On Thu, 2007-02-08 at 12:08 -0500, Mark Stosberg wrote: > Dan Falconer wrote: > > On Wednesday 07 February 2007 2:23 pm, Mark Stosberg wrote: > >> Dan Falconer wrote: > >>> On Wednesday 07 February 2007 1:00 pm, Mark Stosberg wrote: > >>>> Here's a report and a question on log rotation using the altperl tools, > >>>> on FreeBSD. > >>> [ --xxxx-- snip --xxxx--] > >>> > >>>> So I tried the "APACHE_ROTATOR" method, which the altperl tools have > >>>> built-in support for. > >>>> > >>>> That started creating new log files every 10 seconds, which also seemed > >>>> like a bad idea! I created this patch to slon-tools.pm to only create > >>>> one log file per day: > >>> [ --xxxx-- snip --xxxx--] > >>> > >>>> Even then, doesn't that leave me with a pruning job to do? I don't see > >>>> the value of including the time stamp in the log file name. That just > >>>> seems to complicate log file rotation and pruning, and is non-standard > >>>> from what I've seen. > >>> I encountered this in Slony 1.2.6 as well: in 1.1.0, there was a > >>> configuration parameter, "$ROTATE_TIME", which was fed to the apache > >>> rotator instead of the arbitrary "10M". Ours was set to 86400 (every 24 > >>> hours). I was assuming it was an oops, but I removed the APACHE_ROTATOR > >>> variable out of my configuration files so it wouldn't even try to rotate > >>> them... which of course means I have to manually restart slony every once > >>> in a while to avoid running out of space. > >> Does it make more sense then to prune out the timestamp stuff out of the > >> APACHE_ROTATOR line, and add the "86400" value back in? > >> Then, would it still be necessary to purge logs every so often? > >> > >> Mark > > > > The timestamp thing *REALLY* pisses me off. I would at worst having a > > logname like "$dbname-$dateItStarted.log", at best having "$dbname.log" > > which > > is rotated using the standard "$dbname.log.1", "$dbname.log.2", etc... > > considering the size of the logs on my system (350 megs after a few days), > > I > > would prefer them getting zipped, too... but I can do that myself if it > > can't > > be worked in. > > Could Slony be modified to "let go" of the existing log file handle > when an HUP is received, like Apache does? That would allow the > standard features of newsyslog to be used, and would be a lot easier to > manage. > > (For example, I don't have Apache on my database services, so I had to > copy 'rotatelogs' from another machine... ) > > Now I'm using rotatelogs, plus newsyslog for compression, and I've got: > > db_2007-02-07.log > db_2007-02-07.log.0.bz2 > db_2007-02-08.log > db_2007-02-08.log.0.bz2 > > Including even just the date in the name still messes up standard log > rotation practices. > > Mark > As of this release (1.2.6) I am finally able to use syslog for logging purposes which makes rotation a snap. The relevant files and lines:
***from slon.conf: syslog 2 syslog_facility 'LOCAL1' log_level 1 log_timestamp 0 log_pid 1 ***from syslog.conf *.notice;authpriv.none;kern.debug;lpr.info;mail.crit;news.err;LOCAL1.none /var/log/messages LOCAL1.* /var/log/slony.log ***from newsyslog.conf /var/log/slony.log 640 7 * @T00 JC This on a FreeBSD box. I say "finally" because prior to this release I had issues with worker threads not being found and an ever-increasing number of cleanup events per time period. (see my posts from Nov 30 2005 entitled: "[Slony1-general] Worker threads not seen by slon process" Sven _______________________________________________ Slony1-general mailing list [email protected] http://gborg.postgresql.org/mailman/listinfo/slony1-general
