xiangfu0 commented on PR #18103:
URL: https://github.com/apache/pinot/pull/18103#issuecomment-4227330415
Split the pure reproducer into a separate PR for clarity: #18166
The reproducer is in `TDigestMergeOrderReproducerTest` and only uses
t-digest APIs. It mirrors Pinot's pre-aggregated star-tree merge pattern:
- `1024` leaf digests
- `2` values per leaf digest
- batch size `16`
- deterministic seed `5`
- distribution: `97%` in `[0,100)`, `2.5%` in `[9900,9950)`, `0.5%` in
`[0,10000)`
To reproduce locally on the split branch:
```bash
./mvnw -pl pinot-segment-local
-Dtest=TDigestMergeOrderReproducerTest
-Dsurefire.failIfNoSpecifiedTests=false test
```
I also ran that test in 10 fresh Surefire JVMs and it passed `10/10`, so the
sample case is stable in CI-style execution.
The asserted behavior on the current 3.3 branch is:
- compression `100/150/200`: large sequential-vs-hierarchical merge
divergence
- compression `500`: stable again
That should make it easier to discuss the sample case independently from the
production compression change in #18103.
--
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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
