Re: [Gluster-devel] df -kh not reporting correct value, AFR

[DeeDee Park]

Thanks for the info on the -n, I have a much clearer understanding of it.
I didn't see any other docs on the -n option, and I thought I had read all
the docs. After you mentioned it, I did remember the AFR limitation
when doing the :2 to the first two bricks, so the other 3 don't get 
anything.
I had just forgotten about the limitation or wasn't correlating
it with my testcase.

Hmmm, This makes me think tho, if my total dataset is larger than the size 
of
the first two bricks, then replication will fail. I noticed that all my 
files were going
to the last brick that was listed. I had like 300 some odd files in each 
brick, and the last
brick had all the files. I hope that this will get redesigned so that the 
end user doesn't
have to be soo smart about which order the bricks are listed in the 
subvolume command.

Here is a thought for a simple fix on AFR, if I have say 10 bricks, and say 
AFR *:2, then use
a modulus of the available bricks, so you would still use available bricks. 
- eg copy set 1 could
go to 1,3,5,7,9 and copy set 2 could go to the 2,4,6,8, or 10th listed brick 
in the subvolume
command.

I'm shooting for the small number of client and large number of distributed 
server/brick model
and a couple NFS gateways. I would naturally have a lot of server bricks, 
but would only want to
do like 4 or 5 replicas. If I had 10 bricks, that would leave the last 5-6 
bricks not used for replication.


> From: "Daniel van Ham Colchete" <address@hidden>
> To: gluster-devel <address@hidden>
> Subject: Re: [Gluster-devel] df -kh not reporting correct value
> Date: Wed, 11 Jul 2007 21:03:36 -0300
>
> DeeDee,
>
> I read your spec file from another e-mail, and I think I know the answer.
>
> Please any developer correct-me if I'm wrong.
>
> According with
> http://lists.gnu.org/archive/html/gluster-devel/2007-03/msg00106.html the
> AFR translator will return the minimum  size between all it's volumes when
> you df -h.
>
> But, from what I could read at the source, AFR currently will return the
> total space of the first available subvolume.
>
> For fact you can be assured that it does not sum the total space of all 
> it's
> volumes.
>
> As a fellow user, I recommend you to sent more details when asking 
> something
> at the list. It helps me understand your problem more. If it wasn't for the
> other e-mail you sent to the list I would never imagine that you are 
> putting
> all the 5 bricks inside AFR, although you only make 2 copies of each file 
> at
> the first two subvolumes available at open() time and the other three is
> almost never used.
>
> Suggestion:
> http://www.gluster.org/docs/index.php/Understanding_Unify_Translator
>
> Best regards,
> Daniel Colchete
>
> On 7/11/07, Daniel van Ham Colchete <address@hidden> wrote:
> > DeeDee,
> >
> > I'm not a Gluster Developer but I think I can help.
> >
> > First, it is easier if you send your volume files :-) .
> >
> > When you mount a Gluster brick running the glusterfs command, you can use
> > the '-n' option to mount just a 'protocol/client' brick by its name and 
> > see
> > how everything is at the last level. I would start tracing using this. W
> >
> > Best regards,
> > Daniel Colchete
> >
> > On 7/11/07, DeeDee Park < address@hidden> wrote:
> > >
> > > I have 6 bricks totalling 2.3T of space right now in my test setup.
> > > (6GB,
> > > 40GB, 250GB, 500GB, 750GB, 750GB)
> > > I run the 'df' command and it currently shows 967M of space available 
> > at
> > > the client.
> > > It use to show the correct amount a while back. How can i trace this
> > > so I can find out how much each brick is reporting as total space
> > > available?
> > >
> _______________________________________________
> Gluster-devel mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/gluster-devel

_________________________________________________________________
Don't get caught with egg on your face. Play Chicktionary!  
http://club.live.com/chicktionary.aspx?icid=chick_hotmailtextlink2