Skip to content

Parallelize data_size computation in consolidate_shard_caches#4116

Merged
yonromai merged 1 commit intomainfrom
romain/parallel-data-size-consolidation
Mar 25, 2026
Merged

Parallelize data_size computation in consolidate_shard_caches#4116
yonromai merged 1 commit intomainfrom
romain/parallel-data-size-consolidation

Conversation

@yonromai
Copy link
Copy Markdown
Contributor

Summary

  • Parallelizes TreeStore.open + data_size reads across shards using ThreadPoolExecutor(max_workers=32) in consolidate_shard_caches
  • These reads are independent and dominated by remote-storage round-trip latency; parallelizing them removes the main bottleneck in the pre-copy phase
  • Cumulative offset computation remains serial (order-dependent)

Stacked on #4105. Refs #4100.

@yonromai yonromai added the agent-generated Created by automation/agent label Mar 25, 2026
@yonromai yonromai force-pushed the fix/consolidate-metadata-transaction branch from b13c63b to 3da360e Compare March 25, 2026 01:22
Base automatically changed from fix/consolidate-metadata-transaction to main March 25, 2026 13:49
@yonromai yonromai force-pushed the romain/parallel-data-size-consolidation branch from c56cd14 to 7b5633c Compare March 25, 2026 17:12
@yonromai yonromai marked this pull request as ready for review March 25, 2026 17:12
@yonromai yonromai force-pushed the romain/parallel-data-size-consolidation branch from ae461d5 to 118c38a Compare March 25, 2026 17:35
TreeStore.open + data_size reads are independent across shards and
dominated by remote-storage latency. Use a ThreadPoolExecutor to issue
them in parallel, then accumulate offsets serially.

Refs #4100

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@yonromai yonromai force-pushed the romain/parallel-data-size-consolidation branch from 118c38a to b075d72 Compare March 25, 2026 17:37
@yonromai yonromai requested review from dlwh and rjpower March 25, 2026 17:51
@yonromai yonromai merged commit fee874c into main Mar 25, 2026
52 of 54 checks passed
@yonromai yonromai deleted the romain/parallel-data-size-consolidation branch March 25, 2026 19:06
Helw150 pushed a commit that referenced this pull request Apr 8, 2026
## Summary

- Parallelizes `TreeStore.open` + `data_size` reads across shards using
`ThreadPoolExecutor(max_workers=32)` in `consolidate_shard_caches`
- These reads are independent and dominated by remote-storage round-trip
latency; parallelizing them removes the main bottleneck in the pre-copy
phase
- Cumulative offset computation remains serial (order-dependent)

Stacked on #4105. Refs #4100.

Co-authored-by: yoblin <268258002+yoblin@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

agent-generated Created by automation/agent

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants