]
Petr H commented on WFLY-10621:
-------------------------------
It isn't necessarily related to shared sessions, but they may be one of the possible
triggers that lead to the use of the DistributableSessionManager where the problematic
code lies.
Our case is:
- non-HA profile
- single server in the group
- 4 web applications depoyed
- all web modules marked as distributable (in web.xml) - this is likely another trigger
that, similarly to shared sessions, causes that the DistributableSessionManager is put
into the action
- no shared sessions at all
But generally the StampedLock.writeLockInterruptibly() method is able to block forever
which shouldn't be probably allowed during the server stop operation. Sure, it would
be better to find out what does prevent the DistributableSessionManager from getting that
lock, but since it isn't the kind of lock directly visible in thread dumps, I fear
some debugging would be necessary.
I may still try the latest WildFly (13 atm.), but that will take some time. The issue
doesn't even occur consistently (always) so it isn't so straightforward to
reproduce.
Server stop indefinitely blocked by DistributableSessionManager in
lifecycleLock.writeLockInterruptibly()
---------------------------------------------------------------------------------------------------------
Key: WFLY-10621
URL:
https://issues.jboss.org/browse/WFLY-10621
Project: WildFly
Issue Type: Bug
Components: Clustering
Affects Versions: 11.0.0.Final
Environment: CentOS 7.4.1708
Java 1.8.0_172-b11
WildFly 11.0.0.Final
Reporter: Petr H
Assignee: Paul Ferraro
Labels: clustering, sessions
Attachments: threaddump-wfly11-stop-blocked.txt
Note: This is report for the issue reported here first
https://developer.jboss.org/thread/277883
It can be seen at WildFly 11 and EAP 7.1 at least, however as the particular code in the
DistributableSessionManager class seems to be unchanged even in WildFly 13 it's likely
to occur there as well. (We test WildFly 11 because it's closest to the EAP 7.1 which
is being considered for the production use later in this project)
Stop server operation may hang indefinitely due to the thread such as the following one
being blocked:
{noformat}
"ServerService Thread Pool -- 24" #49 prio=5 os_prio=0 tid=0x00000000035d1000
nid=0x4de1 waiting on condition [0x00007f4bddc46000]
java.lang.Thread.State: WAITING (parking)
at sun.misc.Unsafe.park(Native Method)
- parking to wait for <0x00000000c4e61998> (a
java.util.concurrent.locks.StampedLock)
at java.util.concurrent.locks.StampedLock.acquireWrite(StampedLock.java:1119)
at
java.util.concurrent.locks.StampedLock.writeLockInterruptibly(StampedLock.java:413)
at
org.wildfly.clustering.web.undertow.session.DistributableSessionManager.stop(DistributableSessionManager.java:90)
- locked <0x00000000c4e60090> (a
org.wildfly.clustering.web.undertow.session.DistributableSessionManager)
at
io.undertow.servlet.core.DeploymentManagerImpl$3.call(DeploymentManagerImpl.java:599)
at
io.undertow.servlet.core.DeploymentManagerImpl$3.call(DeploymentManagerImpl.java:589)
at
io.undertow.servlet.core.ServletRequestContextThreadSetupAction$1.call(ServletRequestContextThreadSetupAction.java:42)
at
io.undertow.servlet.core.ContextClassLoaderSetupAction$1.call(ContextClassLoaderSetupAction.java:43)
at
org.wildfly.extension.undertow.security.SecurityContextThreadSetupAction.lambda$create$0(SecurityContextThreadSetupAction.java:105)
at
org.wildfly.extension.undertow.security.SecurityContextThreadSetupAction$$Lambda$1005/749195997.call(Unknown
Source)
at
org.wildfly.extension.undertow.deployment.UndertowDeploymentInfoService$UndertowThreadSetupAction.lambda$create$0(UndertowDeploymentInfoService.java:1508)
at
org.wildfly.extension.undertow.deployment.UndertowDeploymentInfoService$UndertowThreadSetupAction$$Lambda$1006/1165195653.call(Unknown
Source)
at
org.wildfly.extension.undertow.deployment.UndertowDeploymentInfoService$UndertowThreadSetupAction.lambda$create$0(UndertowDeploymentInfoService.java:1508)
at
org.wildfly.extension.undertow.deployment.UndertowDeploymentInfoService$UndertowThreadSetupAction$$Lambda$1006/1165195653.call(Unknown
Source)
at
org.wildfly.extension.undertow.deployment.UndertowDeploymentInfoService$UndertowThreadSetupAction.lambda$create$0(UndertowDeploymentInfoService.java:1508)
at
org.wildfly.extension.undertow.deployment.UndertowDeploymentInfoService$UndertowThreadSetupAction$$Lambda$1006/1165195653.call(Unknown
Source)
at
org.wildfly.extension.undertow.deployment.UndertowDeploymentInfoService$UndertowThreadSetupAction.lambda$create$0(UndertowDeploymentInfoService.java:1508)
at
org.wildfly.extension.undertow.deployment.UndertowDeploymentInfoService$UndertowThreadSetupAction$$Lambda$1006/1165195653.call(Unknown
Source)
at
io.undertow.servlet.core.DeploymentManagerImpl.stop(DeploymentManagerImpl.java:603)
at
org.wildfly.extension.undertow.deployment.UndertowDeploymentService.stopContext(UndertowDeploymentService.java:139)
at
org.wildfly.extension.undertow.deployment.UndertowDeploymentService$2.run(UndertowDeploymentService.java:120)
at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
at java.util.concurrent.FutureTask.run(FutureTask.java:266)
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)
at org.jboss.threads.JBossThread.run(JBossThread.java:320)
{noformat}
The related code in the DistributableSessionManager makes use of the
[
java.util.concurrent.locks.StampedLock.writeLockInterruptibly()|https://d...]
method which can theoretically block indefinitely and thus prevent the server stop from
ever completing.
The change was introduced in the following commit [WFLY-7695 Block session manager stop
until all sessions are
closed|https://github.com/wildfly/wildfly/commit/94863deeb96c4536ec5e8266...]
This can occur even in non-clustered environments (non-HA profile) with web applications
just being marked as distributable (web.xml).
We don't know what prevents the lock from being obtained (unable to determine from
the thread dump) so we can't unfortunately do much about it at the moment from the
application side.
We only tried to change the infinispan web local-cache configuration due to _ISPN000136_
errors intermittedly occurring, but that seems to have no effect. We tried various
combinations of *locking isolation* (REPEATABLE_READ, READ_COMMITED), *transaction
locking* (OPTIMISTIC, PESSIMISTIC) and *transaction mode* (NONE, BATCH) settings and it
seems to occur with all of them.
Note: The issue doesn't occur always, only in some cases.