mikemccand commented on issue #15662:
URL: https://github.com/apache/lucene/issues/15662#issuecomment-3864494191

   OK the last ad-hoc run finished and it is STILL SLOW @uschindler!  See 
annotation `JR` across all the charts (e.g. 
[`AndHighHigh`](https://benchmarks.mikemccandless.com/AndHighHigh.html)).
   
   This means the smoking gun is something in the whole-system update across 
many packages (but NOT the kernel!), sheeeeesh.
   
   **And I think this means we shouldn't hold Lucene 10.4.0 release?**
   
   I'll continue trying to isolate the cause.  I could start binary searching 
those many updates, maybe... but it is a complex dep graph.  Hmm maybe someone 
has made a binary search algo that's dep-graph aware!  Fun problem to set 
agentic Opus 4.6 on!  Wait, no, there's a simple way:  Apparently the arch 
package mirrors expose timestamp'd point-in-time snapshots (e.g. 
https://archive.archlinux.org/repos/2026/01/15), so I could just walk backwards 
in time, binary searching.  Maybe at least I will first roll back to my prior 
whole system update and confirm I can reproduce Lucene running fast (normal)...
   
   Java want from `"25.0.1" 2025-09-16` -> `"25.0.2" 2026-01-20`, but it seems 
so unlikely a bugfix release could cause this massive slowdown?  Though that 
[ParallelGC bug / interaction with 
COH](https://bugs.openjdk.org/browse/JDK-8375467) also worries me ... does it 
for sure only manifest with COH enabled... maybe I should separately test G1GC 
for nightly.
   
   Weirdly/confusingly, [`Fuzzy1` (edit-distance=1) actually saw some recovery 
with the LTS kernel 
downgrade](https://benchmarks.mikemccandless.com/Fuzzy1.html), and so did 
[`Fuzzy2`](https://benchmarks.mikemccandless.com/Fuzzy2.html).


-- 
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]

Reply via email to