On Wed, 2009-10-14 at 11:42 -0400, Boemker, Tim wrote:
> If HttpUriRequest.abort() is called at about the same time that the
> request completes, it's possible for an aborted connection to be
> returned to the pool. The next time the connection is used,
> HttpClient.execute fails without retrying, throwing this exception:
>
> java.io.IOException: Connection already shutdown
> at
> org.apache.http.impl.conn.DefaultClientConnection.opening(DefaultClientC
> onnection.java:112)
> at
> org.apache.http.impl.conn.DefaultClientConnectionOperator.openConnection
> (DefaultClientConnectionOperator.java:120)
> at
> org.apache.http.impl.conn.AbstractPoolEntry.open(AbstractPoolEntry.java:
> 147)
> at
> org.apache.http.impl.conn.AbstractPooledConnAdapter.open(AbstractPooledC
> onnAdapter.java:101)
> at
> org.apache.http.impl.client.DefaultRequestDirector.execute(DefaultReques
> tDirector.java:381)
> at
> org.apache.http.impl.client.AbstractHttpClient.execute(AbstractHttpClien
> t.java:641)
> at
> org.apache.http.impl.client.AbstractHttpClient.execute(AbstractHttpClien
> t.java:576)
>
> Steps to reproduce:
> 1) Set a breakpoint in ThreadSafeClientConnManager.releaseConnection
> just after "reusable" is set (and found to be true).
> 2) Run to the breakpoint in releaseConnection.
> 3) Call HttpUriRequest.abort.
> 4) Let releaseConnection complete.
>
> When the connection is next used, the exception will be thrown.
>
> Snippet from ThreadSafeClientConnManager:
> public void releaseConnection(ManagedClientConnection conn, long
> validDuration, TimeUnit timeUnit) {
> ...
> boolean reusable = hca.isMarkedReusable();
> if (log.isDebugEnabled()) { //
> breakpoint here
> if (reusable) {
> log.debug("Released connection is reusable.");
> } else {
> log.debug("Released connection is not reusable.");
> }
> }
> hca.detach();
> if (entry != null) {
> connectionPool.freeEntry(entry, reusable, validDuration,
> timeUnit);
> }
> }
> }
>
> I read the contract of ConnectionReleaseTrigger to say that only the
> first call of either method has effect so that, for example, a call to
> abortConnection made after a call to releaseConnection would be ignored,
> but I don't think that's the way AbstractClientConnAdapter is
> implemented; I think that it's possible for both methods to have effect.
>
> Am I looking at this correctly?
>
Yes, you are. Obviously this is a bug that needs fixing. Could you
please raise a JIRA for this defect? A patch would also be quite
welcome.
https://issues.apache.org/jira/browse/HTTPCLIENT
Oleg
> Thanks for your consideration.
>
> Tim
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [email protected]
> For additional commands, e-mail: [email protected]
>
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
