Hi, The per-job YARN mode also supports HA. Even if the YARN ResourceManager only brings up one JobManager at a time, you require ZooKeeper for storing metadata, such as which jobs are supposed to be running. In fact, the distributed components will still take part in a leader election with only one contender. The keys that are required to be configured for HA are:
high-availability: zookeeper high-availability.zookeeper.quorum: <your zk quorum> high-availability.zookeeper.storageDir: <path on DFS> Best, Gary On Mon, Aug 6, 2018 at 6:25 AM, 杨光 <laolang...@gmail.com> wrote: > Hi, > i am using the " per-job YARN session " mode deploy flink job on yarn. > The document https://ci.apache.org/projects/flink/flink-docs- > release-1.5/ops/jobmanager_high_availability.html#yarn- > cluster-high-availability > says that "we don’t run multiple JobManager (ApplicationMaster) > instances" but still give an example about how to config an Highly > Available YARN Session cluster. > > My question is: does the " per-job YARN session " mode should also > use HA setting ? Because in my test case without using zookeeper HA , > if the YarnApplicationMasterRunner/Jobmanager is failed or killed > ,yarn can only restart the YarnApplicationMasterRunner/Jobmanager > program but no job graph is submitted to the Jobmanager again. > > THANKS >