Skip to content

ParquetFileWriter.end() flushes an incomplete file to storage when finalizing fails #3706

Description

@arg3t

Describe the bug, including details regarding any error messages, version, and platform.

close() in end()'s finally runs before the writer is marked aborted, so a failure serializing indexes/bloom filters/footer flushes the incomplete stream. And withAbortOnFailure only catches IOException, so a RuntimeException/Error (e.g. OOM) never aborts at all. On non-rename storage this leaves a corrupt file at the final path that a retried task silently reuses. Affects the parquet-hadoop writer path on master, any platform.

Component(s)

Core

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions