Hmm, #FreeBSD #tor relay I run worked for years without problems, but now it requires severe tweaking of maximum socket buffers size and total number of sockets, otherwise it eats all available kernel buffer memory (mbuf clusters) and effectively makes the VPS unreachable via the network.
I wonder what has changed recently so that problem arise and whether that's Tor or FreeBSD problem...
@feld not really; normally it keeps under <5K mbuf clusters and at ~6K sockets. That can last for hours and I guess that's the expected load. Then at some moment it eats up over 50K clusters in some minutes, and hits the default kernel limit (sockets count still at ~6K). I've bumped kern.ipc.nmbclusters and restricted tor socket buffers with ConstrainedSockets option. Since it died againI guess more tweaks are still needed. Tor ppl says it may be a DDOS attack on #tor: https://trac.torproject.org/projects/tor/ticket/24687
@AMDmi3 Could it be the new scheduler in Tor?
@lattera dunno, when this started tor haven't had been updated for quite some time, and was at 0.3.1.9.
