.
Description of changes
Drops
--dist loadfilefrom the list of pytest options the README says lives increateCIShallowIntegBuildSpec(in the internalSageMakerMLFPySDKInfraCDKpackage).Why the flag went away: it pins one file's tests to a single xdist worker, and each
test in this suite holds a concurrency slot until its training job reaches a terminal
state (~75s). That serialized the largest file — the 17-test RLVR file alone took
19m45s, against the CodeBuild project's 30-minute timeout, with the rest of the suite
still to run.
The flag was never needed. Job names are unique per invocation rather than per test
function (see
unique_nameinharness.py), precisely so tests can spread acrossworkers, and the wall-clock estimate in
harness.py—(#jobs * drain) / cap, ~8-12min at 84 jobs — assumes they do.
Documentation only. The invocation itself lives in the CDK package, so the behaviour
change ships there, not here.
Testing done
Ran the full shallow suite under the new invocation (Python 3.10,
-n 8, no--dist loadfile) against us-west-2 in the SDK test account: 83 passed, 1 skippedin 6m50s. The single skip is structural —
RLAIFTrainertakes no compute argument,so
test_explicit_compute_is_acceptedskips itself.No code changed in this PR, so there is nothing else to test.
By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.