[
https://issues.apache.org/jira/browse/HIVE-23164?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Eugene Chung updated HIVE-23164:
--------------------------------
Description:
HiveServer2 which receives the deregister command is at first preparing
shutdown. If there's no remaining session, HiveServer2.stop() is called to shut
down. But I found the case that the HiveServer2 JVM is not terminated even if
HiveServer2.stop() has been called and processed. The case is always occurred
when the local(embedded) metastore is used.
I've attached the full thread dump describing the situation.
[^thread_dump_hiveserver2_is_not_terminated.txt]
In this thread dump, you can see some bunch of 'daemon' threads, NO main
thread, and some 'non-daemon' thread(or user thread)s. As specified by
[https://www.baeldung.com/java-daemon-thread], if there is at least one user
thread exists, JVM does not terminate. (Note that DestroyJavaVM thread is
non-daemon but it's special.)
{code:java}
"pool-8-thread-1" #24 prio=5 os_prio=0 tid=0x00007f52ad1fc000 nid=0x821c
waiting on condition [0x00007f525c500000]
java.lang.Thread.State: TIMED_WAITING (parking)
at sun.misc.Unsafe.park(Native Method)
- parking to wait for <0x00000003cfa057c0> (a
java.util.concurrent.locks.AbstractQueuedSynchronizer$ConditionObject)
at java.util.concurrent.locks.LockSupport.parkNanos(LockSupport.java:215)
at
java.util.concurrent.locks.AbstractQueuedSynchronizer$ConditionObject.awaitNanos(AbstractQueuedSynchronizer.java:2078)
at
java.util.concurrent.ScheduledThreadPoolExecutor$DelayedWorkQueue.take(ScheduledThreadPoolExecutor.java:1093)
at
java.util.concurrent.ScheduledThreadPoolExecutor$DelayedWorkQueue.take(ScheduledThreadPoolExecutor.java:809)
at
java.util.concurrent.ThreadPoolExecutor.getTask(ThreadPoolExecutor.java:1067)
at
java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1127)
at
java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
at java.lang.Thread.run(Thread.java:748)
Locked ownable synchronizers:
- None
{code}
The thread above is created by ScheduledThreadPoolExecutor(int coreSize) with
default ThreadFactory which always creates non-daemon thread. If such thread
pool is not destroyed with ScheduledThreadPoolExecutor.shutdown() method, JVM
cannot terminate! The only way to kill is TERM signal. If JVM receives TERM
signal, it ignores non-daemon threads and terminates.
So I have been digging modules which create ScheduledThreadPoolExecutor with
non-daemon threads and I got it. As you may guess, it's the local(embedded)
metastore. ScheduledThreadPoolExecutor is created by
org.apache.hadoop.hive.metastore.HiveMetaStore.HMSHandler#startAlwaysTaskThreads()
and ScheduledThreadPoolExecutor.shutdown() is never called.
Plus, I found another usage of creating such ScheduledThreadPoolExecutor and
not calling its shutdown.
was:
As you know, HiveServer2 which receives the deregister command is at first
preparing shutdown. If there're no remaining sessions, HiveServer2.stop() is
called to shut down. But I found the case that the HiveServer2 JVM is not
terminated even if HiveServer2.stop() is called and properly processed. The
case is always occurred when the local(embedded) metastore is used.
I've attached the full thread dump describing the situation.
[^thread_dump_hiveserver2_is_not_terminated.txt]
In this thread dump, you can see some bunch of 'daemon' threads, NO main
thread, and some non-daemon thread(or user thread)s. As specified by
[https://www.baeldung.com/java-daemon-thread], if there exists at least one
user thread which is alive, JVM does not terminate. (Note that DestroyJavaVM
thread is non-daemon but it's special.)
{code:java}
"pool-8-thread-1" #24 prio=5 os_prio=0 tid=0x00007f52ad1fc000 nid=0x821c
waiting on condition [0x00007f525c500000]
java.lang.Thread.State: TIMED_WAITING (parking)
at sun.misc.Unsafe.park(Native Method)
- parking to wait for <0x00000003cfa057c0> (a
java.util.concurrent.locks.AbstractQueuedSynchronizer$ConditionObject)
at java.util.concurrent.locks.LockSupport.parkNanos(LockSupport.java:215)
at
java.util.concurrent.locks.AbstractQueuedSynchronizer$ConditionObject.awaitNanos(AbstractQueuedSynchronizer.java:2078)
at
java.util.concurrent.ScheduledThreadPoolExecutor$DelayedWorkQueue.take(ScheduledThreadPoolExecutor.java:1093)
at
java.util.concurrent.ScheduledThreadPoolExecutor$DelayedWorkQueue.take(ScheduledThreadPoolExecutor.java:809)
at
java.util.concurrent.ThreadPoolExecutor.getTask(ThreadPoolExecutor.java:1067)
at
java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1127)
at
java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
at java.lang.Thread.run(Thread.java:748)
Locked ownable synchronizers:
- None
{code}
The thread above is created by ScheduledThreadPoolExecutor(int coreSize) with
default ThreadFactory which always creates non-daemon thread. If such thread
pool is not shut down with ScheduledThreadPoolExecutor.shutdown() method, JVM
cannot terminate! The only way to kill is TERM signal. If JVM receives TERM
signal, it ignores non-daemon threads.
