View Full Version : Status Summary for June 2007

July 6th, 2007, 12:47 PM
This is the second month in which we've prepared status updates. We'd
love your feedback!

As always, a fully linked version of this text is available at

June saw the release of Amanda 2.5.2p1 on June 6, a patch release which
was thoroughly tested with the help of the Platform Experts and many
users from the mailing lists and forums. This patch release has proven
quite stable, and development has shifted to feature development for the
next release.

Wiki Changes

* Marc Muehlfeld provided a link to a useful script for performing
backups of MySQL databases.

* Harald Schioeberg provided Lto-barcode, a perl script to generate
barcode labels for LTO Tapes. Chris Hoogendyk suggested barcodewriter,
a PostScript-based barcode utility, as a more general solution.

* Erik W wrote a detailed How To:Install Amanda Server on AIX.

* Harald Schioeberg provided Chg-multiplex, a wrapper to use multiple
Amanda configs in a single restore operation.

* Paul Bijnens contributed a multitude of minor edits all over the wiki,
increasing readability and clarity.

* Dustin J. Mitchell refactored Stefan G. Weichinger's file driver into
How To:Set Up Virtual Tapes, FAQ:Should I use a holdingdisk when the
final destination of the backup is a virtual tape?, How To:Migrate
Virtual Tapes to Real Tapes, and How To:Restore from Virtual Tapes.

Forums Discussions

Many users had questions about configuring Amanda to their needs. Most
questions were about communication problems, and were quickly addressed.
Among the more unusual questions were the following.

* tamiral asked several interesting questions:

* How can I verify that AMANDA has successfully backed-up files? --
Shailen Patel suggested using amrecover to test a full restore, which
is more thorough than the tape verification performed by amverify.

* I'm concerned that because my defined tape size is larger than the
amount of diskspace I have... -- Shailen Patel and Dustin J.
Mitchell responded that yes, this could cause a problem, and
referencing wiki pages describing the optimal configuration.

* Basically, I'd rather use 'dd' for those kind of jobs (even unmount
the filesystem so it won't change), but I guess I can't with AMANDA.
What is the recommended dumptype for this kind of backup...? --
Paddy Sreenivasan suggested a dumptype with always-full, compression
off, and encryption off.

* George Reeke had a client machine which exceeded Amanda's connect
timeout because its hard drives took too long to spin up. Paddy
Sreenivasan suggested increasing ctimeout and etimeout.

* mickeyn has clients with dynamically-assigned addresses. Marc Muehlfeld
suggested using static IP assignments in the DHCP server as a way to
stabilize the clients' addresses.

Mailing List Discussions

* Charlie Reitsma asked if any work had been done on migrating dumps
from vtapes to real tapes. Dustin Mitchell, Gene Heskett, and Jon
LaBadie replied with some pointers, and Charlie started work on an
implementation to suit his needs.

* Philip Weber sparked an interesting conversation about Amanda
automatically detecting disks that are not in Amanda's disklist and
thus are not being backed up. While the problem is difficult to solve
in the general case, a few specific solutions were suggested as
wrappers around Amanda.

* Jon LaBadie tried running amtapetype on a vtape, and the results
touched off an interesting discussion on what amtapetype should do in
such a situation.

* Marc Muehlfeld wondered how Amanda can fill a tape to more than 100%
capacity. Cyrille Bollu and Paul Bijnens answered, pointing out that
tapes differ in actual capacity, and that Amanda only uses the
tapelength for planning purposes, but will happily consume any extra
space it finds.

* Robert P. McGraw, Chris Hoogendyk, and Brian Cuttler discussed the
relative merits of hardware and software compression, noting that
Amanda has a difficult time calculating the number of dumps it can fit
on a tape when hardware compression is enabled.

* Rudy Setiawan asked for advice setting up a large disk-to-disk
configuration, and prompted a lengthy discussion about calculating
space requirements of a configuration, optimizing a system for fast
disk-to-disk backups, and ensuring sufficient redundancy for backup
data in such a configuration. Chris Hoogendyk hijacked the thread to
consider ways of running incremental backups only over a weekend.

