[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-users] Backup solution

  • To: Artur Linhart - Linux communication <AL.LINUX@xxxxxxxxxxx>
  • From: Nico Kadel-Garcia <nkadel@xxxxxxxxx>
  • Date: Wed, 11 Jul 2007 17:01:57 +0100
  • Cc: 'xen-users' <Xen-users@xxxxxxxxxxxxxxxxxxx>, "'Fajar A. Nugraha'" <fajar@xxxxxxxxxxxxx>
  • Delivery-date: Wed, 11 Jul 2007 08:59:33 -0700
  • Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=beta; h=received:message-id:date:from:user-agent:mime-version:to:cc:subject:references:in-reply-to:content-type:content-transfer-encoding; b=Ph1nAcVPsx1djlHIn0mI1j9qGY5c3LPg8I7f0wFRrjjivv47LgiS/wlDl1kNCGwpPokqrv+/jUSS9HEqaNiQxcy8FQIPYWQTM7VvfBWdSBW9BM1QhKzD8ne1rtrrUyTXsKcw772ofY8QhURlhzaGmiPd4kPhZKPBpU0AfnRCF58=
  • List-id: Xen user discussion <xen-users.lists.xensource.com>

Artur Linhart - Linux communication wrote:

        I§ve thought about it and i think the good compromise seems to be
for me

1. start "xm save" to archieve the current and consistent state of the
2. during the "xm save" command is running , the domain is paused, I can
create the snapshot volume during this time (in script could be waited till
the saved memory snapshot file will be created on the disk, if I understand
it correctly, at this moment the domain must be already paused, so now the
snapshot can be created
3. after the save is executed, the xm restore can be done again (or the
command "xm save" can be started with the option -c to leave the domain
running after the memory snapshot will be created, but this can be dangerous
in the case the "xm save" would take too less time to create the snapshot
volume during this time)
This is.... optimistic. The saved state of RAM in such a process is potentially "dirty" with cached data not yet written to disc. And atomic operations, such as database operations, are potentially in mid-process because the write to disc is not complete and the database is still locked at the time of the save. And systems like Linux tend to keep as much possible of the paged material in RAM, rather than necessarily writing to disk. And some filesystems, such as ReiserFS, are *particularly* horrid about not completing file-system operations and then attempting to recover from them.

With all that added together, I'd consider this reasonable if crunched for downtime, but not reasonable for high reliability of sensitive databases such as Oracle, db4, MySQL, or that mess of twisty little undocumented formats all different called an Outlook userfile.

Xen-users mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.