Flink: throttle FLIP-27 source enumerator for split discovery when Flink job is falling behind in streaming execution

[stevenzwu]

Feature Request / Improvement

Right now, FLIP-27 source eagerly discover all available splits from Iceberg table using incremental append scan. If the Flink job is falling behind with a lot of snapshots and data files, eagerly discovering all available splits can overwhelm enumerator with too many pending splits. It can increase memory pressure and enumerator checkpoint size. There is really no benefit of eagerly discovering all splits into memory. It is better to throttle the split discovery when there is certain number (configurable) of pending splits already.

PR #4943 (on the pre FLIP-27 source) is trying to avoid one incremental scan to cover too many snapshots. It can help in the spirit of throttling the scan scope. But it doesn't pause the split discovery if necessary.

Query engine

Flink

[github-actions]

This issue has been automatically marked as stale because it has been open for 180 days with no activity. It will be closed in next 14 days if no further activity occurs. To permanently prevent this issue from being considered stale, add the label 'not-stale', but commenting on the issue is preferred when possible.

[stevenzwu]

this is implemented by PR #6299