fix(ci): enforce post_training isolation in test coordinator and fix missing test markers across test files

[RexBearIU]

Description

This PR fixes leaky test marker expressions in the CI test coordinator and adds missing Pytest markers across 6 test files in MaxText, ensuring pre-training and post-training test suites are strictly isolated and do not accidentally cross-trigger.

Why this change is being made & problem being solved

In .github/workflows/run_tests_coordinator.yml, tpu-post-training-integration and tpu7x-post-training-integration previously checked 'not cpu_only and not gpu_only and integration_test'. Because and post_training was omitted from the filter expression, pre-training integration tests without a post_training mark were leaking into the post-training integration test suite. Furthermore, several test files under tests/post_training/ and tests/integration/ lacked explicit directory/file-level marker declarations (pytestmark = [pytest.mark.post_training], pytestmark = pytest.mark.integration_test), leading to incomplete test suite tagging.

Specific implementation details

1. CI Coordinator Filter Isolation: Added and post_training to tpu-post-training-integration and tpu7x-post-training-integration marker expressions in .github/workflows/run_tests_coordinator.yml.
2. Post-Training Unit Test Tagging: Added pytestmark = pytest.mark.post_training to 3 post-training unit tests:
   • tests/post_training/unit/learn_to_init_test.py
   • tests/post_training/unit/vllm_decode_test.py
   • tests/post_training/unit/vllm_rollout_unroll_test.py
3. Integration Test Tagging: Added pytest.mark.integration_test to 3 integration test files:
   • tests/post_training/integration/grpo_correctness.py
   • tests/integration/check_vma_test.py
   • tests/integration/aot_identical_test.py
4. Engine Isolation Safeguard: Confirmed Native LoRA (use_sft=True in lora_e2e_nnx_test.py) remains marked not post_training as it operates on the pre-training engine.
5. Strict Marker Compliance: Preserved default unit test inheritance (not integration_test and not post_training) for tests/unit/ without introducing unregistered marker tags that would violate --strict-markers in pytest.ini.

Shortcomings & future improvements

In future work, we can consider splitting tpu-tests in ci_pipeline.yml into separate matrix jobs for pre-training and post-training suites to enable direct domain-level job gating.

Tests

• Pre-commit Validation: Passed cleanly on all modified files:

  pre-commit run --files .github/workflows/run_tests_coordinator.yml \
    tests/post_training/unit/learn_to_init_test.py \
    tests/post_training/unit/vllm_decode_test.py \
    tests/post_training/unit/vllm_rollout_unroll_test.py \
    tests/post_training/integration/grpo_correctness.py \
    tests/integration/check_vma_test.py \
    tests/integration/aot_identical_test.py

• CI Test Suite Runs: All required CI workflows completed successfully on GitHub Actions (36/37 checks passing; All Required Tests Passed = Success):
  • cpu-unit tests: PASSED
  • gpu-unit tests: PASSED
  • tpu-unit tests: PASSED
  • cpu-post-training-unit tests: PASSED
  • tpu-post-training-unit tests: PASSED
  • gpu-integration tests: PASSED
  • tpu-integration tests: PASSED
  • maxtext_tpu_pathways_unit_tests: PASSED
  • maxtext_tpu_pathways_integration_tests: PASSED
• Strict Marker Audit: Verified --strict-markers compliance against pytest.ini.

Checklist

Before submitting this PR, please make sure (put X in square brackets):

• I have performed a self-review of my code. For an optional AI review, add the gemini-review label.
• I have necessary comments in my code, particularly in hard-to-understand areas.
• I have run end-to-end tests tests and provided workload links above if applicable.
• I have made or will make corresponding changes to the doc if needed, including adding new documentation pages to the relevant Table of Contents (toctree directive) as explained in our documentation.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!