【Appearance】For jobs with the same configuration (checkpoint interval: 3
minutes, job logic: regular join), flink1.9 runs normally. After flink1.12 runs for a period of time, the checkpoint creation time increases, and finally the checkpoint creation fails. 【Analysis】After learning flink1.10, the checkpoint mechanism is adjusted. The receiver will not cache the data after a single barrier arrives when the barrier is aligned, which means that the sender must wait for credit feedback to transmit data after the barrier is aligned, so the sender will generate certain The cold start of Flink affects the delay and network throughput. Therefore, the checkpoint interval is adjusted to 10 minutes for comparative testing, and it is found that after the adjustment (interval is 10), the job running on flink 1.12 is running normally. issue:https://issues.apache.org/jira/browse/FLINK-16404 【Question】1.Have you encountered the same problem? 2.Can flink1.12 set small checkpoint interval? The checkpoint interval is 3 minutes after the flink1.12 job runs for 5 hours, the checkpoint creation fails, the specific exception stack: org.apache.flink.util.FlinkRuntimeException: Exceeded checkpoint tolerable failure threshold. at org.apache.flink.runtime.checkpoint.CheckpointFailureManager.handleCheckpointException(CheckpointFailureManager.java:96) at org.apache.flink.runtime.checkpoint.CheckpointFailureManager.handleJobLevelCheckpointException(CheckpointFailureManager.java:65) at org.apache.flink.runtime.checkpoint.CheckpointCoordinator.abortPendingCheckpoint(CheckpointCoordinator.java:1924) at org.apache.flink.runtime.checkpoint.CheckpointCoordinator.abortPendingCheckpoint(CheckpointCoordinator.java:1897) at org.apache.flink.runtime.checkpoint.CheckpointCoordinator.access$600(CheckpointCoordinator.java:93) at org.apache.flink.runtime.checkpoint.CheckpointCoordinator$CheckpointCanceller.run(CheckpointCoordinator.java:2038) at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511) at java.util.concurrent.FutureTask.run(FutureTask.java:266) at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$201(ScheduledThreadPoolExecutor.java:180) at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:293) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617) at java.lang.Thread.run(Thread.java:745) -- Sent from: http://apache-flink-user-mailing-list-archive.2336050.n4.nabble.com/ |
Hi From the description, the time used to complete the checkpoint in 1.12 is longer. could you share more detail about the time consumption when running job on 1.9 and 1.12? Best, Congxian Haihang Jing <[hidden email]> 于2021年3月23日周二 下午7:22写道: 【Appearance】For jobs with the same configuration (checkpoint interval: 3 |
Hi Haihang, I think your issue is not related to FLINK-16404, because that change should have small impact on checkpoint time, we already have a micro benchmark for that change (1s checkpoint interval) and no regression is seen. Could you share some more information, for example, the stack of the task which can not finish the checkpoint? Best, Yingjie Haihang Jing <[hidden email]> 于2021年3月25日周四 上午10:58写道: Hi,Congxian ,thanks for your replay. |
Hi Haihang, After scanning the user mailing list, I found some users have reported checkpoint timeout when using unaligned checkpoint, can you share which checkpoint mode do you use? (The information can be found in log or the checkpoint -> configuration tab in webui) Best, Yingjie Yingjie Cao <[hidden email]> 于2021年3月30日周二 下午4:29写道:
|
Free forum by Nabble | Edit this page |