I have a flink cluster configure for zookeeper recovery mode runing on marathon. What's the best practice for submit an application to flink cluster by console?taskmanager log ,jobmanager log and client out
You basically have to configure the flink-conf.yaml with the same ZooKeeper settings as you have used to start the cluster. Given that your ZooKeeper cluster is reachable under zookeeer.foobar.com:2181 and you've set the Flink cluster id to flinkCluster, then you have to insert the following configuration settings into the flink-conf.yaml:
high-availability: zookeeper
high-availability.cluster-id: flinkCluster
high-availability.zookeeper.quorum: zookeeper.foobar.com:2181
The Flink version 1.2.x contains a bug which requires to set the JobManager's address and port via jobmanager.rpc.address and jobmanager.rpc.port in the flink-conf.yaml. Otherwise the ClusterClient fails with an exception or cannot retrieve the proper NIC with which it can talk to the remote JobManager. The problem is described by FLINK-6629.