Re: [External] : Re: StopWithSavepoint() method doesn't work in Java based flink native k8s operator

Posted by Fuyao Li-2 on
URL: http://deprecated-apache-flink-user-mailing-list-archive.369.s1.nabble.com/StopWithSavepoint-method-doesn-t-work-in-Java-based-flink-native-k8s-operator-tp43404p43466.html

Hello Matthias, Austin, Yang,

 

Just to offer a bit more information. It is a demo application and actually there is no real data send into the system when I was trying to test this behavior. So, Flink state is very small.

I also tried to trigger the REST API for stopWithSavepoint() inside the JM pod (the k8s operator by default uses ClusterIP, I can’t access it directly, The screenshot WebUI below is exposed through Ingress). The command I use here is:

 

root@flink-demo-594946fd7b-l4qn5:/opt/flink# curl -H "Content-type: application/json" -d '{

>     "drain" : false,

>     "targetDirectory" : null

> }' 'localhost:8081/jobs/869cc58f996b2c35137f2e874b692537/stop'

{"request-id":"cc36976826a625406e5ead2425317527"}

 

As you can see, I received a request id. After some time (> 5 mins), I don’t see the pod or the deployment is destroyed or shut down. The job is still in running, but NO new checkpoints is triggered. That leaves me the impression of the suspended but not terminated. I can see the screenshot below. As you can see, the pod is still running (That’s the reason why I can see such UI...)

 

 

 

 

According to your statement, this is not the expected behavior… Any ideas to fix this?

 

For the logs, I found a related Flink thread in Chinese[2] and reference the whole problem here [1]. This repo is the starting point of my k8s operator.

 

[1] https://github.com/wangyang0918/flink-native-k8s-operator/issues/4

[2] http://apache-flink.147419.n8.nabble.com/flink-1-11-on-kubernetes-tt4586.html#none

 

Thanks,

 

Best,

Fuyao

 

 

 

 

From: Matthias Pohl <[hidden email]>
Date: Monday, May 3, 2021 at 22:59
To: Fuyao Li <[hidden email]>
Cc: user <[hidden email]>, Yang Wang <[hidden email]>, Austin Cawley-Edwards <[hidden email]>
Subject: [External] : Re: StopWithSavepoint() method doesn't work in Java based flink native k8s operator

Hi Fuyao,

sorry for not replying earlier. The stop-with-savepoint operation shouldn't only suspend but terminate the job. Is it that you might have a larger state that makes creating the savepoint take longer? Even though, considering that you don't experience this behavior with your 2nd solution, I'd assume that we could ignore this possibility.

 

I'm gonna add Austin to the conversation as he worked with k8s operators as well already. Maybe, he can also give you more insights on the logging issue which would enable us to dig deeper into what's going on with stop-with-savepoint.

 

Best,

Matthias

 

On Tue, May 4, 2021 at 4:33 AM Fuyao Li <[hidden email]> wrote:

Hello,

 

Update:

I think stopWithSavepoint() only suspend the job. It doesn’t actually terminate (./bin/flink cancel) the job. I switched to cancelWithSavepoint() and it works here.

 

Maybe stopWithSavepoint() should only be used to update the configurations like parallelism? For updating the image, this seems to be not suitable, please correct me if I am wrong.

 

For the log issue, I am still a bit confused. Why it is not available in kubectl logs. How should I get access to it?

 

Thanks.

Best,

Fuyao

 

From: Fuyao Li <[hidden email]>
Date: Sunday, May 2, 2021 at 00:36
To: user <[hidden email]>, Yang Wang <[hidden email]>
Subject: [External] : Re: StopWithSavepoint() method doesn't work in Java based flink native k8s operator

Hello,

 

I noticed that first trigger a savepoint and then delete the deployment might cause the duplicate data issue. That could pose a bad influence to the semantic correctness. Please give me some hints on how to make the stopWithSavepoint() work correctly with Fabric8io Java k8s client to perform this image update operation. Thanks!

