[Gluster-devel] AFR in actual use

Brent A Nelson brent at phys.ufl.edu
Mon Feb 26 03:16:36 UTC 2007

Based upon an earlier comment, when AFR comes out in 1.3 on Wednesday, it 
will happily replicate data as it is changed, but a node that goes down 
won't resync itself with any changes made while it was down when it comes 
back up.  Resyncing would be done in a future release by an fsck process. 
Is that correct?

If so, some questions:

1) Version 1.3 won't be usable as a fault-tolerant clustering filesystem, 
although 1.4 might be, right? Would there be a crude method of resyncing 
manually (such as rsyncing the underlying filesystems), at least for 
testing until 1.4 is out?
1a) Similarly, what about adding replication to an existing area that 
already has data?

2) In 1.4, would the fsck resync occur after the out-of-sync node has 
already reconnected as a replica? If not, how would you ensure that you 
pull in all the changes made while the fsck is running (but before AFR 
kicks in)?



More information about the Gluster-devel mailing list