[
https://issues.apache.org/jira/browse/HDFS-16881?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17654142#comment-17654142
]
ASF GitHub Bot commented on HDFS-16881:
---------------------------------------
cnauroth commented on code in PR #5268:
URL: https://github.com/apache/hadoop/pull/5268#discussion_r1060859919
##########
hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/FSDirectory.java:
##########
@@ -388,6 +392,12 @@ public enum DirOp {
DFS_PROTECTED_SUBDIRECTORIES_ENABLE,
DFS_PROTECTED_SUBDIRECTORIES_ENABLE_DEFAULT);
+ final long readLockThresholdMs = conf.getLong(
+ DFS_NAMENODE_READ_LOCK_REPORTING_THRESHOLD_MS_KEY,
+ DFS_NAMENODE_READ_LOCK_REPORTING_THRESHOLD_MS_DEFAULT);
+ // use half of read lock threshold
Review Comment:
Is it necessary for this to use half? If so, can you describe why in this
comment?
##########
hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/namenode/TestFSPermissionChecker.java:
##########
@@ -446,4 +447,20 @@ private static INodeFile createINodeFile(INodeDirectory
parent, String name,
parent.addChild(inodeFile);
return inodeFile;
}
+
+ @Test
+ public void testCheckAccessControlEnforcerSlowness() throws Exception {
+ final long thresholdMs = 10;
+ final String message = FSPermissionChecker.runCheckPermission(() -> {
+ try {
+ Thread.sleep(20);
+ } catch (InterruptedException e) {
+ throw new RuntimeException(e);
Review Comment:
I suggest adding `Thread.currentThread().interrupt();` before throwing. It
shouldn't matter much in practice, but JUnit runner threads have some strange
behavior when interrupted status is not restored as expected.
> Warn if AccessControlEnforcer runs for a long time to check permission
> ----------------------------------------------------------------------
>
> Key: HDFS-16881
> URL: https://issues.apache.org/jira/browse/HDFS-16881
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: namanode
> Reporter: Tsz-wo Sze
> Assignee: Tsz-wo Sze
> Priority: Major
> Labels: pull-request-available
>
> AccessControlEnforcer is configurable. If an external AccessControlEnforcer
> runs for a long time to check permission with the FSnamesystem lock, it will
> significantly slow down the entire Namenode. In the JIRA, we will print a
> WARN message when it happens.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
