> > > How are you doing the 2 tapes per day?
> > 
> > I lie to amanda, and tell here there's one tape of size 
> > 80Gb, and run amflush manually each day when I get into 
> > the office.  If I were running 1
> 
> *This* is your problem.  What is your reserve set to, and how 
> much holding disk space do you have?  

Holding disk /tmp/amanda/dump: 57492036 KB disk space available, that's
plenty.

> When amanda hits EOT unexpectedly, it goes into "degraded" 
> mode.  By default, amanda "reserves" 100% of the holding 
> disk space for degraded mode backups, which are incrementals
> only.  If you set reserve to a low number, then you free
> space for full backups in degraded mode.
> 
> So, amanda was probably going into degraded mode every night 
> and didn't have any space for level 0s.  Thus the "can't 
> incremental new disk" and "dumps too big" messages.

That's a good explanation.  I checked my amanda.conf file:

# reserve 30 # percent

so it's reserving 100% by default.

> Also, are you using software or hardware compression?  
> Depending on your data, you should be able to get more
> than 40GB on DLT8K tapes...

Yes, I was trying to get it to *work* first, then tweak for optimization.  I
can definitely up that -- I'm using hardware compression (drawbacks) because
if I don't, it runs for 12 hours doing all the gzipping...

> > tape per day with a 4 week tape cycle, and I'm actually 
> > generating more than 40Gb per day of level 1, then there 
> > will be no space left ever to do the level 0 dumps, 
> > right?
> 
> If amanda estimates that any dump (whether full or 
> incremental) is bigger than a tape, it will tell you 
> that, won't dump the disk, and will do everything 
> else that it can.  Some lies are ok to tell amanda, 
> this one isn't.
> 
> As I see it, you've got a couple of options:
> 
> 1)  Set runtapes to two and use chg-manual.  Not optimal, as 
> I don't think you can run amdump out of CRON as it needs you
> to tell it when you've put in the next tape.

A big drawback. :(

> 2)  Split up any partitions with large incrementals 
>     using tar so that you're sure every disklist
>     entry will fit on a tape.

By that do you mean "make sure the size as reported by /usr/bin/du is less
than 20Gb"?  That's what I do, via a script.  So I make sure it fits on
*half* a tape, to leave the other half for daily incrementals.  But what
happens when nearly every day, the daily incrementals are, say, 30Gb?

> There may be others, as well.  But that's a big, bad lie 
> you're telling amanda there.  You could try cranking
> your reserve down if you want to continue your lying
> ways.

I'll try cranking it down to 10%, and wait a few days to see what she does.
Then maybe I should switch (back) to 

dumpcycle 4 weeks
runspercycle 20
tapecycle 22 tapes
runtapes 1

I think I had a problem with that before.  Yes I remember now, I had a
disklist entry that was larger than my tape, and amanda *wasn't* telling me,
and when it got to level 9 there was a bug that caused amdump to crash.  But
perhaps that's fixed now.  My script that checks for large disks is fixed,
anyway.

Bart
--
Bart Brashers                   MFG Inc.
Air Quality Meteorologist       19203 36th Ave W Suite 101
[EMAIL PROTECTED]        Lynnwood WA 98036-5707
http://www.mfgenv.com           425.921.4000 Fax: 425.921.4040

Reply via email to