Skip to content

[FLINK-40210][checkpoint] Add Per-Region timeout handling and bounded source forced global checkpoint - #28964

Open
raoraoxiong wants to merge 5 commits into
apache:masterfrom
raoraoxiong:raorao/FLINK-40210-regional-checkpoint-timeout
Open

[FLINK-40210][checkpoint] Add Per-Region timeout handling and bounded source forced global checkpoint#28964
raoraoxiong wants to merge 5 commits into
apache:masterfrom
raoraoxiong:raorao/FLINK-40210-regional-checkpoint-timeout

Conversation

@raoraoxiong

Copy link
Copy Markdown
Contributor

Purpose

Completes the Regional Checkpoint feature (FLIP-600 Phase 6) by adding Per-Region timeout handling and bounded source forced global checkpoint, plus end-to-end integration tests.

Changes

Commit 1: Per-Region timeout handling and bounded source forced global checkpoint

  • Per-Region Timeout: when a task neither acknowledges nor declines within checkpoint timeout, CheckpointCanceller marks unacknowledged tasks as declined and triggers tryCompleteRegionalCheckpoint instead of aborting (FLIP-600 Section 9)
  • Bounded Source: DefaultExecutionGraph injects allSourcesFinishedChecker; when all sources finished, next checkpoint is forced global to ensure side effects (e.g. Kafka transactions) are committed before job termination
  • Add RegionalCheckpointTimeoutTest and RegionalCheckpointBoundedSourceTest

Commit 2: Integration tests

  • Add RegionalCheckpointITCase with white-box detection signals (numRegionalFallbackNotifications, numRegionalCompleteNotifications)
  • RegionFailingMapFunction implements CheckpointListener to verify notification dispatch
  • Tests: regional checkpoint during region failover, source data not lost, forced global after consecutive limit, ALL_TO_ALL topology fallback

Testing

  • Unit tests: timeout path, bounded source forced global (Tier 2 abort + reset)
  • Integration tests: end-to-end with MiniCluster

Dependencies

Generated-by: CodeBuddy Code (GLM-5.2)

…iguration and refCheckpointId infrastructure

- Add CheckpointListener.notifyRegionalCheckpointComplete(long, RegionalCheckpointInfo) for healthy-region tasks
- Add CheckpointListener.notifyRegionalCheckpointFallback(long, long) for failed-region tasks
- Add RegionalCheckpointInfo class with fallback checkpoint subtask mapping
- Add OperatorCoordinator.supportsRegionCheckpoint() and checkpointCoordinatorForRegionFallback()
- Wire OperatorCoordinatorCheckpointContext and OperatorCoordinatorHolder for forwarding
- Add 3 config options: region.enabled, region.max-failure-ratio, region.max-consecutive-failures
- Add refCheckpointId field to OperatorSubtaskState for tracking historical state references
- Add MetadataV7Serializer for backward-compatible serialization of refCheckpointId
- Add refCheckpointId to SubtaskStateStats/TaskStateStats for REST API aggregation
- Add CheckpointSubsumeHelper for reference-aware checkpoint cleanup
- Add regional config fields to CheckpointCoordinatorConfiguration

Generated-by: CodeBuddy Code (GLM-5.2)
…d SourceCoordinator fallback

- Add RegionalCheckpointHandler: decline buffering, region failure judgment, state recombination,
  two-tier max-consecutive-failures (Tier 1: force next global; Tier 2: abort + reset)
- Wire CheckpointCoordinator to delegate regional checkpoint logic to RegionalCheckpointHandler
- Add PendingCheckpoint methods: recordDecline, areAllTasksResponded, markUnacknowledgedTasksAsDeclined,
  reportFallbackSubtaskStats, finalizeRegionalCheckpoint
- Add DefaultCompletedCheckpointStore.computeReferencedCheckpointIds for cleaner reference protection
- Implement SourceCoordinator.supportsRegionCheckpoint() and checkpointCoordinatorForRegionFallback()
- Implement SplitAssignmentTracker per-checkpoint assignment history with backward-compatible serialization
- Wire Regional Checkpoint config through StreamGraph
- Add unit tests: success path, consecutive limit, state assembly, deferred abort, cleaner, config

Generated-by: CodeBuddy Code (GLM-5.2)
…dispatch and local state cleanup

- Extend confirmCheckpoint RPC with fallbackCheckpointId parameter (reuses task-side
  checkpoint-complete RPC path so notification survives task restarts)
- Add Task.notifyRegionalCheckpointFallback + CheckpointableTask.notifyRegionalCheckpointFallbackAsync
- Implement StreamTask.notifyRegionalCheckpointFallbackAsync with SubtaskCheckpointCoordinator
- Add SubtaskCheckpointCoordinator.notifyRegionalCheckpointFallback + OperatorChain propagation
- Add AbstractUdfStreamOperator forwarding to user functions
- Add TaskStateManager.pruneStateForCheckpoint for local state cleanup (FLIP-600 Section 9)
- TaskExecutor.confirmCheckpoint dispatches to notifyRegionalCheckpointFallback or notifyCheckpointComplete

Generated-by: CodeBuddy Code (GLM-5.2)
… source forced global checkpoint

- Per-Region Timeout: CheckpointCanceller marks unacknowledged tasks as declined and
  triggers tryCompleteRegionalCheckpoint instead of aborting (FLIP-600 Section 9)
- Bounded Source: DefaultExecutionGraph injects allSourcesFinishedChecker; when all sources
  finished, next checkpoint is forced global to ensure side effects committed before termination
- Add RegionalCheckpointTimeoutTest and RegionalCheckpointBoundedSourceTest

Generated-by: CodeBuddy Code (GLM-5.2)
- Add RegionalCheckpointITCase with white-box detection signals for regional fallback
  notifications (numRegionalFallbackNotifications, numRegionalCompleteNotifications)
- RegionFailingMapFunction implements CheckpointListener to verify notification dispatch
- Tests: regional checkpoint during region failover, source data not lost, forced global
  after consecutive limit, ALL_TO_ALL topology fallback

Generated-by: CodeBuddy Code (GLM-5.2)
@flinkbot

flinkbot commented Aug 13, 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