Squid<->Pinger communication (Squid 2.2STABLE5)

I'm running Squid-2.2STABLE5 on BSD/OS, and have a problem with it. On
BSD/OS, IPC_DGRAM is implemented using socketpair(2). BSD/OS has a pretty
small default SO_RCVBUF and SO_SNDBUF on socketpairs, so I had to throw in
a setsockopt(2) to increase those. But even bumping them to 64K, the
pinger occasionally dies because it gets an ENOBUF. Of course, I can bump
them to even higher limits, but I'm wondering why Squid communicates the
entire 8K packet to the pinger, rather than just letting the pinger take
care of it.

I noticed the tip of the CVS Squid got rid of the use of socketpair()
altogether, but it still seems to communicate the entire payload
between Squid and the pinger, increasing the risk of running out of
network resources.

Would it make sense to delegate allocating the payload to the pinger
process?

Cheers,

                                        -- Bert

Bert Driehuis, MIS -- bert_driehuis@nl.compuware.com -- +31-20-3116119
Dihydrogen Monoxide kills! Join the campaign at http://www.dhmo.org/
Received on Sun Jun 18 2000 - 15:56:39 MDT