[Lxc-users] Cron jobs in containers..
Gordon Henderson
gordon at drogon.net
Thu Jul 8 17:46:20 UTC 2010
Sort of thinking out loud, but looking for suggestions/ideas ...
So I have a host with 20 containers and each of these 'servers' runs cron
jobs - and one of them is to do an MRTG run every 5 minutes, so every 5
minutes, ka-chung, 20 jobs start at the same time which sort of has a bit
of a crowbarring effect...
That's probably the worst example, but there are others - log-file
rotations, locate, updatedb, and so on.. (probably not much impact as
they're run "out of hours" as it were...)
And I'm sure this isn't just an issue with LXC - it must happen in other
virtualisation systems too (KVM, Xen, etc.)
So I'm wondering about getting the host to do some of it which might
serialise it - log-file rotations - it can probably manage, but sending a
signal to syslog in each container? (I guess that's permissible from the
host?)
And I guess there are some things it can't see - e.g. the /proc partition
in each container to read network statistics and so on...
And of-course, maybe it's not that big a problem - I've not actually tried
this with 20 containers yet... It'll still require the same over all
number of cpu cycles whether something is run on the host over all
containers, or run in each container, but hopefully with less CPU
scheduling getting in the way?
Anyway - any thoughts, ideas - even if you think I'm worrying over nothing
are welcome!
Cheers
Gordon
More information about the lxc-users
mailing list