You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "jugosag (Jira)" <ji...@apache.org> on 2019/12/18 10:52:00 UTC
[jira] [Commented] (SPARK-27426) SparkAppHandle states not getting
updated in Kubernetes
[ https://issues.apache.org/jira/browse/SPARK-27426?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16999035#comment-16999035 ]
jugosag commented on SPARK-27426:
---------------------------------
We are observing the exact same problem on Windows with Spark 2.4.4 (patched to resolve SPARK-28921). It works when running in local mode.
From what we understand about the mechanism in place is that SparkLauncher opens a server port on which someone (SparkSubmit?) should report back the state of the application based on which the SparkAppHandle info is then updated.
When running in local mode, we see such a connection from SparkSubmit to our application, but nothing when running the application in k8s mode. Is this just not expected to work for K8s (a.k.a. missing feature) or is this a defect?
> SparkAppHandle states not getting updated in Kubernetes
> -------------------------------------------------------
>
> Key: SPARK-27426
> URL: https://issues.apache.org/jira/browse/SPARK-27426
> Project: Spark
> Issue Type: Bug
> Components: Kubernetes
> Affects Versions: 2.4.0
> Environment: # Cent os 7
> # Kubernetes 1.14
> # Spark 2.4.0
> Reporter: Nishant Ranjan
> Priority: Major
> Labels: SparkAppHandle, kubernetes, sparklauncher
>
> While launching Spark application through "startApplication()" , SparkAppHandle state is not getting updated.
> sparkLaunch = new SparkLauncher()
> .setSparkHome("/root/test/spark-2.4.0-bin-hadoop2.7")
> .setMaster("k8s://https://172.16.23.30:6443")
> .setVerbose(true)
> .addSparkArg("--verbose")
> .setAppResource("local:///opt/spark/examples/jars/spark-examples_2.11-2.4.0.jar")
> .setConf("spark.app.name","spark-pi")
> .setMainClass("org.apache.spark.examples.SparkPi")
> .setConf("spark.executor.instances","5")
> .setConf("spark.kubernetes.container.image","registry.renovite.com/spark:v2")
> .setConf("spark.kubernetes.driver.pod.name","spark-pi-driver")
> .setConf("spark.kubernetes.container.image.pullSecrets","dev-registry-key")
> .setConf("spark.kubernetes.authenticate.driver.serviceAccountName","spark")
> .setDeployMode("cluster")
> ;
> SparkAppHandle handle = sparkLaunch.startApplication();
> Observations:
> # Now, I tried listeners etc but handle.getState() returns UNKNOWN and when Spark application is completed. state changes to LOST.
> # SparkAppHandle is not null
> # handle.getAppId() is always null.
> My best guess is that communication is not working properly between listener and Spark driver in Kubernetes.
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org