Repository navigation
Conversation
adriangb
force-pushed
the
configurable-prefetch
branch
from
September 24, 2025 17:58
3173a4c to
6900198
Compare
Currently we prefetch 1 file. In bandwith limited systems (e.g. object storage) having wider IO than CPU tends to make sense. This makes the prefetch configurable so that we can open >2 files at a time. My main concern is memory use. For now users will have to tune this manually. Long term I think we should track how much prefetched data is 'waiting' using a memory pool + the newly added memory tracking in arrow -> you can set a pretty large prefetch value and it will be throttled by memory. This means that scans with very selective filters (that produce few rows per file) will benefit heavily from wide IO prefetching while scans that produce large amounts of data will do little prefetching if they hit memory limits.
adriangb
force-pushed
the
configurable-prefetch
branch
from
October 4, 2025 20:58
67d9137 to
27fa525
Compare
adriangb
marked this pull request as ready for review
October 5, 2025 04:22
Contributor
Author
|
From my testing this works but doesn't seem to have much of a positive benefit (does not improve query performance). Maybe something is wrong with the implementation. But I think fundamentally the problem is that you want to go wide with IO but not really with other parts (e.g. CPU work applying filters in predicate pushdown), and this is a bit too high of a level for that. @alamb is working on changes to Parquet decoding which will allow prefetching at the byte range level, that sounds much more appropriate since we will be able to do IO-only prefetching. I will continue to do some investigations but I am going to close this for now to keep the PR backlog down. |
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
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.
Currently we prefetch 1 file. In bandwith limited systems (e.g. object storage) having wider IO than CPU tends to make sense. This makes the prefetch configurable so that we can open >2 files at a time. My main concern is memory use. For now users will have to tune this manually. Long term I think we should track how much prefetched data is 'waiting' using a memory pool + the newly added memory tracking in arrow -> you can set a pretty large prefetch value and it will be throttled by memory. This means that scans with very selective filters (that produce few rows per file) will benefit heavily from wide IO prefetching while scans that produce large amounts of data will do little prefetching if they hit memory limits.