[FLINK-18656] Start Delay metric is always zero for unaligned checkpoints

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 1.11.1
• Fix Version/s: 1.12.0, 1.11.2
• Component/s: Runtime / Metrics, Runtime / Network
• Labels: pull-request-available
• Assignee: Piotr Nowojski
• Reporter: Piotr Nowojski
• Created: 21/Jul/20 07:55
• Updated: 29/Jul/20 17:41
• Resolved: 29/Jul/20 12:22

Description

As visible in the attached screenshot, start delay is 0, despite very high end to end time and very quick async/sync times. This is impossible, as "start delay" should be equal to "end_to_end_time - sync_time - async_time".

The bug is probably in CheckpointBarrierUnaligner class, which seems to never invoke CheckpointBarrierHandler#markCheckpointStart.

It's important to make sure this metric is also set for source tasks (time to acquire checkpoint lock).

Issue Links

• is related to: FLINK-15603 Show checkpoint start delay in checkpoint statistics
• relates to: FLINK-18662 Provide more detailed metrics why unaligned checkpoint is taking long time
• links to: GitHub Pull Request #12989
• links to: GitHub Pull Request #13008