[Gluster-devel] Failover question.

[Anand Avati]

>
>     I know it's possible to set up an automatic mirror so that one
> file system is redundant and that failover will happen if you lose one
> or whatever from the mirror.  That's great.  I have a situation in
> which we have a common utillty area and if NFS blows a gut failover is
> problematic at best.  Basically there isn't any.  So Gluster resolves
> this issue nicely.
>
>     But (there's always a but) the problem I have is that the servers
> I have get really really busy at times to the point that wherever this
> common area gets put loading any pograms from it can slow to a crawl. What
> I'd like to be able to do is tell Gluster that if the load on a
> particulat member of the mirror set or the response time gets to a
> certain value, temporarily drop it from the set and recover when it
> returns to normal.  The server never actually drops out it just slows to a
> crawl which is a different problem.
>
>     Any way for Gluster to handle this?  Thanks much.
>

Maybe you can write some scripts which monitors the system load and kills
the glusterfsd, and restart the server later. The client takes care of
failover when the connection breaks.

avati