"scott.marlowe" <scott.marlowe@ihs.com> writes:
> Well, I've found the syslog facility to be noticeable slower than
> rotatelogs with pgsql doing its own logging. So that's why I do it the
> rotatelogs way. Choice is good.
Other reasons: syslog is rumored to drop messages under sufficiently
heavy load (at least on some platforms); syslog inherently can't capture
all messages that might appear on stderr. For example, on most
platforms a link failure in loading a dynamic library is going to be
reported by the dynamic linker on stderr --- we have no way to reroute
it to syslog. If you're trying to debug a problem like "why doesn't
plperl work", those messages are priceless.
regards, tom lane