Re: logging output of a cronjob

(Note: reply-to set to evade the autoresponders).
(Note note: my opinions only; if  you don't like 'em, ignore 'em).

Otavio Exel wrote:

> I've never heard of someone using multilog this way;
> any pitfalls that I should be aware of?

Yep, don't run more than one at once (as you already noted).  I do the
same thing for running postgresql vacuum outputs at a verbose level,
which produces lots of output of which I'm mostly not interested in
until something goes wrong.

> is there a recommended way to prevent 2+ instances of multilog from
> using the same dir?

I use setlock (also daemontools?  It's djbware anyway) to ensure only
once instance of the script is running, either explicitly or implicitly
through a wrapper script.

> is there a way to force multilog to wait for /var/log/xyw/lock instead
> of exiting with code 111?

Should be done before multilog runs otherwise other processes that
make up the pipeline will start and then be aborted (e.g. the postgres
vacuum process mentioned above will be started then terminated).  Best
to ensure that you *have* your resources (e.g. lock) before attempting
the work.

You can also run it as a supervised job with the down flag touched -
just svc -o to start it and allow it to run until completion.  This
is OK if you don't mind cluttering up your process table with supervise
entries (I initially did this, then went the first route because of
this).

> someone suggested (many years ago) supervising a script like that...
> 	#!/bin/sh
> 	test -p my-pipe || mkfifo my-pipe
> 	exec cat my-pipe
> ...and sending to-be-logged-data to my-pipe; seems to me to be a good
> idea (although a bit too complex);
> 
> any comments?

That may not stop your script from running concurrently, which I'm
not sure is what you're wanting to do (I don't remember if a named
fifo can be opened exclusively for writing - a quick test here says
"no").


-- 
Regards,
  Daryl Tester

"We are sexy, sexy Von Neumann machines."  -- http://www.xkcd.org/387/