[Gluster-devel] Request for "read-only" translator

Steffen Grunewald steffen.grunewald at aei.mpg.de
Sat Jun 30 15:01:21 UTC 2007


On Sat, Jun 30, 2007 at 11:41:19AM +0400, Raghavendra G wrote:
> Hi Steffen,
> Sorry there was A mistake in my reply. The Unify schedulers cannot schedule
> based on whether a sub-volume is read-only or not. Hence when a new file
> creation operation comes, the scheduler _cannot_ neglect the read-only
> subvolumes for creating the file. There is a fairly good chance that the
> file creation operation be scheduled on a read-only subvolume which ofcourse
> will fail.

That's a pity.
Please reconsider this: Imagine a storage cluster mainly used for "archival
of data on spinning media". In such a case, one would set up a round-robin
scheduler to fill the storage devices (bricks) one by one, and one would
"freeze" full bricks by setting them read-only so modification will be no
longer possible. I think such a behaviour would make sense. 

So here's the setup again:

> >> brick1a
> >>       > AFR (mirror) > readonly
> >> brick1b                          \
> >>                                  |
> >> brick2a                          |
> >>       > AFR (mirror) > readonly  |
> >> brick2b                           > unify round-robin
> >>                                  |
> >> brick3 (AFR if I'm paranoid)     | - set to read-only when full
> >> brick4 (can be AFR)              | - ...
> >> brick5 (can be AFR)              /
> >> ...

Cheers,
 steffen






More information about the Gluster-devel mailing list