Date: Mon, 14 May 2007 11:29:12 +0100 (BST) From: Robert Watson <rwatson@FreeBSD.org> To: "Marc G. Fournier" <freebsd@hub.org> Cc: stable@FreeBSD.org Subject: Re: UNIX domain sockets MFC's Message-ID: <20070514112716.C24765@fledge.watson.org> In-Reply-To: <FF516AF813E525D10C4A79B4@ganymede.hub.org> References: <20070508132149.A24765@fledge.watson.org> <20070511124830.N24765@fledge.watson.org> <FF516AF813E525D10C4A79B4@ganymede.hub.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, 12 May 2007, Marc G. Fournier wrote: >> The fix for this has now been merged as 1.155.2.22. As there have been no >> new reports of UNIX domain socket problems in the last couple of days, it >> sounds like the MFC of the last batch of fixes and cleanups has not lead to >> problems. > > I've just upgraded my kernel to the latest, to include the MFC'd code above > ... Yes -- I was very specific in my e-mail regarding the MFC's that they were not believed to address the problem you are reporting. I think we have a leak in the way some edge case is handled with regard to UNIX domain socket shutdown. What would be really nice to know is if that persists in 7-CURRENT, in which we've redone the way the socket life cycle works. However, I don't know if you are able to tolerate booting a 7-CURRENT kernel in your environment...? Did we determine whether backing out to before the unpcb socket reference count change made any difference for you? Robert N M Watson Computer Laboratory University of Cambridge > > Just before rebooted, as I've done the past couple of times, I shutdown > everything on the server, so that there were minimal processes running ... > based on the last one, and this one, it looks like the number of "Active > open sockets" is ~4000 ... last time, I was up to 11k sockets open, and it > drop'd to ~7000 once all jails were shut down, but, as reported to > Robert/John, there was a java process in a soclose state, so I wasn't 100% > certain there ... > > This time through, I started at about 8800 sockets open, and shut down all > processes, including all java processes ... using ps auxlw, I checked for any > processes in a soclose state, and there were none ... I waited a full 10 > minutes to let things 'settle', and after 7 of those, it had drop'd down to: > > mars# uptime ; sysctl kern.ipc | grep sock > 2:18PM up 1 day, 13:26, 5 users, load averages: 0.00, 0.47, 2.57 > kern.ipc.maxsockbuf: 262144 > kern.ipc.sockbuf_waste_factor: 8 > kern.ipc.numopensockets: 4835 > kern.ipc.maxsockets: 12328 > > And stuck there for the remaining 3 minutes before I rebooted ... which is what > leads me to believe that there are about 4000 active sockets on this server > when everything is running ... > > > - ---- > Marc G. Fournier Hub.Org Networking Services (http://www.hub.org) > Email . scrappy@hub.org MSN . scrappy@hub.org > Yahoo . yscrappy Skype: hub.org ICQ . 7615664 > -----BEGIN PGP SIGNATURE----- > Version: GnuPG v1.4.5 (FreeBSD) > > iD8DBQFGRfpC4QvfyHIvDvMRAuzoAKDbb5Fndwtw8paTsmLdXIP+FrOBHQCeIVKf > Uhlv8ZRAjVar/fRHD3E6waM= > =yglM > -----END PGP SIGNATURE----- > >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20070514112716.C24765>