On Sat, 30 May 2009 22:14:32 +0100, Matthew Toseland wrote: > On Sunday 10 May 2009 15:51:39 Dennis Nezic wrote: > > On Wed, 15 Apr 2009 12:58:48 -0400, Dennis Nezic wrote: > > > I have seen a couple times my node stall due to exceeding the > > > thread limit, caused by an absurd 333 "HTTP socket > > > handlers" (86.7% of thread usage). (My node's thread limit is > > > 300). > > > > > > I believe this is caused by my browser (dillo) making too many > > > connections to fproxy. But, nevertheless, shouldn't these > > > threads / sockets die after a few minutes? They were all still > > > alive after more than 10mins of stalling my node, after which I > > > had to restart to get things flowing again. > > > > The problem still exists in 1209. (If my Dillo browser opens a page > > with hundreds of thumbnails and no limit to the number of > > connections made to the server/fproxy (It probably relies on http > > server response codes?), the number of HTTP socket handlers will > > sky-rocket above the limit, and stay there. Again, how long are > > these threads supposed to last for and why are they even allowed to > > exist (can't fproxy throttle the number of incoming connections?). > > There still do remain a couple "Pooled threads awaiting work", but > > the node effectively shuts down -- cpu drops to almost nothing, and > > all traffic seems to stop. > > Does it happen in 1212/1213? Are you sure that all these connections > have fetched data and are no longer active? A thread dump of the node > in this condition would be helpful.
Still happens with the latest builds. Loading a page with lots of images, like: u...@qd-hk0vhyg7yvk2bqsjmcud5qsf0tdkgnnf6lnwuh0g,xTFOV9ddCQQk6vQ6G~jfL6IzRUgmfMcZJ6nuySu~NUc,AQACAAE/activelink-index/106/ ... in a browser that apparently doesn't limit the number of connections that it makes to fproxy, causes the number of threads to skyrocket well above the thread limit, stalling all traffic with the node -- although the fproxy interface still works. I did a thread dump, which includes all the hundreds of HTTP socket handler threads, which are all in a WAITING state. All "locked". Here's the full thread dump: http://dennisn.dyndns.org/guest/uploads/tmp/stalled-httpsocket-dump.log.bz2 Either way, how is the thread limit even able to be exceeded? And why aren't these threads dying? _______________________________________________ Support mailing list Support@freenetproject.org http://news.gmane.org/gmane.network.freenet.support Unsubscribe at http://emu.freenetproject.org/cgi-bin/mailman/listinfo/support Or mailto:support-requ...@freenetproject.org?subject=unsubscribe