Skip to content

CPU work done on IO runtime #3589

Open
Open

Description

@alexanderbianchi

Apache Iceberg Rust version

v1.94.0

Describe the bug

Parquet decoding is not consistently isolated from I/O work when using separate CPU and I/O runtimes:

  • Data-file footer and page-index decoding run on the I/O runtime.
  • Data-file page decoding runs on whichever runtime polls the scan stream, which may be the I/O runtime.
  • Delete-file footer parsing, page decoding, and delete-filter construction run on the I/O runtime.

This allows CPU-intensive decoding and decompression to compete with storage requests on I/O workers.

Expected behavior: perform Parquet decoding and batch processing on the configured CPU runtime, while keeping storage opens and byte reads on the I/O runtime. Preserve concurrency limits, backpressure, cancellation, and error propagation.

Expected behavior

Decoding work should be done in the CPU runtime.

Willingness to contribute

Draft PRs:

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions