[FLINK-20087] CheckpointCoordinator waits until all tasks finish initialization of states to trigger checkpoint

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 1.9.0
• Component/s: Runtime / Checkpointing
• Assignee: Unassigned
• Reporter: Jiayi Liao
• Created: 11/Nov/20 12:02
• Updated: 20/Apr/21 15:46
• Resolved: 08/Jan/21 09:46

Description

CheckpointCoordinator will start triggering checkpoint after all tasks send RUNNING status to JobMaster. However, the initializeState could be time-expensive(a few minutes), for example the task needs to rescale with multiple old tasks' rocksdb data. During this process, the triggering of checkpoints are meaningless. 

Issue Links

• duplicates: FLINK-22379 Do not trigger checkpoint when non source tasks are INITIALIZING
• duplicates: FLINK-17012 Expose stage of task initialization