[
https://issues.apache.org/jira/browse/SPARK-14638?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15281477#comment-15281477
]
Younos Aboulnaga commented on SPARK-14638:
------------------------------------------
Thank you very much for your response, Sean. When we get the exception, we get
it for all the apps using our data access layer, but we only get it in episodes
-- when this happens it consistently happens. The problem is that our
troubleshooting gets interrupted by the problem going away. The last time this
happened to me all what I did was go and have lunch; I was trying to run an app
and I was getting the problem for a few days, then one day before lunch I tried
something and it didn't work, but after lunch I tried the same build again and
it worked. And when the problem goes away, it is gone for quite a while. We
haven't been able to reliably reproduce the problem, even when using the same
ProtobufUtil in a POC app.
You are right that we should look at the class initialization code for
ProtobufUtil and any other classes that should have been loaded by the time the
exception happens. Since we could use ProtobufUtil successfully in a POC app,
we have been thinking that the exception is thrown during the initialization of
ProtobufUtil just because it is the first class being used. Maybe it is
actually the culprit but it is a nasty race condition in the initialization of
the class. The ProtobufUtil class does have some static blocks in it, and those
can cause trouble. We will get you some more information about this today.
Do you have a theory why setting the executor classpath with SPARK_CLASSPATH
reliably prevents the problem?
> Spark task does not have access to a dependency in the classloader of the
> executor thread
> -----------------------------------------------------------------------------------------
>
> Key: SPARK-14638
> URL: https://issues.apache.org/jira/browse/SPARK-14638
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
> Affects Versions: 1.2.1, 1.4.1, 1.6.0, 1.6.1
> Environment: > uname -a
> Linux HOSTNAME 3.13.0-74-generic #118-Ubuntu SMP Thu Dec 17 22:52:10 UTC 2015
> x86_64 x86_64 x86_64 GNU/Linux
> > java -version
> java version "1.8.0_77"
> Java(TM) SE Runtime Environment (build 1.8.0_77-b03)
> Java HotSpot(TM) 64-Bit Server VM (build 25.77-b03, mixed mode)
> Reporter: Younos Aboulnaga
>
> We have started to frequently see Spark apps failing because of a
> NoClassDefFoundError thrown despite that the dependency had been added to the
> ClassLoader just before it was thrown. The [Executor.run method adds the
> JAR|https://github.com/apache/spark/blob/v1.6.1/core/src/main/scala/org/apache/spark/executor/Executor.scala#L193]
> containing the class but then a NoClassDefFoundError is thrown subsequently.
> We see log messages from
> [updateDependencies|https://github.com/apache/spark/blob/v1.6.1/core/src/main/scala/org/apache/spark/executor/Executor.scala#L386]
> indicating that the JAR is fetched and added to the class loader. Upon
> inspection of the worker dir, the JAR is there, it is not corrupted, and it
> contains the class that could not be found in the class loader.
> We first saw this when we started writing streaming apps, and we thought it
> is something specific to streaming apps. However, this was wrong as the same
> problem happened with several batch apps.
> We first saw this on a Standalone cluster, and we though that it might be a
> problem caused by the lack of resource manager. Now we installed Mesos and
> the problem still happens.
> I tried to create a POC Spark App that demonstrates the problem, but I
> couldn't reliably reproduce it. The problem would still happen in other apps,
> but it didn't happen in the POC app even though I made it structurally the
> same as any other app we run. The problem seems to be environmental,
> specially because we found a work around for it.
> The work around we found is setting SPARK_CLASSPATH *on the executor nodes*
> to a local copy of the dependency. The problem still happens if we set the
> 'spark.executor.extraClassPath' or 'spark.driver.extraClassPath' or set
> SPARK_CLASSPATH on the driver node. However, if the SPARK_CLASSPATH is set on
> the executor node, then the problem doesn't happen because the JAR doesn't
> need to be added to the class loader by Executor#updateDependencies.
> Other symptoms of the problem are the following:
> 1) Even though there is a 'log4j.properties' in the
> 'spark.executor.extraClassPath', the first line of the stderr of the worker
> says "Using Spark's default log4j profile:
> org/apache/spark/log4j-defaults.properties" The log4j.properties file that is
> shipped with the job is totally neglected.
> 2) Any configuration files on 'spark.executor.extraClassPath' are neglected.
> I am mentioning this because log4j.properties is loaded very early on and in
> a static call, which might sway the troubleshooting into wrong directions.
> Here is the specific example in our case:
> > grep NoClassDef workers/app-20160414111328-0043/0/stderr
> Caused by: java.lang.NoClassDefFoundError: Could not initialize class
> org.apache.hadoop.hbase.protobuf.ProtobufUtil
> Caused by: java.lang.NoClassDefFoundError: Could not initialize class
> org.apache.hadoop.hbase.protobuf.ProtobufUtil
> .. SEVERAL ATTEMPTS ...
> Caused by: java.lang.NoClassDefFoundError: Could not initialize class
> org.apache.hadoop.hbase.protobuf.ProtobufUtil
> Even though, in the same application worker dir:
> > for j in workers/app-20160414111328-0043/0/*.jar ; do jar tf $j | grep
> > ProtobufUtil ; done;
> org/apache/hadoop/hbase/protobuf/ProtobufUtil$1.class
> org/apache/hadoop/hbase/protobuf/ProtobufUtil.class
> There are other examples, specially for configurations not being found. I
> think the SPARK-12279 can also be caused by the same root cause.
> We have been seeing this in several of our clusters and several engineers had
> spent days looking into why their applications suffer from this. We rebuilt
> our infrastructure (always on AWS EC2 nodes) and tested many hypotheses,
> including things that are non-sensical, and we still can't find anything that
> reliably reproduces the problem. The only reliable piece of information is
> that setting SPARK_CLASSPATH *on the executor nodes* prevents the problem
> from happening, because then the dependencies are included in the -cp
> parameter of the java command running the CoarseGrainedExecutorBackend .
> We would appreciate if someone more knowledgeable in Spark internals take a
> look, and we can help by providing as much details as possible.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
