On Jan 10, 2011, at 2:03 PM, Ben Noordhuis wrote: > On Mon, Jan 10, 2011 at 22:22, Neil McKee <neil.mc...@inmon.com> wrote: >> I pushed changes to: >> >> 1. use PIPE_BUF from limits.h (if available). >> 2. use apr_file_pipe_timeout_set(<pipe>, 0) on both ends -- just to make >> absolutely certain that the writes from the critical section are always >> non-blocking. >> 3. use apr_poll() at the read end of the pipe to get my read timeout back. >> 4. make sure any critical-section logging is with APLOG_DEBUG. >> >> http://code.google.com/p/mod-sflow/source/browse/trunk/mod_sflow.c?r=17 > > Much better, Neil. > >> setting the "timeout" to 0 seems to be equivalent to setting O_NONBLOCK? > > Even better, apr_file_pipe_timeout_set() with timeout > -1 sets O_NONBLOCK. >
I think I understand now. I can have: apr_file_pipe_timeout_set(<read_end>, 900000); apr_file_pipe_timeout_set(<write_end>, 0); and remove all that apr_poll() code again: http://code.google.com/p/mod-sflow/source/browse/trunk/mod_sflow.c?r=20 Thanks! Neil >> I don't see any functions for assembling an apr_pollfd_t object > > There indeed aren't, creating them by hand is the (only) way to go.
