[DRBD-user] Snapshot with DRBD

H.D. devnull at deleted.on.request
Thu Feb 14 14:51:04 CET 2008

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


drbd at bobich.net wrote:
> On Thu, 14 Feb 2008, Tomasz Chmielewski wrote:
> I disagree. It's not a solution because syncing the FS isn't sufficient. 
> With LVM you still get the FS level consistency as it is at the time of 
> the snapshot. The problem is that applications like DBs maintain their 
> internal buffers and caches, and these will not necessarily be 
> consistent with the data on disk, FS freeze or not.

Syncing the FS _is_ sufficient. That is why DBMS implement write ahead 
logs. Recovering from a snapshot is the same as recovering from a power 
failure. And if a DBMS can't do that, it doesn't conform to ACID, and no 
one serious would use it. There is no serious DBMS out there that is 
unhappy with FS snapshots, even MySQL / innoDB is fine.

In case of XFS it's actually better than a power failure, both meta data 
and data is flushed, as well as an umount record written.

Have you ever tried running a DBMS on copyfs? There is a reason why no 
one is doing that. First there is no reason to do it, second performance 
is unacceptable.


-- 
Best regards,
H.D.



More information about the drbd-user mailing list