On Nov 7, 2010, at 11:14 PM, Chuck Remes wrote: > I was running a program that allocated a couple of hundred REQ sockets to > connect to a server with a single REP socket via a QUEUE device. > > I noticed that when I allocated >64 sockets I started getting an assertion > failure. > > Assertion failed: nbytes == sizeof (command_t) (mailbox.cpp:193) > > I examined the code and saw a comment explaining how this should *never* > happen since the command size was already confirmed as being less than the > PIPE_BUF. However, I noted a reference to SNDBUF higher up in that same > function. I'm wondering if I am exhausting a system resource (SNDBUF in this > case) by allocating so many sockets. > > BTW, this shell has a very high file descriptor limit (25k) so I know I > haven't hit that particular limit. > > I can show the code but it's ruby and I know that gives some folks hives. :) > > Just imagine a loop that allocates several hundred or several thousand REQ > sockets that are all connecting to the same QUEUE device endpoint.
FYI, running this same code against 2.0.10 results in a hang. cr _______________________________________________ zeromq-dev mailing list zeromq-dev@lists.zeromq.org http://lists.zeromq.org/mailman/listinfo/zeromq-dev