 

Best,

Fuyao

 

 

 

From: Fuyao Li <[hidden email]>
Date: Friday, April 30, 2021 at 18:03
To: user <[hidden email]>, Yang Wang <[hidden email]>
Subject: [External] : Re: StopWithSavepoint() method doesn't work in Java based flink native k8s operator

Hello Community, Yang,

 

I have one more question for logging. I also noticed that if I execute kubectl logs  command to the JM. The pods provisioned by the operator can’t print out the internal Flink logs in the kubectl logs. I can only get something like the logs below. No actual flink logs is printed here… Where can I find the path to the logs? Maybe use a sidecar container to get it out? How can I get the logs without checking the Flink WebUI? Also, the sed error makes me confused here. In fact, the application is already up and running correctly if I access the WebUI through Ingress.

 

Reference: https://github.com/wangyang0918/flink-native-k8s-operator/issues/4

 

 

[root@bastion deploy]# kubectl logs -f flink-demo-594946fd7b-822xk

 

sed: couldn't open temporary file /opt/flink/conf/sedh1M3oO: Read-only file system

sed: couldn't open temporary file /opt/flink/conf/sed8TqlNR: Read-only file system

/docker-entrypoint.sh: line 75: /opt/flink/conf/flink-conf.yaml: Read-only file system

sed: couldn't open temporary file /opt/flink/conf/sedvO2DFU: Read-only file system

/docker-entrypoint.sh: line 88: /opt/flink/conf/flink-conf.yaml: Read-only file system

/docker-entrypoint.sh: line 90: /opt/flink/conf/flink-conf.yaml.tmp: Read-only file system

Start command: $JAVA_HOME/bin/java -classpath $FLINK_CLASSPATH -Xmx3462817376 -Xms3462817376 -XX:MaxMetaspaceSize=268435456 org.apache.flink.kubernetes.entrypoint.KubernetesApplicationClusterEntrypoint -D jobmanager.memory.off-heap.size=134217728b -D jobmanager.memory.jvm-overhead.min=429496736b -D jobmanager.memory.jvm-metaspace.size=268435456b -D jobmanager.memory.heap.size=3462817376b -D jobmanager.memory.jvm-overhead.max=429496736b

ERROR StatusLogger No Log4j 2 configuration file found. Using default configuration (logging only errors to the console), or user programmatically provided configurations. Set system property 'log4j2.debug' to show Log4j 2 internal initialization logging. See https://logging.apache.org/log4j/2.x/manual/configuration.html for instructions on how to configure Log4j 2

WARNING: An illegal reflective access operation has occurred

WARNING: Illegal reflective access by org.apache.flink.api.java.ClosureCleaner (file:/opt/flink/lib/flink-dist_2.11-1.12.1.jar) to field java.util.Properties.serialVersionUID

WARNING: Please consider reporting this to the maintainers of org.apache.flink.api.java.ClosureCleaner

WARNING: Use --illegal-access=warn to enable warnings of further illegal reflective access operations

WARNING: All illegal access operations will be denied in a future release

 

 

-------- The logs stops here, flink applications logs doesn’t get printed here anymore---------

 

^C

[root@bastion deploy]# kubectl logs -f flink-demo-taskmanager-1-1

sed: couldn't open temporary file /opt/flink/conf/sedaNDoNR: Read-only file system

sed: couldn't open temporary file /opt/flink/conf/seddze7tQ: Read-only file system

/docker-entrypoint.sh: line 75: /opt/flink/conf/flink-conf.yaml: Read-only file system

sed: couldn't open temporary file /opt/flink/conf/sedYveZoT: Read-only file system

/docker-entrypoint.sh: line 88: /opt/flink/conf/flink-conf.yaml: Read-only file system

/docker-entrypoint.sh: line 90: /opt/flink/conf/flink-conf.yaml.tmp: Read-only file system

