Repository navigation
fix(arrow): decode data files on the CPU runtime - #3588
Draft
alexanderbianchi wants to merge 2 commits into
Draft
alexanderbianchi wants to merge 2 commits into
alexanderbianchi wants to merge 2 commits into
Conversation
This was referenced Oct 30, 2026
This branch has not been 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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Which issue does this PR close?
Related to #3276 and #3277. No issue is closed by this draft.
Depends on #3587. This branch currently includes that prerequisite commit. The data-file change is the second commit; review the incremental diff here. Rebase onto main after #3587 merges.
What changes are included in this PR?
Data-file footer and page-index decoding currently run on I/O workers, while page decoding runs on whichever runtime polls the scan. Move the complete file-open/planning phase and Parquet stream consumption onto the configured CPU runtime. Storage opens and byte-range reads continue to use the I/O adapter from #3587.
Keep opening/planning in spawned, abort-on-drop tasks so buffered opens and bloom-filter fetches continue progressing when the consumer polls another file. Run bloom-filter evaluation directly inside that CPU task.
Deliver decoded batches through a bounded channel per file. Reserve capacity before polling Parquet, allowing at most one output batch ahead of the consumer. Retain an abort-on-drop producer handle with the returned stream so dropping a scan cancels outstanding work. Await the producer at EOF to propagate decoding failures, panics, and runtime shutdown instead of silently truncating results.
Existing file-concurrency controls are retained. The channel bounds output batches, not Parquet's internal row-group buffers. This adds CPU task/channel overhead; no throughput benchmark has been run.
Are these changes tested?
cargo test -p iceberg --lib arrow::: 291 passed.cargo test -p iceberg --lib scan::: 104 passed.cargo clippy -p iceberg --lib --tests -- -D warnings: passed.cargo fmt --all -- --check: passed.AI Disclosure
Implementation, tests, and this description were prepared with OpenAI Codex assistance.