Skip to content

[FLINK-40519][network] Deliver priority barrier to a credited remote reader while the subpartition is blocked during recovery - #29056

Open
1996fanrui wants to merge 1 commit into
apache:masterfrom
1996fanrui:FLINK-40519
Open

[FLINK-40519][network] Deliver priority barrier to a credited remote reader while the subpartition is blocked during recovery#29056
1996fanrui wants to merge 1 commit into
apache:masterfrom
1996fanrui:FLINK-40519

Conversation

@1996fanrui

Copy link
Copy Markdown
Member

What is the purpose of the change

A blocked PipelinedSubpartition still hands out priority buffers (unaligned-checkpoint barriers) in
pollBuffer(), but isDataAvailableUnsafe() and getBuffersInBacklogUnsafe() still gate on !isBlocked.
A credited remote reader decides availability through those methods, so its one-shot priority notification
is evaluated as "not available", the reader is not enqueued, and the barrier is not delivered until recovery
ends (resumeConsumption()) — the checkpoint can time out. The local channel path is already handled via
hasPendingPriorityEvent; the remote path was not.

Brief change log

  • [FLINK-40519] Extract PipelinedSubpartition#isBlockedForDelivery() and use it in pollBuffer, isDataAvailableUnsafe and getBuffersInBacklogUnsafe so a queued priority barrier is delivered to a credited reader while the subpartition is blocked.

Verifying this change

This change added a unit test:

  • Added a PipelinedSubpartitionTest case asserting that, while blocked with a priority barrier queued, the credited-reader availability check reports available and the barrier is pollable.

Does this pull request potentially affect one of the following parts:

  • Dependencies (does it add or upgrade a dependency): no
  • The public API, i.e., is any changed class annotated with @Public(Evolving): no
  • The serializers: no
  • The runtime per-record code paths (performance sensitive): no
  • Anything that affects deployment or recovery: yes (unaligned-checkpoint barrier delivery during recovery)
  • The S3 file system connector: no

Documentation

  • Does this pull request introduce a new feature? no
  • If yes, how is the feature documented? not applicable

Was generative AI tooling used to co-author this PR?
  • Yes (please specify the tool below)

…reader while the subpartition is blocked during recovery
@flinkbot

flinkbot commented Aug 31, 2026

Copy link
Copy Markdown
Collaborator

CI report:

Bot commands The @flinkbot bot supports the following commands:
  • @flinkbot run azure re-run the last Azure build

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

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants