Attach files

Attach Screenshot

Voters

Watch issue

Watchers

Create sub-task

Link

Clone

Update Comment Author

Replace String in Comment

Update Comment Visibility

Delete Comments

XML

Word

Printable

JSON

Details

Type: Bug
Status: Closed
Priority: Major
Resolution: Fixed
Affects Version/s: 1.13.5, 1.14.3
Fix Version/s: 1.14.5, 1.15.0
Component/s: Runtime / Checkpointing
Labels:
- pull-request-available

Description

After triggerCheckpoint, if checkpoint failed, flink will execute the tolerable-failed-checkpoints logic. But if triggerCheckpoint failed, flink won't execute the tolerable-failed-checkpoints logic.

How to reproduce this issue?

In our online env, hdfs sre deletes the flink base dir by mistake, and flink job don't have permission to create checkpoint dir. So cause flink trigger checkpoint failed.

There are some didn't meet expectations:

JM just log "Failed to trigger checkpoint for job 6f09d4a15dad42b24d52c987f5471f18 since Trigger checkpoint failure" . Don't show the root cause or exception.
user set tolerable-failed-checkpoints=0, but if triggerCheckpoint failed, flink won't execute the tolerable-failed-checkpoints logic.
When triggerCheckpoint failed, numberOfFailedCheckpoints is always 0
When triggerCheckpoint failed, we can't find checkpoint info in checkpoint history page.

All metrics are normal, so the next day we found out that the checkpoint failed, and the checkpoint has been failing for a day. it's not acceptable to the flink user.

I have some ideas:

Should tolerable-failed-checkpoints logic be executed when triggerCheckpoint fails?
When triggerCheckpoint failed, should increase numberOfFailedCheckpoints?
When triggerCheckpoint failed, should show checkpoint info in checkpoint history page?
JM just show "Failed to trigger checkpoint", should we show detailed exception to easy find the root cause?

Masters, could we do these changes? Please correct me if I'm wrong.