Uploaded image for project: 'Flink'
  1. Flink
  2. FLINK-9567

Flink does not release resource in Yarn Cluster mode

    Details

      Description

      After restart the Job Manager in Yarn Cluster mode, sometimes Flink does not release task manager containers in some specific case. In the worst case, I had a job configured to 5 task managers, but possess more than 100 containers in the end. Although the task didn't failed, but it affect other jobs in the Yarn Cluster.

      In the first log I posted, the container with id 24 is the reason why Yarn did not release resources. As the container was killed before restart, but it has not received the callback of onContainerComplete in YarnResourceManager which should be called by AMRMAsyncClient of Yarn. After restart, as we can see in line 347 of FlinkYarnProblem log, 

      2018-06-14 22:50:47,846 WARN akka.remote.ReliableDeliverySupervisor - Association with remote system [akka.tcp://flink@bd-r1hdp69:30609] has failed, address is now gated for [50] ms. Reason: [Disassociated]

      Flink lost the connection of container 24 which is on bd-r1hdp69 machine. When it try to call closeTaskManagerConnection in onContainerComplete, it did not has the connection to TaskManager on container 24, so it just ignore the close of TaskManger.

      2018-06-14 22:50:51,812 DEBUG org.apache.flink.yarn.YarnResourceManager - No open TaskExecutor connection container_1528707394163_29461_02_000024. Ignoring close TaskExecutor connection.

       However, bafore calling closeTaskManagerConnection, it already called requestYarnContainer which lead to numPendingContainerRequests variable in YarnResourceManager increased by 1.

      As the excessive container return is determined by the numPendingContainerRequests variable in YarnResourceManager, it cannot return this container although it is not required. Meanwhile, the restart logic has already allocated enough containers for Task Managers, Flink will possess the extra container for a long time for nothing. 

      In the full log, the job ended with 7 containers while only 3 are running TaskManagers.

      ps: Another strange thing I found is that when sometimes request for a yarn container, it will return much more than requested. Is it a normal scenario for AMRMAsyncClient?

        Attachments

        1. FlinkYarnProblem
          130 kB
          Shimin Yang
        2. fulllog.txt
          42.00 MB
          Shimin Yang

          Activity

            People

            • Assignee:
              dangdangdang Shimin Yang
              Reporter:
              dangdangdang Shimin Yang
            • Votes:
              0 Vote for this issue
              Watchers:
              5 Start watching this issue

              Dates

              • Created:
                Updated:
                Resolved: