Apache Flink：独立集群尝试使用用户名“flink”进行连接

Question

For my master thesis I'm trying to set up a flink standalone cluster on 4 nodes. I've worked along the documentation which pretty neatly explains how to set it up. But when I start the cluster there is a warning and when I'm trying to run a job, there is an error with the same message:

akka.pattern.AskTimeoutException: Ask timed out on [Actor[akka.tcp://flink@MYHOSTNAME:6123/user/jobmanager#-818199108]] after [10000 ms]. Sender[null] sent message of type "org.apache.flink.runtime.messages.JobManagerMessages$LeaderSessionMessage"

Increasing the timeout didn't work. When I open the taskmanagers in web UI, all of them have the following pattern:

akka.tcp://flink@MYHOSTNAME:33779/user/taskmanager

Does anyone have an idea how to solve this to get the cluster working? Thanks in advance!

One last thing: There isn't a user "flink" on the cluster and won't be created. So any advices without telling me I should create that user would be very appreciated! Thanks!

Answer 1

Not sure if it is still relevant, but the way i did it (using Flink 1.5.3):

I setup a HA standalone cluster with 3 master (JobManager) and 20 slaves (TaskManager) in the following way.

1. Define your conf/masters file (hostname:8081 per line)
2. Define your conf/slaves file (each taskmanager hostname per line)
3. Define in the flink-conf.yaml on each master machine its own jobmanager.rpc.address hostname
4. Define in the flink-conf.yaml on each slave machine the jobmanager.rpc.address as localhost
5. Once every is set, execute the bin/start-cluster.sh on any of the master host.

If you need HA then you need to setup a zookeeper quorum and modify the corresponding properties regarding HA (high-availability, high-availability.storageDir, high-availability.zookeeper.quorum)