On Thu, Oct 19, 2023 at 03:09:08PM +0200, Silamael Darkomen wrote: > Hi, > > Today I upgraded to the brand new Squid version 6.3 from ports and noticed, > that Squid no longer starts properly if configured with multiple worker > processes. > > After some debugging the limit from net.unix.dgram.sendspace came up as > cause. The 2k default is way to low. > In ktrace I saw sendmessage calls with messages slightly over 4k. > > Increasing this limit to 16k as net.unix.dgram.recvspace fixes the problem > and Squid can start. > > Perhaps this historically low limit should be adjusted accordingly? > All for all other Unix sockets, sendspace and recvspace share the same > limits, just dgram sockets are out of line. > > PS: Tested this with 7.3 but 7.4 seems to have the same limitations. >
It is a SOCK_DGRAM socket, it is supposed to be limited. >From the man-page: A SOCK_DGRAM socket supports datagrams (connectionless, unreliable messages of a fixed (typically small) maximum length). The program should set the socket buffer size via setsockopt() using SO_SNDBUF. It seems squid just YOLOs this and hopes for the best. So the best way to fix this is in squid itself. -- :wq Claudio
