[ 
https://issues.apache.org/jira/browse/POOL-315?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Buğra Gedik updated POOL-315:
-----------------------------
    Description: 
The {{close}} method is implemented as follows:

{code}
public void close() {
        if(!this.isClosed()) {
            Object var1 = this.closeLock;
            synchronized(this.closeLock) {
                if(!this.isClosed()) {
                    this.startEvictor(-1L);
                    this.closed = true;
                    this.clear();
                    this.jmxUnregister();
                    this.idleObjects.interuptTakeWaiters();
                }
            }
        }
    }
{code}

The line {{this.startEvictor(-1L);}} calls 
{{EvictionTimer.cancel(this.evictor);}} from {{BaseGenericObjectPool}} and that 
in turn calls the following method in {{EvictionTimer}}:

{code}
    static synchronized void cancel(TimerTask task) {
        task.cancel();
        --_usageCount;
        if(_usageCount == 0) {
            _timer.cancel();
            _timer = null;
        }
    }
{code}

Here, {{_timer}} is a {{java.util.TimerTask}}. If you look at the documentation 
of it, you'll see that it does not block on the currently executing task. Even 
though {{task.cancel()}} is called, there is no code to wait for it to 
complete. The end result is that, even if you close the pool, there may still 
be an eviction thread running. Yes, it will eventually go away, but in some 
rare cases it takes a bit of time to go away.

When running code under Tomcat, this results in the 
{{"commons-pool-EvictionTimer"}} thread (created in the class 
{{EvictionTimer}}) to be reported as leaking, despite the pool being closed. 
This happens rarely, since most of the time the timer thread goes away before 
Tomcat checks for leaking threads.

In my opinion a fix can be put into {{startEvictor}} in 
{{BaseGenericObjectPool}}:

{code}
    final void startEvictor(long delay) {
        synchronized (evictionLock) {
            if (null != evictor) {
                EvictionTimer.cancel(evictor);
                // HERE: evictor.waitForCompletion();
                evictor = null;
                evictionIterator = null;
            }
            if (delay > 0) {
                evictor = new Evictor();
                EvictionTimer.schedule(evictor, delay, delay);
            }
        }
    }
{code}



  was:
The {{close}} method is implemented as follows:

{code}
public void close() {
        if(!this.isClosed()) {
            Object var1 = this.closeLock;
            synchronized(this.closeLock) {
                if(!this.isClosed()) {
                    this.startEvictor(-1L);
                    this.closed = true;
                    this.clear();
                    this.jmxUnregister();
                    this.idleObjects.interuptTakeWaiters();
                }
            }
        }
    }
{code}

The line {{this.startEvictor(-1L);}} calls 
{{EvictionTimer.cancel(this.evictor);}} from {{BaseGenericObjectPool}} and that 
in turn calls the following method in {{EvictionTimer}}:

{code}
    static synchronized void cancel(TimerTask task) {
        task.cancel();
        --_usageCount;
        if(_usageCount == 0) {
            _timer.cancel();
            _timer = null;
        }
    }
{code}

Here, {{_timer}} is a {{java.util.TimerTask}}. If you look at the documentation 
of it, you'll see that it does not block on the currently executing task. Even 
though {{task.cancel()}} is called, there is no code to wait for it to 
complete. The end result is that, even if you close the pool, there may still 
be an eviction thread running. Yes, it will eventually go away, but in some 
rare cases it takes a bit of time to go away.

When running code under Tomcat, this results in the 
{{"commons-pool-EvictionTimer"}} thread (created in the class 
{{EvictionTimer}}) to be reported as leaking, despite the pool being closed. 
This happens rarely, since most of the time the timer thread goes away before 
Tomcat checks for leaking threads.







> GenericObjectPool close() does not wait for the current eviction task
> ---------------------------------------------------------------------
>
>                 Key: POOL-315
>                 URL: https://issues.apache.org/jira/browse/POOL-315
>             Project: Commons Pool
>          Issue Type: Bug
>            Reporter: Buğra Gedik
>
> The {{close}} method is implemented as follows:
> {code}
> public void close() {
>         if(!this.isClosed()) {
>             Object var1 = this.closeLock;
>             synchronized(this.closeLock) {
>                 if(!this.isClosed()) {
>                     this.startEvictor(-1L);
>                     this.closed = true;
>                     this.clear();
>                     this.jmxUnregister();
>                     this.idleObjects.interuptTakeWaiters();
>                 }
>             }
>         }
>     }
> {code}
> The line {{this.startEvictor(-1L);}} calls 
> {{EvictionTimer.cancel(this.evictor);}} from {{BaseGenericObjectPool}} and 
> that in turn calls the following method in {{EvictionTimer}}:
> {code}
>     static synchronized void cancel(TimerTask task) {
>         task.cancel();
>         --_usageCount;
>         if(_usageCount == 0) {
>             _timer.cancel();
>             _timer = null;
>         }
>     }
> {code}
> Here, {{_timer}} is a {{java.util.TimerTask}}. If you look at the 
> documentation of it, you'll see that it does not block on the currently 
> executing task. Even though {{task.cancel()}} is called, there is no code to 
> wait for it to complete. The end result is that, even if you close the pool, 
> there may still be an eviction thread running. Yes, it will eventually go 
> away, but in some rare cases it takes a bit of time to go away.
> When running code under Tomcat, this results in the 
> {{"commons-pool-EvictionTimer"}} thread (created in the class 
> {{EvictionTimer}}) to be reported as leaking, despite the pool being closed. 
> This happens rarely, since most of the time the timer thread goes away before 
> Tomcat checks for leaking threads.
> In my opinion a fix can be put into {{startEvictor}} in 
> {{BaseGenericObjectPool}}:
> {code}
>     final void startEvictor(long delay) {
>         synchronized (evictionLock) {
>             if (null != evictor) {
>                 EvictionTimer.cancel(evictor);
>                 // HERE: evictor.waitForCompletion();
>                 evictor = null;
>                 evictionIterator = null;
>             }
>             if (delay > 0) {
>                 evictor = new Evictor();
>                 EvictionTimer.schedule(evictor, delay, delay);
>             }
>         }
>     }
> {code}



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to