[Wylug-discuss] Rebooting Unix/Linux boxen...

[Nigel Metheringham]

On Thu, 2004-07-29 at 13:43, Robert Speed wrote:
> Can someone settle a dispute for me.
>
> I'm trying to argue why a Unix/Linux box should need a reboot (other
> than for a kernel upgrade.)

Other major upgrades (libc) often suggest a reboot, but its normally
only a kernel update that requires one.

However I do have a standing recommendation that some of our main
service boxes should have a reboot scheduled once a month (as it happens
they never get done so the things get rebooted 6 monthly or so).

This is mainly to get round memory leak and similar problems:-
      * This week I found a box that had been up for 6 months or so had
        a process that monitors the disk controller status which had
        bloated to a little over 0.5GB of memory.  This was not good.
      * Our ipsec boxes periodically do something odd - a reboot every
        few months heads this off.
      * We appear to have a situation where syslog dies when SIGHUPed by
        logrotate.  This happens (say) one time in 200, so an occasional
        reboot helps restart them.
      * After some rather longer uptime the jiffies count wraps - this
        tends to have bad juju, and rebooting is the fastest fix.

> I'm trying to argue that it shouldn't be necessaryu to reboot a Unix/Linux
> box.
>
> So what I suppose I need to know is that what are the implications
> of rebooting a Linux/Unix box on a daily basis (other than perhaps cache
> contents...)

Daily is far too often - it means you tend to force disk fscks every so
often which can take for ever.

I personally think its safer to have a degree of regular reboot
scheduling - however I would ideally do it in the few month category.

	Nigel.
--
[ Nigel Metheringham           Nigel.Metheringham at InTechnology.co.uk ]
[ - Comments in this message are my own and not ITO opinion/policy - ]