Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Hi
I am exploring the use of DRBD as a backup system for some servers.
So far I've installed it on a couple of test servers, and it's worked
well -
once I understood some idiosynchrasies of the command line :)
But my objective is a little more ambitious than just making a single
online
backup. What I'd ideally like to do - and I've seen this asked for
before
- is achieve incremental backup also, so I have both the current backup
of the server and also a backup from last week, for example.
Of course, I can do this by taking down the DRBD connection, and using
a simple DD to backup the secondary. But my servers are quite big -
100gb or so - and it's hard to do this in a realistic time.
I have figured out a number of creative solutions to this:
- Run DBRD on top of LVM
- Run Ext3Cow (copy-on-write filesystem) on top of DRBD
- Run DRBD on top of cowloop (a copy-on-write block driver)
But there's a snag with each of them:
- LVM seems inefficient if you add multiple snapshots
- Ext3Cow has no way of deleteing snapshots
- DRBD won't actually load on top of cowloop ...
I guess I should say that this is largely an experimental endeavour -
if I can make it work, it would solve a problem, but I am also
just curious about what's possibe.
So.. I have some questions:
How to take backups of the secondary ?
----------------------------------------------------------
Imagine this case:
- My main server, 'alpha' is primary
- My backup server, 'beta' is secondary
At midnight, I want to take an additional backup of the
secondary. I only need RO access, but I'm not supposed
to mount the secondary.
This is my approach:
- Disconnect DRDB
- Change beta to primary
- Mount it
- Make a backup
- Unmount
- Change it back to secondary
- Reconnect
It this correct?
What about this:
- Disconnect DBRD
- Sync drives
- Mount either the DRBD device or the underlying physical disk RO
- Make backup..
Can you tell if there are sectors waiting to sync?
-------------------------------------------------------------------
Imagine we have the previous example, and while DRBD is disconnected
someone on the primary server copies a 100mb file to the drive. This
means
we have 100mb waiting to sync..
I tried exactly this case, copying a 100mb file, then reconnecting. I
then
immediate ran cat /proc/dbrd - but it did not show sync in progress.
If there are sectors being written because of a reconnect, does DRBD
show
them on proc?
(It occurs to me now that the 100mb file I copied was probably sparse,
so it may have had a very small amount of real data, although the copy
took a long time)
How does DRBD decide which way to Sync?
----------------------------------------------------------------
Just curious...
Imagine the scenario above again, but this time I write sectors to both
drives while disconnected.
How does DRBD decide which one is primary? If I make one secondary
before the reconnect, does it assume that one is to be overwritten?
I guess I should try it out...
Incremental Backup
----------------------------
Has anyone tried any of the incremental backup solutions listed above?
Finally - has anyone patched DBRD to include snapshots? Or is this a
possible future feature?
Best Regards
Martin Green
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linbit.com/pipermail/drbd-user/attachments/20050206/0b9689f4/attachment.htm>