fishfishfishfishaa opened a new pull request, #7963:
URL: https://github.com/apache/paimon/pull/7963
### Purpose
Following **PIP-30 (Improvement For Paimon Committer In Flink)** , this PR
implements the **Paimon Write Coordinator (PWC)** to replace the current
`CommitOperator` with a JobManager-level `OperatorCoordinator`, eliminating the
network shuffle bottleneck for commit messages.
### Key Design Decision: Custom HDFS State
Instead of using Flink's native `StateBackend`, I chose custom HDFS state
management because:
1. **Timing mismatch**: Flink's operator state snapshot occurs after
`snapshotState()` returns, but PWC needs to persist aggregated messages
**before** acknowledging WriteOperators.
2. **Explicit recovery control**: Enables clean `resetToCheckpoint()` logic
with idempotent commit.
3. **Decoupled cleanup**: HDFS state deletion independent of Flink
checkpoint retention policies.
**State path:**
`<flink-checkpoint-dir>/pwc/<operatorId>/checkpoint-{ckId}.state`
### Current Scope
- **Supported**: `FixedBucketSink`
### Testing
- Added unit tests for recovery logic
### Related Issue
fix #2641
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