Start command: $JAVA_HOME/bin/java -classpath $FLINK_CLASSPATH -Xmx697932173 -Xms697932173 -XX:MaxDirectMemorySize=300647712 -XX:MaxMetaspaceSize=268435456 org.apache.flink.kubernetes.taskmanager.KubernetesTaskExecutorRunner -D taskmanager.memory.framework.off-heap.size=134217728b -D taskmanager.memory.network.max=166429984b -D taskmanager.memory.network.min=166429984b -D taskmanager.memory.framework.heap.size=134217728b -D taskmanager.memory.managed.size=665719939b -D taskmanager.cpu.cores=1.0 -D taskmanager.memory.task.heap.size=563714445b -D taskmanager.memory.task.off-heap.size=0b --configDir /opt/flink/conf -Djobmanager.memory.jvm-overhead.min='429496736b' -Dpipeline.classpaths='file:usrlib/quickstart-0.1.jar' -Dtaskmanager.resource-id='flink-demo-taskmanager-1-1' -Djobmanager.memory.off-heap.size='134217728b' -Dexecution.target='embedded' -Dweb.tmpdir='/tmp/flink-web-d7691661-fac5-494e-8154-896b4fe30692' -Dpipeline.jars='file:/opt/flink/usrlib/quickstart-0.1.jar' -Djobmanager.memory.jvm-metaspace.size='268435456b' -Djobmanager.memory.heap.size='3462817376b' -Djobmanager.memory.jvm-overhead.max='429496736b'

ERROR StatusLogger No Log4j 2 configuration file found. Using default configuration (logging only errors to the console), or user programmatically provided configurations. Set system property 'log4j2.debug' to show Log4j 2 internal initialization logging. See https://logging.apache.org/log4j/2.x/manual/configuration.html for instructions on how to configure Log4j 2

WARNING: An illegal reflective access operation has occurred

WARNING: Illegal reflective access by org.apache.flink.shaded.akka.org.jboss.netty.util.internal.ByteBufferUtil (file:/opt/flink/lib/flink-dist_2.11-1.12.1.jar) to method java.nio.DirectByteBuffer.cleaner()

WARNING: Please consider reporting this to the maintainers of org.apache.flink.shaded.akka.org.jboss.netty.util.internal.ByteBufferUtil

WARNING: Use --illegal-access=warn to enable warnings of further illegal reflective access operations

WARNING: All illegal access operations will be denied in a future release

Apr 29, 2021 12:58:34 AM oracle.simplefan.impl.FanManager configure

SEVERE: attempt to configure ONS in FanManager failed with oracle.ons.NoServersAvailable: Subscription time out

 

 

-------- The logs stops here, flink applications logs doesn’t get printed here anymore---------

 

 

Best,

Fuyao

 

 

From: Fuyao Li <[hidden email]>
Date: Friday, April 30, 2021 at 16:50
To: user <[hidden email]>, Yang Wang <[hidden email]>
Subject: [External] : StopWithSavepoint() method doesn't work in Java based flink native k8s operator

Hello Community, Yang,

 

I am trying to extend the flink native Kubernetes operator by adding some new features based on the repo [1]. I wrote a method to release the image update functionality. [2] I added the

triggerImageUpdate(oldFlinkApp, flinkApp, effectiveConfig);

 

under the existing method.

triggerSavepoint(oldFlinkApp, flinkApp, effectiveConfig);

 

 

I wrote a function to accommodate the image change behavior.[2]

 

Solution1:

I want to use stopWithSavepoint() method to complete the task. However, I found it will get stuck and never get completed. Even if I use get() for the completeableFuture. It will always timeout and throw exceptions. See solution 1 logs [3]

 

Solution2:

I tried to trigger a savepoint, then delete the deployment in the code and then create a new application with new image. This seems to work fine. Log link: [4]

 

My questions:

1. Why solution 1 will get stuck? triggerSavepoint() CompleteableFuture could work here… Why stopWithSavepoint() will always get stuck or timeout? Very confused.
2. For Fabric8io library, I am still new to it, did I do anything wrong in the implementation, maybe I should update the jobStatus? Please give me some suggestions.
3. For work around solution 2, is there any bad influence I didn’t notice?

 

 

[1] https://github.com/wangyang0918/flink-native-k8s-operator

[2] https://pastebin.ubuntu.com/p/tQShjmdcJt/

[3] https://pastebin.ubuntu.com/p/YHSPpK4W4Z/

[4] https://pastebin.ubuntu.com/p/3VG7TtXXfh/

 

Best,

Fuyao