Force replication restructure

dormando dormando at rydia.net
Tue Nov 6 20:35:11 UTC 2007


Rob wrote:
> I started my MogileFS "cluster" with just a single storage node.
> I just added two more storage nodes, and am looking for a way to force
> re-replication of it's files to the other nodes. Improve redundancy
> and load distribution, etc. What is the recommended way to go about
> this?
> 
> I was thinking either flag the devices on the first node as down
> (slowly, one at a time), or increase the replication count for my
> classes, wait for replication to complete, then put it back to normal.
> 

What's your devcount set to?

Since previously the hostcount was 1, I'd recommend first running a 
'fsck' - which will find policy violations and re-replicate files. I'm 
almost sure if you have a devcount of at least 2 on the class for those 
files, and they're stuck on one host, it'll cause a policy violation and 
make an extra copy.

Instead of marking devices as down (which doesn't actually do anything) 
or dead, you can mark them as 'drain' one by one, which shouldn't affect 
the system's ability to actually find and replicate the data. This may 
be optional, run a fsck first and see what happens ;)

-Alan


More information about the mogilefs mailing list