[
https://issues.apache.org/jira/browse/AMQ-7362?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Knut Saua Mathiesen updated AMQ-7362:
-------------------------------------
Description:
I'm trying to hunt down why my application, when using a vm broker, will not
shutdown gracefully.
Though I have not found the problem (yet) I might have found a concurrency bug
in the VMTransportServer code:
{code:java}
VMTransport client = new VMTransport(location) {
public void stop() throws Exception {
if (!disposed.get()) {
super.stop();
if (connectionCount.decrementAndGet() == 0 && disposeOnDisconnect) {
VMTransportServer.this.stop();
}
}
};
};{code}
My debugger tells me that I've managed to get connectionCount less than 0 which
does not make any sense.
>From what I can see: If _disposed.get()_ returns _false_ it will call
>_super.stop()_ which does a compareAndSet on dispose:
{code:java}
@Override
public void stop() throws Exception {
// Only need to do this once, all future oneway calls will now
// fail as will any asnyc jobs in the task runner.
if (disposed.compareAndSet(false, true)) {
...
{code}
One could easily imagine that two threads both come to the conclusion that
disposed.get() returns false, while only one of the will be allowed into the
body of VMTransport#stop().
The second thread will still decrement the connectionCount which as far as I
can see is wrong.
I don't see why this couldn't just be:
{code:java}
VMTransport client = new VMTransport(location) {
public void stop() throws Exception {
super.stop();
if (!disposed.get()) {
if (connectionCount.decrementAndGet() == 0 && disposeOnDisconnect) {
VMTransportServer.this.stop();
}
}
};
};
{code}
was:
I'm trying to hunt down why my application, when using a vm broker, will not
shutdown gracefully.
Though I have not found the problem (yet) I might have found a concurrency bug
in the VMTransportServer code:
{code:java}
VMTransport client = new VMTransport(location) {
public void stop() throws Exception {
if (!disposed.get()) {
super.stop();
if (connectionCount.decrementAndGet() == 0 && disposeOnDisconnect) {
VMTransportServer.this.stop();
}
}
};
};{code}
My debugger tells me that I've managed to get connectionCount less than 0 which
does not make any sense.
>From what I can see: If _disposed.get()_ returns _false_ it will call
>_super.stop()_ which does a compareAndSet on dispose:
{code:java}
@Override
public void stop() throws Exception {
// Only need to do this once, all future oneway calls will now
// fail as will any asnyc jobs in the task runner.
if (disposed.compareAndSet(false, true)) {
...
{code}
One could easily imagine that two threads both come to the conclusion that
disposed.get() returns false, while only one of the will be allowed into the
body of VMTransport#stop().
The second thread will still decrement the connectionCount which as far as I
can see is wrong.
> Concurrency bug in closing of VMTransportServer?
> ------------------------------------------------
>
> Key: AMQ-7362
> URL: https://issues.apache.org/jira/browse/AMQ-7362
> Project: ActiveMQ
> Issue Type: Bug
> Affects Versions: 5.15.11
> Reporter: Knut Saua Mathiesen
> Priority: Major
>
> I'm trying to hunt down why my application, when using a vm broker, will not
> shutdown gracefully.
> Though I have not found the problem (yet) I might have found a concurrency
> bug in the VMTransportServer code:
> {code:java}
> VMTransport client = new VMTransport(location) {
> public void stop() throws Exception {
> if (!disposed.get()) {
> super.stop();
> if (connectionCount.decrementAndGet() == 0 &&
> disposeOnDisconnect) {
> VMTransportServer.this.stop();
> }
> }
> };
> };{code}
> My debugger tells me that I've managed to get connectionCount less than 0
> which does not make any sense.
> From what I can see: If _disposed.get()_ returns _false_ it will call
> _super.stop()_ which does a compareAndSet on dispose:
> {code:java}
> @Override
> public void stop() throws Exception {
> // Only need to do this once, all future oneway calls will now
> // fail as will any asnyc jobs in the task runner.
> if (disposed.compareAndSet(false, true)) {
> ...
> {code}
> One could easily imagine that two threads both come to the conclusion that
> disposed.get() returns false, while only one of the will be allowed into the
> body of VMTransport#stop().
> The second thread will still decrement the connectionCount which as far as I
> can see is wrong.
>
> I don't see why this couldn't just be:
> {code:java}
> VMTransport client = new VMTransport(location) {
> public void stop() throws Exception {
> super.stop();
> if (!disposed.get()) {
> if (connectionCount.decrementAndGet() == 0 &&
> disposeOnDisconnect) {
> VMTransportServer.this.stop();
> }
> }
> };
> };
> {code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
