A quick follow-up: Further testing shows this same hanging thread/running-out-of-connections problem happens with successful (200) GET requests, even if we call entity.exhaust() or clientResource.release(), both of which seem like likely candidates for successfully closing down the connections. Also, I notice that there is something called HttpIdleConnectionReaper, which looks like it's supposed to clean up idle connections, but even when we wait a while (> 15 seconds) between requests we still end up with all the connections in use and subsequent requests hanging.
So far, the only thing that has worked, and _only_ on successful (200) requests, is to call entity.getStream().close(). Even calling entity.getText() does not release the connection- I guess something needs to close the inputstream... -Dave Fogel On Mon, Jul 12, 2010 at 4:43 PM, David Fogel <[email protected]> wrote: > Hi all- > > We've started using Restlet's HttpClient extension (along with Apache > HttpClient 4.01). > > We've hit what seems like a pretty big bug: when we make a client > request using the ClientResource API, if we GET a resource that does > not exist, that process seems to consume and never release an > HttpClient connection object. After GETing a few times with 404 > responses, all subsequent requests to that host will block forever. > Since the connection objects are pooled, and the default is 2, we are > seeing a hung system reproducibly on the 3rd attempt. > > I believe something like the following code will trigger this condition: > > ClientResource resource = new > ClientResource("http://www.restlet.org/DoesNotExist.html";); > Representation entity = null; > > // first time > try { > entity = resource.get(); > } catch (ResourceException re) { > re.printStackTrace(); // <---- this will show a 404 Status > } > //second time > try { > entity = resource.get(); > } catch (ResourceException re) { > re.printStackTrace(); // <---- this will show a 404 Status > } > // third time > try { > entity = resource.get(); // <---- this will hang forever! > } catch (ResourceException re) { > re.printStackTrace(); > } > > (Obviously our code is not actually like this- we would have these > multiple client GET requests triggered by use of our web app's > resources, which are in turn fetching some (often absent) data from > another server.) > > Since we don't have an entity in a response, we can't call > representation.exhaust() on it, and when we add finally() clauses to > the above which call resource.release(), nothing is different. (I'm > not clear at all under what circumstances we're supposed to call > ClientResource.release(), but it didn't help.) > > The problem goes away when we turn off the org.restlet.ext.httpclient > extension and use the built-in client connectors. We'd prefer to use > HttpClient, since we have a dependency on it anyhow from other code in > our system. (Is the built-in client connector production-ready? Will > it scale up to many concurrent requests?) > > Any suggestions? > > Thanks, > -Dave Fogel > ------------------------------------------------------ http://restlet.tigris.org/ds/viewMessage.do?dsForumId=4447&dsMessageId=2632456
