flink-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Till Rohrmann <trohrm...@apache.org>
Subject Re: Docker-Flink Project: TaskManagers can't talk to JobManager if they are on different nodes
Date Mon, 06 Nov 2017 13:48:59 GMT
I'm not entirely sure how docker swarm works but from the Flink perspective
there mustn't be two TaskManagers running on the same host (meaning an
entity where you share the same address) if you set the TaskManager data
port to a fixed value (otherwise only one of them can be started due to
port conflicts). If you can ensure that this is the case, then it should be
save to specify a port for the data transmission.

Cheers,
Till

On Mon, Nov 6, 2017 at 2:37 PM, Vergilio, Thalita <
t.vergilio4822@student.leedsbeckett.ac.uk> wrote:

> Hi Till,
>
>
> Thanks a lot for your answer.
>
>
> Is the taskmanager.data.port unique per TaskManager? The documentation
> says it is assigned at runtime by the OS. My thinking here is that you
> would need to know what that is at service creation time, which would go
> against the whole idea of how services are scaled in Docker Swarm.
>
>
> When you create a Swarm service using 'docker stack deploy' or 'docker
> service create', the configuration that is used at that point is the same
> that will be used by all instances of the service. If you then scale
> TaskManager to 8 or 10 containers, each of them gets the same service
> configuration(the one used to create the service).
>
>
> I have in fact tried to map specific ports in the TaskManager service
> configuration, but then I got "port already in use" when I tried to scale
> up the service.
>
>
> I wonder if there is a way around it.
>
>
> Maybe the people who developed the create-docker-swarm-service.sh script
> in the docker-flink project would be able to shed some light?
>
>
> ------------------------------
> *From:* Till Rohrmann <trohrmann@apache.org>
> *Sent:* 06 November 2017 12:40:33
> *To:* Piotr Nowojski
> *Cc:* Vergilio, Thalita; user@flink.apache.org; Patrick Lucas
>
> *Subject:* Re: Docker-Flink Project: TaskManagers can't talk to
> JobManager if they are on different nodes
>
> Hi Thalita,
>
> in order to make Flink work, I think you have to expose the JobManager RPC
> port, the Blob server port and make sure that the TaskManager can talk to
> each other by exposing the `taskmanager.data.port`. The query server port
> is only necessary if you want to use queryable state.
>
> I've pulled in Patrick who has more experience with running Flink on top
> of Docker. He'll definitely be able to provide more detailed
> recommendations.
>
> Cheers,
> Till
>
> On Mon, Nov 6, 2017 at 9:22 AM, Piotr Nowojski <piotr@data-artisans.com>
> wrote:
>
> Till, is there somewhere a list of ports that need to exposed that’s more
> up to date compared to docker-flunk README?
>
> Piotrek
>
> On 3 Nov 2017, at 10:23, Vergilio, Thalita <t.vergilio4822@student.leedsb
> eckett.ac.uk> wrote:
>
> Just an update: by changing the JOB_MANAGER_RPC_ADDRESS to the public IP
> of the JobManager and exposing port 6123 as {{PUBLIC_IP}}:6123:6123, I
> manged to get the TaskManagers from different nodes and even different
> subnets to talk to the JobManager.
>
> This is how I created the services:
>
> docker network create -d overlay overlay
>
> docker service create --name jobmanager --env JOB_MANAGER_RPC_ADDRESS=
> {{PUBLIC_IP}}  -p 8081:8081 -p{{PUBLIC_IP}}:6123:6123 -p 48081:48081 -p
> 6124:6124 -p 6125:6125 --network overlay --constraint 'node.hostname ==
> ubuntu-swarm-manager' flink jobmanager
>
> docker service create --name taskmanager --env JOB_MANAGER_RPC_ADDRESS=
> {{PUBLIC_IP}}  -p 6121:6121 -p 6122:6122  --network overlay --constraint
> 'node.hostname != ubuntu-swarm-manager' flink taskmanager
>
> However, I am still encountering errors further down the line. When I
> submit a job using the Web UI, it fails because the JobManager can't talk
