[
https://issues.apache.org/jira/browse/SENTRY-2299?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16540673#comment-16540673
]
Na Li commented on SENTRY-2299:
-------------------------------
[~Tagar] To verify this is the case, can you enable debug in NameNode, so we
can see the details of the path and table that causes this issue? Before the
exception, you can see debug message containing "renameAuthzObject" before
exception that shows the old name and path as well as the new name and path.
> NPE In Sentry HDFS Sync Plug
> ----------------------------
>
> Key: SENTRY-2299
> URL: https://issues.apache.org/jira/browse/SENTRY-2299
> Project: Sentry
> Issue Type: Bug
> Components: Sentry
> Affects Versions: 2.1.0
> Reporter: Na Li
> Assignee: Na Li
> Priority: Major
>
> Sentry HDFS ACL synchronization stopped working and throws
> NullPointerException. The HDFS logs showed repeating errors like the
> following:
> {code}
> 11:16:15.743 AM WARN SentryAuthorizationInfo
> Failed to update, will retry in [30000]ms, error:
> java.lang.NullPointerException
> at org.apache.sentry.hdfs.HMSPaths$Entry.access$200(HMSPaths.java:146)
> at org.apache.sentry.hdfs.HMSPaths.renameAuthzObject(HMSPaths.java:879)
> at
> org.apache.sentry.hdfs.UpdateableAuthzPaths.applyPartialUpdate(UpdateableAuthzPaths.java:118)
> at
> org.apache.sentry.hdfs.UpdateableAuthzPaths.updatePartial(UpdateableAuthzPaths.java:81)
> at
> org.apache.sentry.hdfs.SentryAuthorizationInfo.processUpdates(SentryAuthorizationInfo.java:211)
> at
> org.apache.sentry.hdfs.SentryAuthorizationInfo.update(SentryAuthorizationInfo.java:139)
> at
> org.apache.sentry.hdfs.SentryAuthorizationInfo.run(SentryAuthorizationInfo.java:232)
> at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
> at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:308)
> at
> java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:180)
> at
> java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:294)
> at
> java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
> at
> java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
> at java.lang.Thread.run(Thread.java:748)
> The customer checked the Sentry logs and didn't see any corresponding errors.
> The issue stopped occurring, apparently not through any specific user
> intervention. (The customer tried manually failing over the active NameNode,
> with no change.)
> {code}
> Arjun mentioned the reason is that some delta update from sentry server was
> lost, so the oldEntry at line HMSPaths.java:879 was null. That caused null
> exception.
> {code}
> void renameAuthzObject(String oldName, List<List<String>> oldPathElems,
> String newName, List<List<String>> newPathElems) {
> if (LOG.isDebugEnabled()) {
> LOG.debug(String.format("%s renameAuthzObject({%s, %s} -> {%s, %s})",
> this, oldName, assemblePaths(oldPathElems), newName,
> assemblePaths(newPathElems)));
> }
> if (oldPathElems == null || oldPathElems.isEmpty() ||
> newPathElems == null || newPathElems.isEmpty() ||
> newName == null || newName.equals(oldName)) {
> LOG.warn(String.format("%s renameAuthzObject({%s, %s} -> {%s, %s})" +
> ": invalid inputs, skipping",
> this, oldName, assemblePaths(oldPathElems), newName,
> assemblePaths(newPathElems)));
> return;
> }
> // if oldPath == newPath, that is path has not changed as part of rename
> and hence new table
> // needs to have old paths => new_table.add(old_table_partition_paths)
> List<String> oldPathElements = oldPathElems.get(0);
> List<String> newPathElements = newPathElems.get(0);
> if (!oldPathElements.equals(newPathElements)) {
> Entry oldEntry = root.find(oldPathElements.toArray(new String[0]),
> false);
> Entry newParent = root.createParent(newPathElements);
> oldEntry.moveTo(newParent, newPathElements.get(newPathElements.size() -
> 1)); -> oldEntry is null
> }
> {code}
> There are several possible reasons why some delta changes are lost.
> {code}
> 1. Sentry server does not save the rename update as delta update. The chance
> is really low
> 2. The delta change is lost from sentry server to name node plugin. The
> chance is also low
> 3. When applying delta change about old entry, it is lost
> {code}
> The fix for this issue
> 1. Check if oldEntry is null. If so, don't call oldEntry.moveTo. Instead,
> throw exception, which will be caught by its caller and causes sentry client
> at name node plugin gets path full snapshot from sentry server.
> 2. Find out why the oldEntry is null and fix it.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
