[Gluster-users] health monitoring of replicated volume

Vikas Gorur vikas at gluster.com
Sun Jun 13 02:28:23 UTC 2010


On Jun 12, 2010, at 9:24 AM, Deyan Chepishev wrote:

> Hello,
> 
> My main idea is that currently no one can tell you if the data is consistent on all nodes, for example very simple case:
> 
> 2 replicated nodes, node1 and node2
> 
> I reboot node1 and during this time, someone writes a file which will be written on node2. At this time you have out of sync replication.
> 
> When node1 comes back up in a few minutes, there is no way to know if this node is with consistent data or not. Some people suggest to run ls -lR but you have to agree that this is not applicable if you have huge file system tree with millions of files.

Please note that the "ls -lR" is only needed to trigger self-heal pre-emptively. Even if you don't do it, replicate will at all times serve read requests from the up-to-date copy, and any access to a file will automatically trigger self-heal.

I do agree, however, that a way to monitor the state would be helpful.

------------------------------
Vikas Gorur
Engineer - Gluster, Inc.
------------------------------











More information about the Gluster-users mailing list