Hi,
your config is wrong, see comments later on...

Gertjan van Oosten schrieb:

Is there still nobody here who can answer my question?

As quoted from Gertjan van Oosten <gertjan AT West DOT NL>:

I'm having a problem configuring Amanda-2.4.4p2 (on sun-sparc-solaris8)
to always do a full dump of a filesystem.  The filesystem in question is
over 100 GByte, and no matter what I try, when Amanda starts, Amanda's
planner asks for a size estimate for the level 0 dump of this filesystem
(good), and then for a size estimate for a level 1 dump of this
filesystem (not good).  The level 1 estimate takes too long, so Amanda
aborts with the (in)famous:

FAILURE AND STRANGE DUMP SUMMARY: datahost /data lev 0 FAILED [Estimate timeout from datahost]

In my disklist I have:

 datahost /data always-full


Excerpt from my amanda.conf:

 inparallel 4
 dumporder "Ssss"
 netusage  10000 Kbps

 dumpcycle 1 day

set this to 0

 runspercycle 5

and this to 1 as you won't do 5 runs in a 0 day cycle.

 tapecycle 5 tapes

 bumpsize 20 Mb
 bumpdays 1
 bumpmult 4

 etimeout 300
 dtimeout 1800
 ctimeout 30

 tapebufs 30

 runtapes 1
 tpchanger "chg-zd-mtx"
 tapedev "/dev/rmt/0bn"
 rawtapedev "/dev/null"
 changerfile "/opt/amanda-2.4.4p2/etc/amanda/tapehost/changer"
 changerdev "/dev/scsi/changer/c3t0d0"

 maxdumpsize -1

 tapetype HP-LTO-2
 define tapetype HP-LTO-2 {
     comment "HP LTO-2 Ultrium (hardware compression off)"
     length 201216 mbytes
     filemark 0 kbytes
     speed 23663 kps
 }

 define dumptype global {
     comment "Global definitions"
     index yes
 }

 define dumptype always-full {
     global
     comment "Full dump of this filesystem always"
     compress none
     priority high
     dumpcycle 0
     strategy noinc
 }


What I want is simple: Amanda should do a level 0 dump of that
filesystem, and not try to find out how large a level 1 dump of this
filesystem would be (it wastes a *HUGE* amount of time, almost an hour).
I want a level 0 or nothing at all.  Is that possible, and if so, how?

By the way, even though the above specifies:

 etimeout 300       # number of seconds per filesystem for estimates.
 dtimeout 1800      # number of idle seconds before a dump is aborted.

the estimate is not aborted after 300 seconds but only after 1800
seconds.  That doesn't seem right, does it?

AFAIK this is normal, amanda takes the etimeout-value, multiplies it
with the number of DLE's for that host and waits for that time, as it
gets the estimates of all dle's of one host with one request to that
host.

Furthermore, even after the planner decides to timeout and cut the
connection, the amandad/sendsize/ufsdump processes keep on running on
the client (until they're done, in fact, but they have nowhere to send
the data to).  Shouldn't they be terminated as well?

No, this is perfectly normal behavior as amanda has no way to reach out
via the network to the remote machine and kill these processes while
they are running, because amandad on that machine is busy until they
terminate.

Regards,

Christoph