2:24 p.m.
[
https://jira.jboss.org/jira/browse/JBIDE-4925?page=com.atlassian.jira.plu...
]
Darryl Miles commented on JBIDE-4925:
-------------------------------------
Just got around to look at patch. Again this is a 10000 foot view of it.
The patch looks like it will deal with the "hung UI" part of the issue.
However.... the thread that is pushing the JMX command over the network socket will
continue to be running ? is t.interrupt() enough to ensure it finishes? it may just
retry the network IO method and block again right away.
So the next comment is... can you invalidate the "connection" so that it
effectively closes the socket from underneath the thread trying to push the command, a
"force close of socket", this should make the JMX thread wakeup and
return/cleanup since there is no network socket to retry IO on and the error of a closed
socket is terminal. Can I also ask if the "connection" object is then
invalidated from future use by the publishing tools ? i.e. if a future publish operation
requires a JMX connection (in a future publish operation) it needs to go and create a new
one (i.e. reconnect).
Another issue, you've now added upto an additional 1 second delay to processing, (the
time left on the remaining Thread.sleep(1000)) this really SHOULD wake up immediately if
either the monitor changes or the thread terminates. If anything upon Thread terminating
SHOULD be immediately but the cancellation request CAN have some seconds delay. This
design decision is on the basis of the most frequently used code path, choices impacting
developer productivity and that the cancellation feature is hardly used by an important
feature.
The way out of this is to setup a basic Object lock/wait before thread creation, at exit
of sub-thread signal a wakeup object.interrupt() and use the standard object.sleep()
mechanism instead of Thread.sleep() all with appropriate synchronized locking.
Re the 4 seconds.. maybe this is a bit short... yes most of the time it should happen
really fast but.... it can early take over 4 seconds to page in eclipse + jboss, it can
take eclipse longer than that for the UI to wake up after going for lunch (due to paging
on a 6Gb, dual cpu 64bit system). Also you have now fixed the ability to cancel the
publish so if the wait it too long the user can now cancel and start over. Now people
never use the cancel feature you just added since they never press it before 4 seconds are
up, maybe 10 or 15 is better IMHO.
final Object waitObject = new Object();
// Create thread
// Inside sub-thread code
//.....
} catch (Exception ex) {
exception = ex; // Do this to local storage (not directly to a shared variable)
}
synchronized(waitObject) {
e[0] = exception; // Only assign the shared variable while we hold the lock
waitObject.notifyAll();
}
// end of sub-thread-code
// ...
// start thread.....
// for wait loop
while(t.isAlive() && !monitor.isCanceled() && count <= 4000) {
count+= 1000;
synchronized(waitObject) {
try {
waitObject.sleep(1000);
} catch(InterruptedException eat) {
}
}
}
if( t.isAlive()) {
connection.forceSocketClosureThreadSafeImplementation();
connection.invalidateFromFutureUse(); // this removes this instance of
'connection' from the server tools (and forces the next operation to acquire a new
one)
// Do existing logging error stuff
}
JBossAS Publish, Modal dialog blocking UI (until timeout, not
supporting job cancellation)
------------------------------------------------------------------------------------------
Key: JBIDE-4925
URL: https://jira.jboss.org/jira/browse/JBIDE-4925
Project: Tools (JBoss Tools)
Issue Type: Bug
Components: JBossAS
Affects Versions: 3.1.0.M2
Reporter: Darryl Miles
Assignee: Rob Stryker
Fix For: 3.1.0.M4
Attachments: JBIDE-4925_premature.patch
The tooling can often attempt to publish to JBoss AS but it never completes. Maybe JBoss
AS product failed.
The Eclipse runtime job manager displays the publishing job with a job cancellation
button, JBoss Tooling fails to attempt or process this cancellation request so the UI can
end up hung at a modal "public job in progress" dialog without any availity to
background it. The only way out is to kill the process ID on the system and restart
eclipse.
