avoids unneeded meta index read#6024
Merged
keith-turner merged 1 commit intoapache:2.1from Dec 18, 2025
Merged
Conversation
CachableBlockFile.Reader.getBCFile() always read the RFile root block from the index cache even if would not use it. Modified the method to only read from the index cache when the data is actually needed. Noticed this while working on apache#6010, saw for every data block read from a rfile it would also read from the index cache. Adjusted ScanTracingIT because before this change there was an rfile index cache read per rfile data block read. After this change the number of rfile index reads align w/ the number of rfiles opened, which in this case aligns with the number of tablets in the test tables. The existing behavior was likely not a performance problem if the data was in cache, but could potentially cause thread contention and extra CPU usage. So its probably good to fix, also nice to make the tracing stats align better w/ expectations.
kevinrr888
approved these changes
Dec 17, 2025
Member
kevinrr888
left a comment
kevinrr888
left a comment
There was a problem hiding this comment.
LGTM
I don't follow the test changes but seeing as you recently wrote the test, I don't have any problems
Contributor
Author
Yeah those test are hard to follow and not sure how to improve it. They were created from manually inspecting the new stats to see if they made sense and then creating some loose bounds that matched the stats seen. |
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
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
CachableBlockFile.Reader.getBCFile() always read the RFile root block from the index cache even if would not use it. Modified the method to only read from the index cache when the data is actually needed. Noticed this while working on #6010, saw for every data block read from a rfile it would also read from the index cache.
Adjusted ScanTracingIT because before this change there was an rfile index cache read per rfile data block read. After this change the number of rfile index reads align w/ the number of rfiles opened, which in this case aligns with the number of tablets in the test tables.
The existing behavior was likely not a performance problem if the data was in cache, but could potentially cause thread contention and extra CPU usage. So its probably good to fix, also nice to make the tracing stats align better w/ expectations.