[OpenAFS] Re: Bos server troubles with Suse 9.0

[Jeffrey Hutzelman]

On Monday, January 24, 2005 13:32:33 -0800 Russ Allbery <rra@stanford.edu> 
wrote:

> ted creedon <tcreedon@easystreet.com> writes:
>
>> The restart was not set up intentionally. It was a default.
>
>> Somewhere in the documentation the subject was mentioned, so I assumed
>> it was purposeful.
>
> Yeah, it's been part of the Transarc installation instructions from time
> immemorial, or at least longer than I've been running AFS.  I've always
> disagreed with it.  There's no need to restart AFS services regularly; it
> just incurs unnecessary downtime while the file servers reattach volumes.
> We never restart any of the daemons except for the kaserver (and there
> only to rotate logs, since we crank the logging way up from the default),
> and have been running that way for at least a decade without any
> difficulties.
>
> We do have file servers crash every once in a while, maybe one every six
> months or so, but it's not clear to me that restarts would do anything to
> prevent that.  I've never noticed anything like memory leaks.

We actually did notice memory leaks in the fileserver, a very long time 
ago.  We happened to run AFS on a particularly bizarre platform which was 
capable of dynamically increasing the size of its swap file to accomodate 
increased demands on memory.  The only problem was, it did not cope 
gracefully when there was no disk space to do so -- instead, the system 
would panic.  So we had to monitor the sizes of the swapfiles and schedule 
reboots of fileserver machines when they were getting close to running out 
of disk space into which to grow the swapfile.  This worked out to be about 
once a month, on irregular schedules.  Of course, that was a long time ago. 
We no longer run fileservers (or anything else) on Mach, and we've never 
done regular fileserver restarts.


-- Jeff