[
https://issues.apache.org/jira/browse/FLINK-36227?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Alexander Fedulov updated FLINK-36227:
--------------------------------------
Fix Version/s: 1.19.2
> NullPointerException when starting flink with logback logger
> ------------------------------------------------------------
>
> Key: FLINK-36227
> URL: https://issues.apache.org/jira/browse/FLINK-36227
> Project: Flink
> Issue Type: Bug
> Components: API / Core
> Affects Versions: 1.20.0
> Environment: The error rise when flink is staring with logback as
> logger
> Reporter: Szymon
> Priority: Critical
> Labels: pull-request-available
> Fix For: 1.19.2, 1.20.1
>
>
> I think that there is a bug in flink code in class MdcUtils.
>
> When I run job on standalone flink or unit test with MiniCluster I get
> below error. This error rise only when logback is used, as logging adapter.
> I use logback version 1.2.13 because in flink documentation
> [https://nightlies.apache.org/flink/flink-docs-release-1.20/docs/deployment/advanced/logging/]
> is written “Logback 1.3+ requires SLF4J 2, which is currently not supported.”
>
> _14:33:41.111 [flink-pekko.actor.default-dispatcher-5] INFO
> org.apache.flink.runtime.rpc.pekko.PekkoRpcActor - The RpcEndpoint
> taskmanager_0 failed._
> _java.lang.NullPointerException: null_
> _at java.base/java.util.HashMap.putMapEntries(HashMap.java:497)_
> _at java.base/java.util.HashMap.putAll(HashMap.java:781)_
> _at
> java.base/java.util.Collections$SynchronizedMap.putAll(Collections.java:2604)_
> _at
> ch.qos.logback.classic.util.LogbackMDCAdapter.setContextMap(LogbackMDCAdapter.java:197)_
> _at org.slf4j.MDC.setContextMap(MDC.java:264)_
> _at
> org.apache.flink.util.MdcUtils.lambda$withContext$0(MdcUtils.java:48)_
>
> I checked flink code and in my opinion there should be additional check in
> class MdcUtils in method
> _public static MdcCloseable withContext(Map<String, String> context) {_
> _final Map<String, String> orig = MDC.getCopyOfContextMap();_
> _MDC.setContextMap(context);_
> _return () -> MDC.setContextMap(orig);_
> _}_
>
> this method use MDC.getCopyOfContextMap();. Description of this method says
> that returned value may be null. So when variable orig is null and is passed
> to “MDC.setContextMap”. At the end this null is passed to class
> LogbackMDCAdapter to method setContextMap where this null Is passed to
> newMap.putAll(contextMap), and then NullPointerException rise. I checked this
> for logback 1.2.13 where setContextMap is
> _public void setContextMap(Map<String, String> contextMap) {_
> _lastOperation.set(WRITE_OPERATION);_
> _Map<String, String> newMap = Collections.synchronizedMap(new
> HashMap<String, String>());_
> _newMap.putAll(contextMap);_
> _// the newMap replaces the old one for serialisation's sake_
> _copyOnThreadLocal.set(newMap);_
> _}_
>
> This error should not appear when logback version is 1.3.3 or higher
> _@SuppressWarnings("unchecked")_
> _public void setContextMap(Map contextMap) {_
> _if (contextMap != null) {_
> _readWriteThreadLocalMap.set(new HashMap<String,
> String>(contextMap));_
> _} else {_
> _readWriteThreadLocalMap.set(null);_
> _}_
> _nullifyReadOnlyThreadLocalMap();_
> _}_
>
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
