bugfix: prevent background merges on the realtime lucene index#13050
Merged
Jackie-Jiang merged 1 commit intoapache:masterfrom May 3, 2024
Merged
bugfix: prevent background merges on the realtime lucene index#13050Jackie-Jiang merged 1 commit intoapache:masterfrom
Jackie-Jiang merged 1 commit intoapache:masterfrom
Conversation
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## master #13050 +/- ##
============================================
+ Coverage 61.75% 62.17% +0.41%
+ Complexity 207 198 -9
============================================
Files 2436 2502 +66
Lines 133233 136658 +3425
Branches 20636 21179 +543
============================================
+ Hits 82274 84961 +2687
- Misses 44911 45397 +486
- Partials 6048 6300 +252
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. |
Contributor
|
I saw this test being flaky. Do you think it is related? |
Contributor
Author
|
I don't think it's related, but is probably caused by this behavior: https://julien.ponge.org/blog/not-all-java-single-threaded-executors-are-created-equal-a-java-finalizer-horror-story/ Did you see the flakiness is Java 21? It should be fixed there, but could misbehave in the suites run with 11/17. Do you mind if I put the fix into another PR since it's not related to the fix for the reuse feature, or want me to bundle it into this one? (there's a couple other nits I'd rather bundle together if no preference). |
Contributor
|
A general question: if we turn off merge, will it result in too many Lucene segments for realtime part and cause query perf issues? |
Jackie-Jiang
approved these changes
May 3, 2024
Contributor
Author
In theory it could, but in practice Pinot's typical segment sizes prevent this (>5-10GB segments isn't normal), and we don't see a query perf regression with even a small 10MB buffer size and 1.5GB segment size. FDs are probably more of a concern for systems w/ low limits, but I think can be addressed if needed w/ lucene buffer size/compound file format/Pinot segment size tuning - all of which have existing configs. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
#12744 changed the Lucene index build process to commit the realtime Lucene index, and reuse it in the immutable segment.
.commit()is called at the beginning of segment conversion, and the Lucene index directory is copied to another location. SinceIndexWritercan merge segments in the background even without insertions/updates to the index, these merges non-deterministically can trigger aFileNotFoundException:By using
NoMergeScheduler, the async merges will never occur. This patch was tested in our staging/prod env and the intermittent segment build failures disappeared.tag:
bugfix