[prev in list] [next in list] [prev in thread] [next in thread] 

List:       amanda-hackers
Subject:    Re: "skip-incr" ignored?
From:       David MILLER <dpm () iassf ! easams ! com ! au>
Date:       1995-11-10 1:06:05
[Download RAW message or body]

Excerpts from amanda: 12-Oct-95 Re: "skip-incr" ignored? Michael C.
Matthews@pd.s (3723*)


> >Hello,
> >    I've been putting together a new amanda configuration based on my
> >experience with the local machines. I defined new dumptypes in the
> >amanda.config file using the option "skip-incr" and used them
> >for 3 small, relatively static partitions. The first time through
> >everything was fine with the level 0 dumps. The second time through
> >there were 3 different, unexpected behaviors. I got a level 1 dump on
> >one and "FAILED" messages on the other two. The latter cases were
> >'too big' and 'no estimate or historical data'. The machine is a Sparc2
> >with SunOS and it is both server and client. I built amanda-2.2.6 with
> >gcc-2.5.0.
> >    Is this a known problem or have I managed to foul up my installation?
> >Some of the 'amdump' log-file is excerpted below.
> > - Dave Covey

I got something similar this week, and I've tracked it down to a problem
with planner. What happens is that is a filesystem is skip-incr the
initial estimates are skipped, the getting of an estimate is skipped,
but the filesystem ends up on the failed queue. An estimate is then
calculted anyway (and it's NOT from historical data) and that estimate
is 0! This causes a division by zero (I think) later on in planner.

I have done a quick fix so that if the file system is skip-incr then it
won't get added to the failq. This works, but I think that it is
basically a hack. While it shouldn't appear on the failq it the
estimation should use historical data (but I didn't have any in this
circumstance anyway!).

Anyway, here is the patch. I'm not on amanda-hackers so perhaps someone
already has this, but ...

*** server-src/planner.c	Fri Nov 10 11:49:14 1995
--- server-src/planner.c.new	Fri Nov 10 10:51:22 1995
***************
*** 795,801 ****
--- 795,805 ----
      }

      for(dp = hostp->disks; dp != NULL; dp = dp->hostnext)
+     {
+ 	if(est(dp)->level[0] == -1) continue;	/* ignore this disk */
+ 						/* because skip-incr*/
  	enqueue_disk(destqp, dp);
+     }
  }

  static disk_t *lookup_hostdisk(hp, str)


David Miller, Unix System Administrator
Easams Australia
Direct +61-2-367 4572	Fax +61-2-367 4566
Unit 5, 2 Giffnock Ave, North Ryde, NSW 2113

[prev in list] [next in list] [prev in thread] [next in thread]