wenjin272 opened a new pull request, #1343: URL: https://github.com/apache/flink-agents/pull/1343
Linked issue: N/A (diagnostic-only draft). ### Purpose of change Python CI can print passing summaries for both test groups and still finish with exit code 2. This draft records the original command exit codes before `tools/ut.sh` normalizes them, and runs three independent Flink 1.20 Python integration jobs per CI run to collect evidence in the upstream runner environment. Example failure: https://github.com/apache/flink-agents/actions/runs/40848545591/job/123361969642 Both commands run as before. After the second command returns, the script prints the Flink version, E2E `rc1`, and integration `rc2`, before the existing exit-5 handling and logical-OR aggregation. Logging after both commands avoids inserting output between the two test invocations. The uv and pip paths both report these values. ### Behavioral Semantics - Original command statuses are logged even when either group fails; `rc2=5` is logged before normalization. - Both zero statuses still produce script exit 0; any nonzero status still produces script exit 2. - Flink 1.20 gets three separately named jobs with independent runner setup. Existing `fail-fast: false` keeps other samples running after a failure. Other Flink versions retain one job each. - Test commands, model setup, runner selection, and Python shutdown behavior are unchanged. This draft collects evidence; it does not suppress failures or claim a fix. ### Tests | Contract | Validation | | --- | --- | | Raw values remain visible and final status is unchanged | 24 temporary full-script checks using stub commands across uv/pip, including 0, 1, 2, 5, 120, 134, and 139 | | Three independent 1.20 samples; other versions run once | Parsed workflow matrix and checked distinct entries and `fail-fast: false` | | Existing script behavior | `bash tools/test/run.sh` completed successfully (321 cases reported, with environment-dependent skips); `/bin/bash -n tools/ut.sh`; `git diff --check` | Not verified locally: actual Flink E2E execution or reproduction of the post-test failure. The upstream draft CI run is intended to supply that evidence. Actionlint could not be run because its release asset download timed out; YAML parsing and explicit matrix checks passed. ### API No public API changes. The script adds diagnostic output; exit-code semantics are unchanged. ### Documentation - [ ] `doc-needed` - [x] `doc-not-needed` - [ ] `doc-included` ### Was this patch authored or co-authored using generative AI tooling? - [x] Yes - [ ] No Generated-by: Codex CLI 0.153.4 (GPT-6) -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
