Make DelayableWriteable compress its contents #126988
Merged
Conversation
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
original-brownbear
added
>non-issue
:Analytics/Aggregations
elasticsearchmachine
added
the
Team:Analytics
Collaborator
|
Pinging @elastic/es-analytical-engine (Team:Analytics) |
rjernst
reviewed
Apr 19, 2025
Member
rjernst
left a comment
rjernst
left a comment
There was a problem hiding this comment.
The concepts here seem good (using fixed-size sizes and compressing). I have one question.
| out.writeInt(serialized.length()); | ||
| serialized.writeTo(out); | ||
| } else { | ||
| out.writeBytesReference(serialized); |
Member
There was a problem hiding this comment.
It seems the above comment is no longer accurate with the new branch since we will no longer copy the raw bytes, but instead reencode?
Contributor
Author
There was a problem hiding this comment.
We still copy the raw bytes, it's just that the prefix is a full int now.
writeBytesReference is nothing else but:
out.writeVInt(serialized.length());
serialized.writeTo(out);all that happens here is going from a vint to an int?
Contributor
Author
|
Thanks Ryan! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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Outdated suggestions cannot be applied.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.
In light of data from recent escalations and the introduction of batched execution we can make two improvements to this logic.
For one, we should prefix with a fixed length length field so that we don't need to do any copying when serializing to account for the vint. This outright halves the memory bandwidth required relative to the previous implementation.
More importantly maybe, we should compress these bytes. The wire-format for aggregations is rather inefficient when working with nested bucket aggregations since the type strings are repeated over and over. These don't contribute to the peak heap requirements because they are translated into Java types, but blow up the message size considerably (among other things). Practically, it seems that we often get compression ratios of ~10x for aggregations.
Given that we generally have more memory issues than CPU issues during the reduce-step it seems like an easy tradeoff to trade a little CPU for compression for serious heap savings here.