* Sebastian Marten asked about the relationship of runtapes to
tape_splitsize. Marc Muehlfeld replied to say that splitting dumps
across tapes is impossible with only one tape (runtapes=1). Sebastian
does not have a changer, but nightly backups are larger than his media
(DVD-RAM), so Marc suggested using multiple DVD drives, spooling some
dumps to holding and flushing between dumps, or using chg-manual to
use two DVD-RAMs per dump.

* Phillip Weber wondered if he could run two Amanda configs against the
same tape changer. Ian Turner, Joshua Baker-LePain, and Jean-Francois
Malouin chipped in to say that this configuration could work, but that
some extra machinery was required to ensure that only one
configuration was making requests of the changer at any given time.

* "fedora" saw "file changed as we read it" in his backup report for
MySQL data files. This triggered a lengthy discussion: Oliver Nicole
pointed out that this is just a "STRANGE" warning, not an error, and
that it is caused by tar being unable to back up a large file while
that file is changing. Marc Muehlfeld, Bruce Thompson and Linda
Pahdoco offered pointers to scripts for backing up MySQL databases via
mysqldump or mysqlhotcopy. Dustin J. Mitchell pointed out that ZRM
for MySQL is designed to back up MySQL databases without making
textual dumps.

* Glenn Gillis had trouble with estimates failing on a partition with a
very large, constantly changing file. Kevin Till and Oliver Nicole
agreed that this was a result of a bad tar version returning an
incorrect status.

* Paul Crittenden discovered an error in the file-locking support
detection, which caused Solaris builds of Amanda to use no file
locking at all. Chris Hoogendyk, Jon LaBadie, Jean-Louis Martineau,
and Dustin J. Mitchell chased down a solution.

* Alan Jedlow discovered a bug in chg-zd-mtx, which was fixed by Dustin
J. Mitchell.

* Jean-Francois Malouin had trouble with estimates timing out after 6
hours. Jean-Louis Martineau provided a patch to reset this limit for
each estimate.

* Jordan Desroches had errors backing up a subdirectory of an NFS mount.
Paul Bijnens, Dustin J. Mitchell, Olivier Nicole, and Jean-Louis
Martineau contributed advice, eventually determining that the amanda
user did not have adequate permissions on the subdirectory. Harald
Schioeberg had trouble with Amanda not finishing when /tmp was full.
Dustin J. Mitchell pointed out that many of the external programs on
which Amanda relies use /tmp, and that the failure most likely
occurred there.

* Fedora had trouble getting client-server communication to work, and
the eventual resolution was that host-based firewalls like netfilter
and iptables can interfere with Amanda's communication.

* Ladislav Michnovic alerted amanda-hackers to a regression in SUSE
Linux, reported by Matthias Andree, relating to multihomed machines.
The issue has stalled without any feedback from Matthias.

Code Development

Most of the development work in June was focused on testing and
bugfixes for the 2.5.2p1 release. Subsequent to the release, Dustin J.
Mitchell committed a number of minor changes that had been queued up
awaiting release, including improved holding-disk management,
streamined internationalization, and debugging support.

Jean-Louis Martineau
* fixed a bug discovered by Marc Muehlfeld where hosts running in a mixed
IPv4/IPv6 environment would encounter errors (r397)
* fixed a bug discovered by Christopher McCrory where amfetchdump sent
messages to stdout, corrupting the resulting tarfiles (r410)
* fixed spurious detection of GNU Tar's backup as a dump-compatible
program (r411)

Dustin J. Mitchell
* streamlined the handling of exit statuses from the programs Amanda
invokes (r420)
* fixed broken file-locking support on Solaris, as reported by Paul
Crittenden (r422)
* clarified the error message resulting from incorrect entries in
.amandahosts (r423)

Franz Fischer submitted a patch to make amtapetype close the tape device
after rewinding, to work around a hardware/OS bug on his system. The
patch was modified and committed by Jean-Louis Martineau (r416)