[Gluster-devel] High abailability question...

[Steffen Grunewald]

On Tue, May 27, 2008 at 10:50:56AM +0530, Krishna Srinivas wrote:
> Victor,
> 
> Here is your steps:
> 1 open file
> 2 read
> 3 bring first child down
> 4 read continues with a seamless failover
> 5 bring first child up, and bring second child down.
> 6 read fails.
> 
> Now retrying again from first server would be difficult as all the "states"
> associated with that server would have got lost when it "went down"
> It could have been a temporary network disconnect or the server could
> have got rebooted (in which case it would have lost the open file
> descriptor)

So does this mean that the client keeps track not only of the file position,
but also of the server(s) available? (I can see that this would make sense
for files opened for write, but for read-only files this looks unnecessary)

> The file has to be reopened in this case.

Which makes HA non-existent in the user's eye. Or at least one-way only.
In the real world I can imagine scenarios where to-and-fro failovers would
be necessary. Even for single files. Without the need to re-open (which
app would do that?)

Hmmm.

Steffen

-- 
Steffen Grunewald * MPI Grav.Phys.(AEI) * Am Mühlenberg 1, D-14476 Potsdam
Cluster Admin * http://pandora.aei.mpg.de/merlin/ * http://www.aei.mpg.de/
* e-mail: steffen.grunewald(*)aei.mpg.de * +49-331-567-{fon:7233,fax:7298}
No Word/PPT mails - http://www.gnu.org/philosophy/no-word-attachments.html