DanielLeens commented on PR #13177:
URL: https://github.com/apache/seatunnel/pull/13177#issuecomment-6783265661

   Hi @goutamadwant, a short follow-up to my last review of `827a5083`. The 
fork `Build` run on this head has finished 
(https://github.com/goutamadwant/seatunnel/actions/runs/42765841447), and one 
failure comes from a test added in this PR, so my earlier "no code change is 
needed" no longer holds.
   
   **Related to this PR**
   - `unit-test (17, windows-latest)`: 
`EmbeddingModelDimensionTest.testHttpProvidersPropagateRetryAfterAndStopAtCap{String}[1]`
 (the `OPENAI` case) failed at line 99 with `expected: <1> but was: <2>`. The 
`Retry-After` and `429` assertions on lines 97-98 passed, so the final response 
was the expected 429, but the mock server saw two requests. The same test 
passed on `unit-test (11, windows-latest)` in this run and on both Windows jobs 
in the previous run, and this head's Java code is unchanged since then, so it 
is intermittent.
   - My reading (an inference from the code, not reproduced): the `/embeddings` 
handler in this test never reads the POST body before `exchange.close()`. On 
Windows, closing a socket with unread input can reset the connection before the 
client reads the 429, so the first attempt fails at the transport layer and is 
sent again. Your sibling test `testOpenAIWaitsForRetryAfterCooldown` already 
drains the body for this reason ("Consume the request before reusing the HTTP 
connection").
   - Minimum fix: drain `exchange.getRequestBody()` at the start of that 
handler, the same way the sibling test does. This is a test-only change; the 
production retry logic does not need to change.
   
   **Not related to this PR**
   - `connector-file-sftp-it` (11/17): the removed `atmoz/sftp:alpine-3.7` 
image, fixed on `dev` by #13203.
   - `engine-v2-it (11)`: `SystemLoadAllocateStrategyIT.testSystemLoadStrategy` 
timed out in the engine slot-allocation test. This PR does not touch the engine.
   - `transform-v2-it-part-2 (11)`: cancelled at the job time limit. Your E2E 
(`TestEmbeddingIT`) runs in part 1, not part 2.
   
   Suggested next step: push the test fix together with a merge of the latest 
`dev` (which includes #13203). One fresh run then covers both. If 
`engine-v2-it` or `transform-v2-it-part-2` fail again for the same reasons, 
re-running only those jobs should be enough.
   
   My source-review conclusion is otherwise unchanged; I will approve once 
`Build` is green on the new head.


-- 
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