[FLINK-13270] CheckpointCoordinator abort the checkpoint

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Duplicate
• Affects Version/s: 1.6.3
• Component/s: Runtime / Checkpointing
• Assignee: Unassigned
• Reporter: molsion mo
• Created: 15/Jul/19 11:46
• Updated: 15/Jul/19 12:07
• Resolved: 15/Jul/19 12:07

Description

1. Submitting a job that source is Kafka, and Kafka partition is 1 , if set source Parallelism to 4,the job will be submitted successfully and source operator is RUNNING status but three Parallelism task is FINISHED
2. checkpoint will abort when org.apache.flink.runtime.checkpoint.CheckpointCoordinator#triggerCheckpoint trigger and return CheckpointTriggerResult(CheckpointDeclineReason.NOT_ALL_REQUIRED_TASKS_RUNNING), this job do not have a successful checkpoint and lost runtime status
3. Here is the log

July 15th 2019, 19:29:43.230 Checkpoint triggering task Source: org.apache.flink.streaming.connectors.kafka.FlinkKafkaConsumer@132e3594-Stream -> Calc(where: (CAST(msg) LIKE '%我们%'), select: (CAST(msg) AS kkey, CAST(key) AS msg)) -> SinkConversion to Row (1/4) of job 2e1b28950a38d62803a9840609b55612 is not being executed at the moment. Aborting checkpoint. 


July 15th 2019, 19:28:43.230 Checkpoint triggering task Source: org.apache.flink.streaming.connectors.kafka.FlinkKafkaConsumer@132e3594-Stream -> Calc(where: (CAST(msg) LIKE '%我们%'), select: (CAST(msg) AS kkey, CAST(key) AS msg)) -> SinkConversion to Row (1/4) of job 2e1b28950a38d62803a9840609b55612 is not being executed at the moment. Aborting checkpoint.

 

We want to be able to trigger checkpoint and store it if we have a finished node in operator,or we want to be able to check Parallelism problem when we submit the job(eg : if Parallelism  > kafkaPartition then submit unsuccessfully)

 

 

 

 

Issue Links

• duplicates: FLINK-2491 Support Checkpoints After Tasks Finished