Re: BUG #8225: logging options don't change after reload

On Jun 13, 2013, at 7:48 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:

> Jeff Frost <jeff(at)pgexperts(dot)com> writes:
>> What I don't understand is the new log file being created from the new
>> log_filename setting but then nothing being logged into it. Is it the
>> postmaster which creates that file? I would've thought it would be
>> the logger process?
>
> Hm, I hadn't focused on that --- that *is* pretty bizarre. The
> postmaster creates the log file initially before forking the syslogger,
> but subsequent rotations are handled by the syslogger process.
>
> Is it possible that your systems are running on the hairy edge of ENFILE
> limits? I notice that the syslogger will silently fail to rotate if it
> gets ENFILE while trying to open the new log file. That doesn't look
> like it'd explain the lack of log_checkpoint activity, though. Also,
> usually people notice this state because everything else on the box
> starts to fall over ...

These are definitely busy systems, but usually not running close to the edge, I'll see if I can make a test case using pgbench on 9.2.4.