On Fri, Aug 22, 2008, Paul Walsh wrote: > Michael Alger wrote: >> On Thu, Aug 21, 2008 at 05:13:22PM +0200, Chris Joelly wrote: >>> On Thu, Aug 21, 2008 at 11:59:15AM +0200, Andrew Beekhof wrote: >>>> On Aug 21, 2008, at 1:01 AM, Chris Joelly wrote: >>>>> i have a question on handling logrotate with services which are >>>>> managed using heartbeat / crm: >>>>> >>>>> if e.g. apache is startet/stopped using the resource manager, >>>>> and logrotate restarts it when logrotate rotates the apache >>>>> logfiles, how would crm respond to this issue? is crm able to >>>>> handle thios correct or >>>>> is crm seeing a resource failure and then not be able to start >>>>> apache again as it is stopped and then restartet outside of crm? >>>> Its entirely up to the apache resource agent. >>>> If it reports the apache failed, then we'll restart it. >>> ok, i see i messed things up... >>> but what is the best practice to handle logrotate triggered >>> restarts of services (e.g. apache) which forces the cluster >>> software to restart the service using the resource agent? >>> >>> is the way to go to use some sort of database logging (using >>> mod_log_sql for apache) or for other services use syslog-ng with >>> output to databases and disabling logrotate on this services? >>> >>> or is there another way staying with the textfile based logging? >> >> Quite intriguing. Most services don't require a restart in order to >> rotate logs. However, having done a bit of a quick search, it appears >> this likely is the case for Apache 1.x. It looks like a common solution >> is to have Apache write its logs via a separate process which handles >> rotation without having to restart the server: >> >> http://httpd.apache.org/docs/1.3/programs/rotatelogs.html >> >> This may contain some useful information as well: >> >> http://www.freebsddiary.org/rotatelogs.php >> > > We've tended to configure Apache to log to /some/where/access.log and > /some/where/error.log with those being symlinks to > /some/where/access_log.YYMM and /some/where/error_log.YYMM respectively. > A script is then run at 00:00 on the 1st of each month which stops > Apache, removes the symlink, creates new empty access_log.YYMM and > error_log.YYMM files, re-creates the symlink then restarts Apache. > > What I haven't done (yet) is implement the same on a heartbeat cluster. > I'm planning to achieve this (see the thread "use of crm_resource" 15 Jul > - 18 July) by having the script issue the command > > crm_resource -r Apache --meta -p target_role -v stopped > > before switching the log files over then > > crm_resource -r Apache --meta -p target_role -v started > > afterwards.
I think i use apacht2ctl graceful after logrotate rotates the Apache log files. I saw that newer distributions (e.g. Debian Lenny) uses apache2ctl graceful, so i'll use this approach on Debian Etch too ... Samba e.g. uses SIGHUP to force its deamons (smbd, nmbd) to reload the config and switch over to new logfiles, at least in Etch and Lenny too. So i hope that other services has a similar approach that doesn't need to bring down the service. Thanks 2 all Chris _______________________________________________ Linux-HA mailing list [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