So I have been digging modules which create ScheduledThreadPoolExecutor with
non-daemon threads and now I got it. As you may guess, it's local(embedded)
metastore. It's created by
org.apache.hadoop.hive.metastore.HiveMetaStore.HMSHandler#startAlwaysTaskThreads()
and ScheduledThreadPoolExecutor.shutdown() is never called.
Plus, I found another usage of creating such ScheduledThreadPoolExecutor and
not calling its shutdown.
> server is not properly terminated because of non-daemon threads
> ---------------------------------------------------------------
>
> Key: HIVE-23164
> URL: https://issues.apache.org/jira/browse/HIVE-23164
> Project: Hive
> Issue Type: Bug
> Components: HiveServer2
> Reporter: Eugene Chung
> Assignee: Eugene Chung
> Priority: Major
> Attachments: HIVE-23164.01.patch,
> thread_dump_hiveserver2_is_not_terminated.txt
>
>
> HiveServer2 which receives the deregister command is at first preparing
> shutdown. If there's no remaining session, HiveServer2.stop() is called to
> shut down. But I found the case that the HiveServer2 JVM is not terminated
> even if HiveServer2.stop() has been called and processed. The case is always
> occurred when the local(embedded) metastore is used.
> I've attached the full thread dump describing the situation.
> [^thread_dump_hiveserver2_is_not_terminated.txt]
> In this thread dump, you can see some bunch of 'daemon' threads, NO main
> thread, and some 'non-daemon' thread(or user thread)s. As specified by
> [https://www.baeldung.com/java-daemon-thread], if there is at least one user
> thread exists, JVM does not terminate. (Note that DestroyJavaVM thread is
> non-daemon but it's special.)
>
> {code:java}
> "pool-8-thread-1" #24 prio=5 os_prio=0 tid=0x00007f52ad1fc000 nid=0x821c
> waiting on condition [0x00007f525c500000]
> java.lang.Thread.State: TIMED_WAITING (parking)
> at sun.misc.Unsafe.park(Native Method)
> - parking to wait for <0x00000003cfa057c0> (a
> java.util.concurrent.locks.AbstractQueuedSynchronizer$ConditionObject)
> at java.util.concurrent.locks.LockSupport.parkNanos(LockSupport.java:215)
> at
> java.util.concurrent.locks.AbstractQueuedSynchronizer$ConditionObject.awaitNanos(AbstractQueuedSynchronizer.java:2078)
> at
> java.util.concurrent.ScheduledThreadPoolExecutor$DelayedWorkQueue.take(ScheduledThreadPoolExecutor.java:1093)
> at
> java.util.concurrent.ScheduledThreadPoolExecutor$DelayedWorkQueue.take(ScheduledThreadPoolExecutor.java:809)
> at
> java.util.concurrent.ThreadPoolExecutor.getTask(ThreadPoolExecutor.java:1067)
> at
> java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1127)
> at
> java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
> at java.lang.Thread.run(Thread.java:748)
> Locked ownable synchronizers:
> - None
> {code}
> The thread above is created by ScheduledThreadPoolExecutor(int coreSize) with
> default ThreadFactory which always creates non-daemon thread. If such thread
> pool is not destroyed with ScheduledThreadPoolExecutor.shutdown() method, JVM
> cannot terminate! The only way to kill is TERM signal. If JVM receives TERM
> signal, it ignores non-daemon threads and terminates.
> So I have been digging modules which create ScheduledThreadPoolExecutor with
> non-daemon threads and I got it. As you may guess, it's the local(embedded)
> metastore. ScheduledThreadPoolExecutor is created by
> org.apache.hadoop.hive.metastore.HiveMetaStore.HMSHandler#startAlwaysTaskThreads()
> and ScheduledThreadPoolExecutor.shutdown() is never called.
> Plus, I found another usage of creating such ScheduledThreadPoolExecutor and
> not calling its shutdown.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
