SEPURI-SAI-KRISHNA commented on PR #12290:
URL: https://github.com/apache/seatunnel/pull/12290#issuecomment-5924452409
Correcting my previous comment. The rerun on `4d64a4d8` has finished and it
is red again, but with **three different failures**, none of them the two I
named earlier. `engine-v2-it` passed this time on both JDK legs, so #12311 and
#12313 are not what is holding this PR now.
| leg | failure | status |
| --- | --- | --- |
| `transform-v2-it-part-1 (11)` |
`TestFilterRowKindIT.testFilterRowKindMultiTable` | #12116 |
| `iceberg-connector-it (8)` | `DependencyResolutionException`, no test ran
| infrastructure |
| `jdbc-connectors-it-part-7 (11)` | `MetalakeIT.testMetalake`, fails in
`startUp` | not currently tracked |
Run `36725638109`: 80 jobs successful, 3 failed, 11 skipped. `engine-v2-it`
is green on both JDK 8 and JDK 11, which is what rules out the two I named
before.
**`transform-v2-it-part-1`** is the one that deserves scrutiny here, because
the failing test has "MultiTable" in the name and this PR touches multi-table
sink routing. It is not this PR, and the reason is that the change cannot alter
routing at all:
```java
- index = (object.hashCode() & Integer.MAX_VALUE) % blockingQueues.size();
+ index = HashUtils.bucketIndex(object.hashCode(), blockingQueues.size());
```
`HashUtils.bucketIndex(int, int)` is `checkBucketCount(bucketCount); return
(hash & Integer.MAX_VALUE) % bucketCount;`. For any bucket count of 1 or more
the two expressions are the same arithmetic, so every row routes to the same
queue before and after. The only behavioural addition is the guard rejecting a
bucket count of zero or less, which a running job cannot reach.
The failure itself is `expected: <0> but was: <1>` at
`TestFilterRowKindIT.java:53`, an exit-code assertion on a non-Zeta container
leg. #12116 records that exact signature for that exact test, and attributes it
to `AssertSinkWriter` using static JVM-wide counters evaluated per subtask
`close()`, on Flink legs only. Suite totals were 219 tests, 1 failure, 0 errors.
**`iceberg-connector-it`** never reached a test. Maven failed dependency
resolution, so there is nothing in it to attribute.
**`MetalakeIT`** fails during fixture setup rather than in the test body:
```
org.awaitility.core.ConditionTimeoutException: ... expected: <0> but was:
<7> within 3 minutes
at MetalakeIT.startUp(MetalakeIT.java:158)
```
I could not find an open issue for it. I am deliberately not filing one on a
single observation, since one occurrence is an anecdote; if it shows up again I
will sample the nightly runs properly and file it with evidence.
So the position is unchanged in substance: this PR's diff is one file,
`+5/-7`, provably routing-neutral, and nothing red on it is attributable to it.
What has changed is which flakes happened to land, which is itself the point.
Three reruns have now produced three different sets of unrelated failures.
@nzw921rx @davidzollo I would rather not keep rerunning and burning CI on
this. If you are satisfied the above is unrelated, this is ready from my side;
if you would prefer to wait for #12116 and the rest to clear, say so and I will
park it until then.
--
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]
