Repository navigation
Conversation
v0.23.0 merges the loader's 100,000-row batches into stored objects of about 25 MB under a memory budget. ./load picks the pipeline depth, read cursors and budget from MemTotal, so the same recipe loads on 4 GB to metal machines. q28 averages octet_length, which matches ClickHouse's byte length(). The stateless tag is dropped so the load time is shown.
pmoust
had a problem deploying
to
benchmark-approval
October 9, 2026 02:16 — with 
GitHub Actions
Error
The real-S3 figures were from v0.16.1 with 4 MB objects. Replace them with a v0.23.0 stock pass on c6a.4xlarge against a same-region bucket, beside the RustFS result from the same release.
pmoust
requested a deployment
to
benchmark-approval
October 9, 2026 03:39 — with
GitHub Actions
Waiting
This branch is waiting to be deployed
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Pins the
ravelentry to Ravel v0.23.0 and sizes its load to the machine.What changes for this benchmark:
--target-bytes 1850000000,--max-flush-delay 30s), instead of one ~4 MB object per batch. That is about 410 objects instead of about 2,600. Above the stock fetch policy's ranged-read break-even, a cold scan reads only the columns a statement needs instead of whole objects, and that is most of the change in the cold column../loadpicks the pipeline depth, read cursors and that budget fromMemTotal(table in the README). Every value can be overridden from the environment.octet_length("URL")instead oflength("URL"). ClickHouse'slength()counts bytes and DataFusion's counts characters, so this is the faithful translation; q29 is unchanged.statelessis dropped fromtemplate.json, so the load time is shown../start. It passes no performance flags, the same as before.Measurement. Published v0.23.0 binaries, checked against the release's
SHA256SUMS, on a fresh c6a.4xlarge using this repository's cloud-init with this branch's entry:results/20261004)Other machines. The same load (which does not depend on the server's configuration) completed on c6a.large, c6a.xlarge, c6a.2xlarge, c8g.4xlarge, c6a.metal, c7a.metal-48xl and c8g.metal-48xl:
The c6a.large load completes where v0.21.0's failed. Those passes ran a tuned server configuration, so they are not stock query results and are not included here.
The README's figures for the object layout, the resolved pools and the tuned configuration are updated for v0.23.0.