perf(dwio): Reuse context in ZSTD_decompress #15854
Open
+4
−4
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
Summary: ## History Some time ago similar PR was landed (facebookincubator#6699, D49501856) and caused SEV S369242 in Meta. That time `ZSTD_DCtx` context was created and reused on decompressor level. The optimization was reverted due to OOMs. ## Getting back to it again The optimization still makes sense. For example: - In Presto Adhoc we spend 0.07% of CPU cycles in `ZSTD_createDCtx_internal`: https://fburl.com/strobelight/2lt1wn4v - In Presto batch we spend 0.1% of CPU cycles in `ZSTD_createDCtx_internal`: https://fburl.com/strobelight/js4kn8za ## The fix Instead of creating `ZSTD_DCtx` per decompressor, we should create it per thread. Then we will be able to reuse the allocation and don't consume so much memory in FlatMaps. Differential Revision: D89716393
✅ Deploy Preview for meta-velox canceled.
|
meta-cla
bot
added
the
CLA Signed
|
@kletkavrubashku has exported this pull request. If you are a Meta employee, you can view the originating Diff in D89716393. |
kletkavrubashku
changed the title
Yuhta
approved these changes
Dec 24, 2025
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Outdated suggestions cannot be applied.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.
Summary:
History
Some time ago similar PR was landed (#6699, D49501856) and caused SEV S369242 in Meta. That time
ZSTD_DCtxcontext was created and reused on decompressor level. The optimization was reverted due to OOMs.Getting back to it again
The optimization still makes sense. For example:
ZSTD_createDCtx_internal: https://fburl.com/strobelight/2lt1wn4vZSTD_createDCtx_internal: https://fburl.com/strobelight/js4kn8zaThe fix
Instead of creating
ZSTD_DCtxper decompressor, we should create it per thread. Then we will be able to reuse the allocation and don't consume so much memory in FlatMaps.Test plan
I did multiple experiments. The biggest one runs 100k shadow queries on the same cluster with 2 packages: https://fburl.com/scuba/presto_queries/ddi8pxmd
Same upstream revision for prod and test builds. Only this diff on top.
Here is what I got:
https://pxl.cl/8Dsk1
Execution time is 9% better. The data is not probably really accurate, but at least it's better. I appreciate suggestions here.
The memory usage didn't grow: https://pxl.cl/8Dsjn
Differential Revision: D89716393