Amanda-Users

Re: Backup after LVM snapshot

2005-07-01 11:39:15
Subject: Re: Backup after LVM snapshot
From: Maurice Poisson Adriasola <mpoisson AT aconcagua DOT cl>
To: amanda-users AT amanda DOT org
Date: Fri, 1 Jul 2005 11:19:58 -0400

On Jul 1, 2005, at 7:53 AM, Paul Bijnens wrote:

Maurice Poisson Adriasola wrote:
I intend to backup a host without stopping its services. To achieve it
I've put Debian Sarge's "amdump" in a wrapping bash script that does the
following:
1) Open an ssh connection, as root, to the host to be backed up and run
on it a script that creates and mounts a snapshot of the /var
filesystem.
2) Run Sarge's amdump script
3) Open an ssh connection, as root, to the already backed up host and
run a script that unmounts the snapshot and removes it.
For the time being the only host that has to be backed up is the amanda
server itself. Backups seem to be done right, but once every two runs
the hosts hangs and needs to be rebooted just after doing the backup.


Are you hitting this bug?

https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=132057

I have been hitting it in hosts with very little physical memory. Maybe there is barely enough memory on this host too. I'm gonna check it.

I never noticed the regularity -- once every two runs -- or is that just
a rough proportion of the crashes/successes?

It has been once every two runs (or once every 48 hours) if and only if the amdump-wrapper script is launched by cron (as user backup). Instead, if the same script is run as a shell job by user backup, it can be run more times without failure.


amcheck is done a couple of hours before amdump. It is also preceded by
snapshot creation and followed by snapshot removal and it never fails,
Thus I believe I must have overlooked something in amanda's
configuration.

Again a clue that it could be load related: during amcheck there is very
little disk activity.

To avoid the trouble with lvm2, I also avoid any conditions like
running out of space, just in case they might tickle the bug.
Maybe you have a larger diskactivity or larger dump or... once every
two runs, tickling the bug?

Failure seems to happen only every time amanda's planner schedules an incremental backup of the (snapshoted) /var filesystem and never when a full dump is due.


<Prev in Thread] Current Thread [Next in Thread>