> to the TaskManager on port 35033. I presume this is the
> taskmanager.data.port, which needs to be set to a range and this range
> exposed when I create the service?
>
> Are there any other ports that I need to open at service creation time?
>
> Connecting the channel failed: Connecting to remote task manager + '/{{IP_ADDRESS_OF_MANAGER}}:35033'
has failed. This might indicate that the remote task manager has been lost.
> 	at org.apache.flink.runtime.io.network.netty.PartitionRequestClientFactory$ConnectingChannel.waitForChannel(PartitionRequestClientFactory.java:196)
> 	at org.apache.flink.runtime.io.network.netty.PartitionRequestClientFactory$ConnectingChannel.access$000(PartitionRequestClientFactory.java:131)
> 	at org.apache.flink.runtime.io.network.netty.PartitionRequestClientFactory.createPartitionRequestClient(PartitionRequestClientFactory.java:83)
> 	at org.apache.flink.runtime.io.network.netty.NettyConnectionManager.createPartitionRequestClient(NettyConnectionManager.java:59)
> 	at org.apache.flink.runtime.io.network.partition.consumer.RemoteInputChannel.requestSubpartition(RemoteInputChannel.java:112)
> 	at org.apache.flink.runtime.io.network.partition.consumer.SingleInputGate.requestPartitions(SingleInputGate.java:433)
> 	at org.apache.flink.runtime.io.network.partition.consumer.SingleInputGate.getNextBufferOrEvent(SingleInputGate.java:455)
> 	at org.apache.flink.streaming.runtime.io.BarrierTracker.getNextNonBlocked(BarrierTracker.java:91)
> 	at org.apache.flink.streaming.runtime.io.StreamInputProcessor.processInput(StreamInputProcessor.java:213)
> 	at org.apache.flink.streaming.runtime.tasks.OneInputStreamTask.run(OneInputStreamTask.java:69)
> 	at org.apache.flink.streaming.runtime.tasks.StreamTask.invoke(StreamTask.java:263)
> 	at org.apache.flink.runtime.taskmanager.Task.run(Task.java:702)
> 	at java.lang.Thread.run(Thread.java:748)
>
>
>
> ------------------------------
> *From:* Piotr Nowojski <piotr@data-artisans.com>
> *Sent:* 02 November 2017 14:26:32
> *To:* Vergilio, Thalita
> *Cc:* user@flink.apache.org
> *Subject:* Re: Docker-Flink Project: TaskManagers can't talk to
> JobManager if they are on different nodes
>
> Did you try to expose required ports that are listed in the README when
> starting the containers?
>
> https://github.com/apache/flink/tree/master/flink-contrib/docker-flink
>
> Ports:
> • The Web Client is on port 48081
> • JobManager RPC port 6123 (default, not exposed to host)
> • TaskManagers RPC port 6122 (default, not exposed to host)
> • TaskManagers Data port 6121 (default, not exposed to host)
>
> Piotrek
>
> On 2 Nov 2017, at 14:44, javalass <t.vergilio4822@student.leedsb
> eckett.ac.uk> wrote:
>
> I am using the Docker-Flink project in:
> https://github.com/apache/flink/tree/master/flink-contrib/docker-flink
>
> I am creating the services with the following commands:
> docker network create -d overlay overlay
> docker service create --name jobmanager --env
> JOB_MANAGER_RPC_ADDRESS=jobmanager -p 8081:8081 --network overlay
> --constraint 'node.hostname == ubuntu-swarm-manager' flink jobmanager
> docker service create --name taskmanager --env
> JOB_MANAGER_RPC_ADDRESS=jobmanager --network overlay --constraint
> 'node.hostname != ubuntu-swarm-manager' flink taskmanager
>
> I wonder if there's any configuration I'm missing. This is the error I get:
> - Trying to register at JobManager akka.tcp://flink@jobmanager:6123/
> user/jobmanager (attempt 4, timeout: 4000 milliseconds)
>
>
>
>
>
>
>
> --
> Sent from: http://apache-flink-user-mailing-list-archive.2336050.
> n4.nabble.com/
>
>
> To view the terms under which this email is distributed, please go to:-
> http://disclaimer.leedsbeckett.ac.uk/disclaimer/disclaimer.html
>
>
>
> To view the terms under which this email is distributed, please go to:-
> http://disclaimer.leedsbeckett.ac.uk/disclaimer/disclaimer.html
>
>

Mime
View raw message