On Tue, Dec 09, 2003 at 09:53:59AM -0500, Brian Cuttler wrote:
> Hello,
>
> Hate to say but I'm running (probably) one of the odder configurations.
>
> We run two amanda servers concurrently on a single system (Solaris 9,
> Amanda 2.4.4). We are able to do this because one of the servers
> backups up only "local" ie UFS partitions and the other only "remote"
> clients (clients not of localhost). We are writing to two separate
> jukeboxes (each with a single LTO tape drive).
>
> I have started seeing that the remote backups fail, the following
> is the typical message
>
> panther /data lev 1 FAILED [no more holding disk space]
>
> I take it to mean that the local backups, running much more quickly
> (fast processor, no network) than remote backups and _shareing the
> same physical /amanda/work partition_ are filling up the amanda/work
> area and the remote backups having no place to write are failing.
>
> Is this the case ? Why don't the (remote) dumps simply delay until
> spool clears - or perhaps put one or two into degraded mode, allowing
> the other amanda server to write its own drive and clear the spool
> and allowing the remote backups to resume use of spool.
>
> I don't really want to limit the amount of spool available to either
> server, I suppose I could probably stagger the two servers in cron but
> not certain which way to skew them.
>
> Ideal situation, single amanda server - write completed dumps to
> alternate jukeboxes, output any of n-dumps to first available or
> best fit drive. Sort of a round-robin on the jukeboxes. Doable ?
>
I did not consider this carefully,
quick off top of my head possibilities.
Local config, set all DLEs to same spindle number.
That will limit them to max one dump at a time.
At the same time, limit the holding disk to accomodate
the two largest local DLEs. That will allow one done
DLE taping with one more dumping; but will limit the
holding disk usage. You might then be able to leave
the remote config unchanged.
Another possibility is to consider the two LTO drives
a "changer" using the chg-multi script designed for
multiple drives, not changers. Then use a single
script for all DLE's and runtapes of 2.
--
Jon H. LaBadie jon AT jgcomp DOT com
JG Computing
4455 Province Line Road (609) 252-0159
Princeton, NJ 08540-4322 (609) 683-7220 (fax)
|