[Gluster-devel] How to make out-of-sync files visible in replication setup

Ed W lists at wildgooses.com
Sat Mar 27 13:54:07 UTC 2010


On 25/03/2010 18:54, Ian Rogers wrote:
> Some random thoughts:
>
> 1. A client will be making a note like this when it's trying to write 
> a file and spots that one of the replica sub-volumes is down
> 1.1. so the client shouldn't try to store this info in the sub-volumes 
> as, by definition, they're in a degraded state.
> 1.2. the info should include which sub-volumes were up/down to help 
> the clients make an informed decision for self-heal
>

So presumably you can figure out which volumes were down because they 
wouldn't have the links shown on them?

The logic would seem that you do an "ls" of the "degraded" directory and 
this causes all the links to be created across all volumes (if they 
weren't already).  You can then delete all the links if you are sure 
that all volumes are up.

Interestingly if one volume went down as you tried to delete the excess 
links then you would presumably have extra links once that volume came 
back up again - this seems safe in that at worst you scan some extra files

Ed W






More information about the Gluster-devel mailing list