unbridled-41 commented on PR #2833:
URL: 
https://github.com/apache/rocketmq-dashboard/pull/2833#issuecomment-5467545956

   Full-suite control experiment for the one intermittent failure mentioned 
above 
(`OpenAiCompatibleLlmGatewayTest.successfulAndFailedStreamsEmitOneTerminalSequence`):
   
   | full `mvn -B test` run | code under test | result |
   |---|---|---|
   | 1 | this PR (v1) | 1 failure — this test |
   | 2 | this PR (v2) | 1 failure — this test |
   | 3 | this PR (v2) | 1 failure — this test |
   | control 1 | unmodified branch | green |
   | control 2 | unmodified branch | **1 failure — same test, same assertion** |
   
   So the failure is a pre-existing, load-sensitive intra-test race, unrelated 
to this change: the test runs two sequential chats on a `ThreadPoolExecutor(0, 
1, SynchronousQueue, AbortPolicy)` and awaits only the first stream's 
terminal-event latch before submitting the second chat; if the first worker 
thread has not returned to the pool yet, the second submission is rejected and 
surfaces as `llm.gateway.overloaded` instead of `llm.provider.failed`. It 
reproduces on the unmodified branch as well (control 2), and by surefire class 
order the AI tests execute before the message-provider classes touched by this 
PR are loaded.
   


-- 
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