Here is a stack trace of the only active thread with org.jboss.* in it. It appears to be
waiting for something from a network socket and it appear to be using blocking network I/O
to get it. I'm don't fully understand the JMX stuff myself (but I guess its some
kind of management network channel), in this situation and the situation of publishing
stuff the command it is attempting to send should be given a maximum timeout before the
command (and network connection) is aborted and Eclipse attempt to reconnect to the AS (at
least 1 time). If the JMX command is stateless or the first command that would begin a
stateful interaction (then quickly realizing the JMX connection is no longer working and
reopening the connection to retry the JMX command make sense.
But the main issue here is the fact that the JMX network handling is not being done with
a regard for the Eclipse job monitor, I think this can be effected easily by making the
thread which handles the public register itself just before it begin any unit-of-work that
may block for an unknown about of time. Then the socket I/O is interrupted causing an
exception be thrown (unblocking that thread).
Here is a stack trace of the only active thread with org.jboss.* in it.
"Worker-39" prio=10 tid=0x08c5bc00 nid=0x227a runnable [0xf310b000]
java.lang.Thread.State: RUNNABLE
at java.net.SocketInputStream.socketRead0(Native Method)
at java.net.SocketInputStream.read(SocketInputStream.java:129)
at java.io.BufferedInputStream.fill(BufferedInputStream.java:218)
at java.io.BufferedInputStream.read(BufferedInputStream.java:237)
- locked <0x4b08b708> (a java.io.BufferedInputStream)
at java.io.DataInputStream.readByte(DataInputStream.java:248)
at sun.rmi.transport.StreamRemoteCall.executeCall(StreamRemoteCall.java:195)
at sun.rmi.server.UnicastRef.invoke(UnicastRef.java:142)
at org.jboss.invocation.jrmp.server.JRMPInvoker_Stub.invoke(Unknown Source)
at
org.jboss.invocation.jrmp.interfaces.JRMPInvokerProxy.invoke(JRMPInvokerProxy.java:133)
at
org.jboss.invocation.InvokerInterceptor.invokeInvoker(InvokerInterceptor.java:365)
at org.jboss.invocation.InvokerInterceptor.invoke(InvokerInterceptor.java:197)
at
org.jboss.jmx.connector.invoker.client.InvokerAdaptorClientInterceptor.invoke(InvokerAdaptorClientInterceptor.java:66)
at org.jboss.proxy.SecurityInterceptor.invoke(SecurityInterceptor.java:68)
at
org.jboss.proxy.ClientMethodInterceptor.invoke(ClientMethodInterceptor.java:74)
at org.jboss.proxy.ClientContainer.invoke(ClientContainer.java:101)
at $Proxy4.invoke(Unknown Source)
at
org.jboss.ide.eclipse.as.core.server.internal.JBossServerBehavior.suspendDeployment(JBossServerBehavior.java:235)
at
org.jboss.ide.eclipse.as.core.server.internal.JBossServerBehavior$3.run(JBossServerBehavior.java:194)
at
org.jboss.ide.eclipse.as.core.extensions.jmx.JMXSafeRunner.run(JMXSafeRunner.java:71)
at
org.jboss.ide.eclipse.as.core.extensions.jmx.JMXSafeRunner.run(JMXSafeRunner.java:52)
at
org.jboss.ide.eclipse.as.core.extensions.jmx.JBossServerConnection.run(JBossServerConnection.java:92)
at
org.jboss.ide.eclipse.as.core.extensions.jmx.JBossServerConnectionProvider.run(JBossServerConnectionProvider.java:51)
at
org.jboss.ide.eclipse.as.core.server.internal.JBossServerBehavior.publishStart(JBossServerBehavior.java:198)
at
org.eclipse.wst.server.core.model.ServerBehaviourDelegate.publish(ServerBehaviourDelegate.java:847)
at
org.eclipse.wst.server.core.model.ServerBehaviourDelegate.publish(ServerBehaviourDelegate.java:708)
at org.eclipse.wst.server.core.internal.Server.publishImpl(Server.java:2690)
at org.eclipse.wst.server.core.internal.Server$PublishJob.run(Server.java:272)
at org.eclipse.core.internal.jobs.Worker.run(Worker.java:55)
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
https://jira.jboss.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira