Optimize sort settings default providers#138711
Conversation
|
Benchmark results running the elastic/logs many-shards-quantitative benchmark in esbench. Baseline ref: 22a7491 Results: |
…ngs-default-providers-2
|
Ok, I ran another benchmark with the optimizations from this PR cherry-picked onto the first commit that introduced the regression. This eliminates any slowdowns due to other changes since the original regression was introduced. Baseline ref: 22a7491 |
SummaryThis PR fixes a regression in the initial-indices step of the elastic/logs many-shards-quantitative benchmark.
Since the performance returned to the baseline throughput (within noise) when the fix was cherry-picked on top of the original regression, I believe this fix completely resolves the original regression. |
|
Pinging @elastic/es-storage-engine (Team:StorageEngine) |
martijnvg
left a comment
There was a problem hiding this comment.
LGTM 👍
Thanks for presenting the results in such a clear way.
However, there still remains some slowdown due to other changes since that initial regression that likely should be investigated separately.
Did flame graphs suggest what the other slowdown(s) could be?
I'm not certain, but possibly something security related? I'll post more details on the original investigation. |
This PR optimizes the default providers for the sort settings
index.sort.field,index.sort.order,index.sort.mode, andindex.sort.missingto fix a performance regression in the many-shards-quantitative benchmark associated with the introduction of these default providers in #135886.