Re: sysklogd steps on logrotate's foot
- To: KELEMEN Peter <email@example.com>
- Cc: firstname.lastname@example.org
- Subject: Re: sysklogd steps on logrotate's foot
- From: Jaakko Niemi <email@example.com>
- Date: Sat, 02 Feb 2002 17:07:50 +0200
- Message-id: <[🔎] firstname.lastname@example.org>
- In-reply-to: <20020130172158.GB17308@chiara.elte.hu> (KELEMEN Peter's message of "Wed, 30 Jan 2002 18:21:59 +0100")
- References: <20020127181827.GA14749@chiara.elte.hu> <20020127213918.GA23710@finlandia.infodrom.north.de> <20020130172158.GB17308@chiara.elte.hu>
KELEMEN Peter <email@example.com> writes:
> * Martin Schulze (firstname.lastname@example.org) [20020127 22:39]:
>> Err... why is logrotate touching logfiles "owned" by sysklogd?
>> A package must not touch other packages' files.
> What I meant was if an administrator of a busy site applies log
> rotating policies to logfiles, she wants to use one product to
> do the rotation. So, she customises logrotate.conf according to
> her policy, only to find out that sysklogd rotates logs on its
> own and blows up log archiving by messing up serial numbers and
> What she'll miss is an easy way of telling sysklogd: "hands off
> pal, I will do it myself" with regards to log rotation.
This does not solve the problem with sysklogd, but as an alternative,
I use syslog-ng and have it logging into $facility/$year/$month/$day/foo.log
under /var/log. This gives a neat directory structure, and a cronjob
compresses all files in it that have not been touched in a couple days.
However one should locate the directory tree to somewhere else, as some
packages just assume for example /var/log/messages existing and being a file.