SEZ9 commented on issue #12058:
URL: https://github.com/apache/seatunnel/issues/12058#issuecomment-5594885645

   @Rangsh thanks for the baseline reproduction — running it against 
`origin/dev` @ `8bea8c681` rather than a candidate change is the right way to 
characterize the original symptom.
   
   What your run establishes so far: on a single M1 / Corretto 11.0.26 host 
with the existing JMH defaults (`SingleShotTime`, 3 forks, 3 warmup / 5 
measurement iterations), `checkpointOverviewIncrementalUpdate` shows 190.087 ± 
65.997 us/op, CV ≈ 32.5%, range 128.4–393.3 us/op, with the two outliers (240.3 
in fork 0, 393.3 in fork 2) sitting inside otherwise fairly tight forks. So the 
dispersion reproduces outside the Actions runner. It does not yet tell us which 
mechanism produces the spikes, and I agree with DanielLeens that this should 
not be read as validating any fix.
   
   Your comment appears to be cut off after "Within-fork spikes ex" — could you 
repost the rest of the Observations section?
   
   To close out this baseline step:
   
   1. Attach the `results.json` from `-rf json -rff results.json` and the `-v 
EXTRA` console output for all three forks, so the raw iteration samples can be 
inspected.
   2. Add GC logging to the fork JVM args and attach the per-fork GC logs, so 
we can see whether the 240.3 / 393.3 iterations line up with collections.
   3. Run the same command again, unchanged, on the same machine (A/A) and 
report whether the outlier count and CV are stable across the two runs.
   4. Please also try the diagnostics tool nzw921rx pointed to above and attach 
its output alongside the raw JMH artifacts.
   
   Once those are in, we can agree on one narrowly scoped test to isolate the 
mechanism (JIT, GC, `file:///` MapStore writes, or scheduling) before anyone 
proposes a change.
   
   <!-- streview-comment:909 -->


-- 
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