You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "Stavros Kontopoulos (JIRA)" <ji...@apache.org> on 2019/06/03 10:43:00 UTC
[jira] [Comment Edited] (SPARK-27900) Spark on K8s will not report
container failure due to oom
[ https://issues.apache.org/jira/browse/SPARK-27900?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16854449#comment-16854449 ]
Stavros Kontopoulos edited comment on SPARK-27900 at 6/3/19 10:42 AM:
----------------------------------------------------------------------
A better approach is to try use -XX:+ExitOnOutOfMemoryError or set the SparkUncaughtExceptionHandler [https://github.com/apache/spark/blob/master/core/src/main/scala/org/apache/spark/executor/Executor.scala#L88] in the driver (eg. SparkContext).
[~srowen@scient.com] do you know why we dont set the handler at the driver side?
was (Author: skonto):
A better approach to try here:
http://raindev.io/detecting-jvm-oome.html
[https://jarekprzygodzki.wordpress.com/2017/10/15/on-why-and-how-exit-jvm-on-onoutofmemoryerror|https://jarekprzygodzki.wordpress.com/2017/10/15/on-why-and-how-exit-jvm-on-onoutofmemoryerror/]
> Spark on K8s will not report container failure due to oom
> ---------------------------------------------------------
>
> Key: SPARK-27900
> URL: https://issues.apache.org/jira/browse/SPARK-27900
> Project: Spark
> Issue Type: Bug
> Components: Kubernetes
> Affects Versions: 3.0.0, 2.4.3
> Reporter: Stavros Kontopoulos
> Priority: Major
>
> A spark pi job is running:
> spark-pi-driver 1/1 Running 0 1h
> spark-pi2-1559309337787-exec-1 1/1 Running 0 1h
> spark-pi2-1559309337787-exec-2 1/1 Running 0 1h
> with the following setup:
> {quote}apiVersion: "sparkoperator.k8s.io/v1beta1"
> kind: SparkApplication
> metadata:
> name: spark-pi
> namespace: spark
> spec:
> type: Scala
> mode: cluster
> image: "skonto/spark:k8s-3.0.0-sa"
> imagePullPolicy: Always
> mainClass: org.apache.spark.examples.SparkPi
> mainApplicationFile: "local:///opt/spark/examples/jars/spark-examples_2.12-3.0.0-SNAPSHOT.jar"
> arguments:
> - "1000000"
> sparkVersion: "2.4.0"
> restartPolicy:
> type: Never
> nodeSelector:
> "spark": "autotune"
> driver:
> memory: "1g"
> labels:
> version: 2.4.0
> serviceAccount: spark-sa
> executor:
> instances: 2
> memory: "1g"
> labels:
> version: 2.4.0{quote}
> At some point the driver fails but it is still running and so the pods are still running:
> 19/05/31 13:29:20 INFO DAGScheduler: Submitting ResultStage 0 (MapPartitionsRDD[1] at map at SparkPi.scala:34), which has no missing parents
> 19/05/31 13:29:23 INFO MemoryStore: Block broadcast_0 stored as values in memory (estimated size 3.0 KiB, free 110.0 MiB)
> 19/05/31 13:29:23 INFO MemoryStore: Block broadcast_0_piece0 stored as bytes in memory (estimated size 1765.0 B, free 110.0 MiB)
> 19/05/31 13:29:23 INFO BlockManagerInfo: Added broadcast_0_piece0 in memory on spark-pi2-1559309337787-driver-svc.spark.svc:7079 (size: 1765.0 B, free: 110.0 MiB)
> 19/05/31 13:29:23 INFO SparkContext: Created broadcast 0 from broadcast at DAGScheduler.scala:1180
> 19/05/31 13:29:25 INFO DAGScheduler: Submitting 1000000 missing tasks from ResultStage 0 (MapPartitionsRDD[1] at map at SparkPi.scala:34) (first 15 tasks are for partitions Vector(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14))
> 19/05/31 13:29:25 INFO TaskSchedulerImpl: Adding task set 0.0 with 1000000 tasks
> Exception in thread "dag-scheduler-event-loop" java.lang.OutOfMemoryError: Java heap space
> at scala.collection.mutable.ResizableArray.ensureSize(ResizableArray.scala:106)
> at scala.collection.mutable.ResizableArray.ensureSize$(ResizableArray.scala:96)
> at scala.collection.mutable.ArrayBuffer.ensureSize(ArrayBuffer.scala:49)
> Mem: 2295260K used, 24458144K free, 1636K shrd, 48052K buff, 899424K cached
> $ kubectl describe pod spark-pi2-driver -n spark
> Name: spark-pi2-driver
> Namespace: spark
> Priority: 0
> PriorityClassName: <none>
> Node: gke-test-cluster-1-spark-autotune-46c36f4f-x3z9/10.138.0.44
> Start Time: Fri, 31 May 2019 16:28:59 +0300
> Labels: spark-app-selector=spark-74d8e5a8f1af428d91093dfa6ee9d661
> spark-role=driver
> sparkoperator.k8s.io/app-name=spark-pi2
> sparkoperator.k8s.io/launched-by-spark-operator=true
> sparkoperator.k8s.io/submission-id=spark-pi2-1559309336226927526
> version=2.4.0
> Annotations: <none>
> Status: Running
> IP: 10.12.103.4
> Controlled By: SparkApplication/spark-pi2
> Containers:
> spark-kubernetes-driver:
> Container ID: docker://55dadb603290b42f9ddb71959edf0224ddc7ea621ee15429941d3bcc7db9b71f
> Image: skonto/spark:k8s-3.0.0-sa
> Image ID: docker-pullable://skonto/spark@sha256:6268d760d1a006b69c7086f946e4d5d9a3b99f149832c63cfc7fe39671f5cda9
> Ports: 7078/TCP, 7079/TCP, 4040/TCP
> Host Ports: 0/TCP, 0/TCP, 0/TCP
> Args:
> driver
> --properties-file
> /opt/spark/conf/spark.properties
> --class
> org.apache.spark.examples.SparkPi
> spark-internal
> 1000000
> State: Running
> In the container processes are in _interruptible sleep_:
> PID PPID USER STAT VSZ %VSZ CPU %CPU COMMAND
> 15 1 185 S 2114m 7% 0 0% /usr/lib/jvm/java-1.8-openjdk/bin/java -cp /opt/spark/conf/:/opt/spark/jars/* -Xmx500m org.apache.spark.deploy.SparkSubmit --deploy-mode client --conf spar
> 287 0 185 S 2344 0% 3 0% sh
> 294 287 185 R 1536 0% 3 0% top
> 1 0 185 S 776 0% 0 0% /sbin/tini -s – /opt/spark/bin/spark-submit --conf spark.driver.bindAddress=10.12.103.4 --deploy-mode client --properties-file /opt/spark/conf/spark.prope
> Liveness checks might be a workaround but rest apis may be still working if threads in jvm still are running as in this case (I did check the spark ui and it was there).
>
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org