DanielLeens commented on PR #12261:
URL: https://github.com/apache/seatunnel/pull/12261#issuecomment-5714489216

   Status correction from the author: this PR does not fix the flake it was 
opened for, so please do not merge it as a stabilization.
   
   Its own CI (fork run `34974425990`, both 
`updated-modules-integration-test-part-1` legs) still fails 
`PostgresCDCIT#testPostgresCdcSnapshotOnlyAndCommittedOffsetStartupModes`, and 
at a different place than the one this PR changes:
   
   ```
   expected: <1> but was: <0> within 3 minutes.
        at 
PostgresCDCIT.testPostgresCdcSnapshotOnlyAndCommittedOffsetStartupModes(PostgresCDCIT.java:629)
   ```
   
   That is the final phase: after `savepoint` + `restore` of the 
committed-offset job, the row inserted once the restored slot is active (`id = 
15`) never reaches the sink. The seed-phase LSN wait this PR reworks has 
already passed by then. The same failure is on `dev` itself 
(`all-connectors-it-7 (11)` in push runs `34995029902` and `35063682504`) and 
on unrelated PRs (#12364, #12360), always at ~260 s, i.e. that 3 minute await.
   
   What the restored job's log shows in the failing run: the split restores 
with `lsn=LSN{0/227D970}, lastCommitLsn=LSN{0/227D970}`; the restored and 
freshly discovered row types are identical (no `RestoreTableSchemaEvent`); 
logical decoding on the slot starts twice 0.5 s apart (Debezium's WAL position 
search, then the real stream); checkpoints 5, 6, 7, ... keep completing every 5 
s; `Committing offset` advances once, from `0/227D970` to `0/227FCC0` at 
checkpoint 6, and never again. So the source consumed WAL past the insert, but 
no row was written. Whether the event is dropped by the WAL resume position 
filter or lost further down is not visible, because the Zeta e2e container does 
not print `io.debezium` INFO logs.
   
   I am keeping this PR open only as the place to continue that investigation; 
the current diff should be treated as insufficient.
   


-- 
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