Re: Niced cron jobs
Craig Sanders <cas@taz.net.au> writes:
> niced cron jobs [can be] disastrous on heavily loaded servers that
> never have any idle time to service the niced jobs.... this is not
> just speculation, it's experience. i've seen it happen because i've
> tried it myself.
Very interesting real world info. (Of course, NOT using nice still
doesn't guarantee that jobs will complete within (even lengthy)
realtime constraints.)
> most cron jobs are written with the assumption that only one
> instance will be running at a time, which can cause locking or other
> contention problems
Such cron jobs are arguably buggy.
Well, I don't have a firm opinion on this issue, but here are a few
points to consider:
- Preventing multiple jobs can be done by using Anacron instead of
cron.
- If we use a default nice level for cron (and Anacron jobs), the
local admin can, of course, change it. Specifying the default nice
level in a single place would make changing it easier.
- Different kernels schedule tasks differently. So, e.g., nicing
something by 10 levels results in slightly different behavior on a
Linux 2.0 kernel vs. a 2.2 SMP kernel. (And what about non-Linux
Debian systems?) These differences should ordinarily be
inconsequential, but folks with very heavily loaded systems like
Mr. Sanders could be affected by slight differences in the meaning
of priority levels.
-ccwf
Reply to: