[Gluster-devel] AFR write scheduling

Krishna Srinivas krishna at zresearch.com
Wed Nov 21 10:47:33 UTC 2007


On Nov 21, 2007 4:33 AM, Kevan Benson <kbenson at a-1networks.com> wrote:
> Krishna Srinivas wrote:
> > On Nov 21, 2007 2:09 AM, Székelyi Szabolcs <cc at avaxio.hu> wrote:
> >> Kevan Benson wrote:
> >>> I don't remember anything about the status of this, and I can't find it
> >>> in the archives from a few minutes of searching, so I'll just ask.
> >>>
> >>> Is afr write scheduling planned for the future?  My understanding of AFR
> >>> currently is that if you have a 4 node AFR, and you specify all .html
> >>> files to be redundant at 2 locations and all .db files to be redundant
> >>> in 3, you can expect the first 2 defined nodes to contain redundant
> >>> .html files and the first 3 nodes to contain redundant .db files and the
> >>> fourth node to be essentially empty (until the other nodes fill up?).
> >>>
> >>> Are there any plans to have it use a scheduler to decide which AFR nodes
> >>> to write to?
> >>>
> >>> Is the solution just to wrap multiple AFRs in a unify and use the unify
> >>> scheduler to achieve this?
> >>>
> >> See http://www.mail-archive.com/gluster-devel@nongnu.org/msg02201.html
> >>
> >
> > Correct.
> > Note that previously supported "option replicate *html:2,*txt:1"
> > pattern matching feature
> > is moved out of AFR. It should be implemented using unify's switch.case
> >
> > Also check the following link which I just updated:
> > http://www.gluster.org/docs/index.php/Understanding_AFR_Translator
>
> Thanks both.  I know it was there somewhere, I was just having trouble
> locating it.
>
> Krishna, any info on how switch.case is going to work?  Is it going to
> replace the block-size parameter of stripe as well?
>

switch.case is not going to affect block-size of stripe...

In unify you can have the following scheduler config:
option switch.case *.tar.gz:afr-node;*:child1,child2,child3

This will put all the tar.gz files in the afr which will be
replicated. Other non important
files will be created on the child[123] nodes (they will be created on
round robin basis)

Krishna





More information about the Gluster-devel mailing list