Ian Clarke <i.clarke at dynamicblue.com> wrote: > Has anyone noticed things getting quite quiet once their node is removed > from inform.php? > > I am concerned that once nodes are removed from inform.php the > mechanisms we have implemented to ensure that other nodes remain aware > of them (and are thus able to make use of their storage-space) may not > be sufficient - although the only evidence I have for this so-far is > that my node's log was a string of "writing datastore ot disk" messages, > having been quite active yesterday when I reactivated it.
I have the suspicion that this may result from a TCP problem. My node is having this problem too, and when I try to connect to it manually, no ConnectionHandler entry appears in the log. I got suspicious and checked using netstat: longitude.doc.ic.ac.uk% netstat Active Internet connections (w/o servers) Proto Recv-Q Send-Q Local Address Foreign Address State tcp 1 0 longitude.doc.ic.:19114 n43-129.dhnet.ufl.:4112 CLOSE_WAIT tcp 1 0 localhost:19114 localhost:25890 CLOSE_WAIT tcp 0 0 longitude.doc.ic.:25891 cm-24-142-61-146.:19114 ESTABLISHED tcp 0 0 longitude.doc.ic.:19114 cm-24-142-61-146.:62307 ESTABLISHED tcp 40533 0 longitude.doc.ic.:19114 danky-1.dsl.speake:1470 ESTABLISHED tcp 49385 0 longitude.doc.ic.:19114 n43-129.dhnet.ufl.:4117 ESTABLISHED tcp 0 0 longitude.doc.ic.:25897 angry.gangofsnail:19114 ESTABLISHED tcp 0 0 longitude.doc.ic.:19114 147-89.waldenweb.c:4111 ESTABLISHED tcp 1 0 longitude.doc.ic.:25899 danky-1.dsl.speak:19114 CLOSE_WAIT tcp 0 1 longitude.doc.ic.:19114 rm01-24-29-193-145:2093 CLOSE tcp 0 1 longitude.doc.ic.:19114 rm01-24-29-193-145:2093 CLOSE tcp 1 0 longitude.doc.ic.:19114 rm01-24-29-193-145:2093 CLOSE_WAIT tcp 0 0 longitude.doc.ic.:19114 226-132.adsl2.avt:18912 ESTABLISHED tcp 1 0 longitude.doc.ic.:19114 rm01-24-29-193-145:2184 CLOSE_WAIT tcp 0 1 longitude.doc.ic.:19114 cma-d0095.resnet.u:2166 CLOSE tcp 0 1 longitude.doc.ic.:19114 cma-d0095.resnet.u:2166 CLOSE tcp 0 1 longitude.doc.ic.:19114 cma-d0095.resnet.u:2166 CLOSE tcp 1 0 longitude.doc.ic.:19114 cma-d0095.resnet.u:2166 CLOSE_WAIT tcp 1 0 longitude.doc.ic.:19114 rm01-24-29-193-145:2341 CLOSE_WAIT tcp 1 0 longitude.doc.ic.:19114 rm01-24-29-193-145:2416 CLOSE_WAIT tcp 1 0 longitude.doc.ic.:19114 rm01-24-29-193-145:2552 CLOSE_WAIT tcp 1 0 longitude.doc.ic.:19114 rm01-24-29-193-145:2650 CLOSE_WAIT tcp 1 0 longitude.doc.ic.:19114 rm01-24-29-193-145:2688 CLOSE_WAIT tcp 39181 0 longitude.doc.ic.:19114 rm01-24-29-193-145:2809 CLOSE_WAIT tcp 0 29215 longitude.doc.ic.:25908 angry.gangofsnail:19114 ESTABLISHED tcp 1 0 longitude.doc.ic.:19114 rm01-24-29-193-145:2857 CLOSE_WAIT tcp 1 0 longitude.doc.ic.:25933 danky-1.dsl.speak:19114 CLOSE_WAIT tcp 1 0 longitude.doc.ic.:19114 rm01-24-29-193-145:2893 CLOSE_WAIT tcp 0 0 longitude.doc.ic.:26136 ro02-24-29-236-15:19114 ESTABLISHED tcp 0 0 longitude.doc.ic.:26196 nas-32-156.chicag:19114 ESTABLISHED tcp 0 1 longitude.doc.ic.:19114 142.92.129.37:3526 CLOSE tcp 0 1 longitude.doc.ic.:19114 142.92.129.37:3526 CLOSE [...] tcp 106 0 longitude.doc.ic.:19114 chiba.3jane.net:1361 ESTABLISHED tcp 107 0 longitude.doc.ic.:19114 place.org:2839 ESTABLISHED This is just the first page, but you get the idea. I had over 300 connections open! The ESTABLISHED ones I think are not being closed for some reason, the CLOSE_WAIT ones are supposed to be closed but are hanging around. Now I don't know what would cause the the CLOSE_WAIT problem... here's some maybe-relevant hints I found on mud-dev and linux-kernel -- I'm not sure what they mean, though. theo --- forwarded message --- Date: 20 Aug 1998 23:27:36 +0200 >From: Andi Kleen <ak#muc,de> To: Andy Sloane <andude#guildsoftware,com> Cc: linux-kernel#vger,rutgers.edu, ak#muc,de Subject: Re: Sockets permanently in CLOSE_WAIT state. This is a known 2.0 deficiency, it doesn't do thread locking in net/socket.c. You may say that sockets are not really thread-safe in 2.0 This is fixed in 2.1, and I can't reproduce it with 2.1.115. -Andi --- forwarded message --- Subject: Socket stuck in CLOSE_WAIT state >From: Alex Buell <alex.buell at tahallah.demon.co.uk> Date: Mon, 26 Apr 1999 23:30:22 +0100 (GMT) There's a DoS attack going around on the net: this particular attack focuses on IDENT (port 113) and leaves sockets in CLOSE_WAIT state. Here's an example (from netstat -ntu) tcp 34 0 194.222.9.116:113 204.60.30.67:1386 CLOSE_WAIT Seems if there's data in the queue, the socket will never close. This is on 2.2.6. Someone else reported the same thing with 2.2.5, he had a socket in CLOSE_WAIT over 7 days! Cheers, Alex _______________________________________________ Freenet-dev mailing list Freenet-dev at lists.sourceforge.net http://lists.sourceforge.net/mailman/listinfo/freenet-dev
