[
https://issues.apache.org/jira/browse/NIFI-6817?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16961082#comment-16961082
]
Bryan Bende commented on NIFI-6817:
-----------------------------------
The issue is a JVM limitation that limits a native library to being loaded by
only one class loader. The issue did become more noticeable when the HDFS
processors moved to per-instance class loading, instead of NAR class loading,
but it would still be an issue across NARs. For example, if the Hive NAR loads
snappy and the HDFS NAR loads snappy, only one of them will work.
> When there are multiple PutHDFS with snappy compression, will throw
> java.lang.UnsatisfiedLinkError
> --------------------------------------------------------------------------------------------------
>
> Key: NIFI-6817
> URL: https://issues.apache.org/jira/browse/NIFI-6817
> Project: Apache NiFi
> Issue Type: Bug
> Components: Extensions
> Affects Versions: 1.9.0, 1.9.2
> Environment: nifi-1.9.2
> centos 7.3
> Reporter: lixh
> Priority: Major
>
> The first flow put hdfs use snappy compression codec work correct. now copy
> the flow and start it, the second PutHDFS will throw
> java.lang.UnsatisfiedLinkError.
> The hadoop native lib path config in bootstrap.conf:
> java.arg.17=-Djava.library.path=/usr/lib/hadoop/native
> I guess，Because AbstractHadoopProcessor.java has
> @RequiresInstanceClassLoading(cloneAncestorResources = true),so
> org.apache.hadoop.util.NativeCodeLoader will be load by diffrent
> instanceClassloader
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
