CryoThrust commented on PR #12051:
URL: https://github.com/apache/seatunnel/pull/12051#issuecomment-6944684201

   Thanks for checking. I traced all callers on `022e216e6`:
   
   1. `stopCoordinatorSchedulers()` has one call site, inside 
`newMockCoordinatorService()`, before the fixture is returned. It is not called 
from test teardown or a `finally` block, so its assertion cannot replace a 
failure from the test body. All ownership-controlled fixtures now start with 
`isMasterNode() == false`; the other mock servers also return false by default. 
The assertion would still deliberately reject a fixture whose executor was 
unexpectedly shut down during construction.
   2. The constructor assigns `executorService` before starting either 
scheduler. `clearCoordinatorService()` shuts that executor down but does not 
null it, so an already-cleared instance reaches the shutdown assertion. The 
existing getter reports an absent/inaccessible field as `AssertionError("Failed 
to get coordinator executorService by reflection")`. It is not a general 
null-safe reflection utility: `ReflectionUtils.getField()` uses `Optional.of`, 
but the current coordinator lifecycle never assigns null to this field. No 
extra reflection or null guard was added in this revision.
   3. The ordering is enforced by the existing calls: the fixture factory calls 
`stopCoordinatorSchedulers()` before returning; `shutdownScheduler()` calls 
`shutdownNow()` and requires `awaitTermination(5, SECONDS)` to succeed. A 
timeout or interruption throws rather than returning a fixture. The test only 
enables master ownership or mutates `isActive` after the factory returns. The 
comment explains that ordering; the synchronous termination checks enforce it. 
Those checks already landed via #12050.
   
   The [fork Build for this exact 
head](https://github.com/CryoThrust/seatunnel/actions/runs/43884520828) 
completed with 84 successful jobs, 11 conditional skips and two failures. All 
four Linux/Windows JDK 11/17 unit jobs and both JDK 11/17 engine-v2 integration 
jobs passed. I downloaded both failed job logs: `all-connectors-it-4 (17)` 
could not download Maven Wrapper after repeated Central HTTP 429 responses and 
produced no test reports; `jdbc-connectors-it-part-2 (17)` stopped resolving 
the Milvus SDK POM with Central HTTP 429. These are dependency/bootstrap 
failures, not failed fixture assertions. This is fork evidence, not a green 
Apache-side Build.
   
   I requested the wrapper-failed job's retry, which was accepted. The second 
job's retry was rejected because the workflow is now running; I will retry that 
remaining failed job once the active attempt finishes. No source or assertion 
changes were made for these download failures.
   


-- 
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]

Reply via email to