Thanks Domas, that's very helpful. I have just a few more questions ...
* A
"locator" is a string which tells MW in which replication
cluster to
find the revision.
locator tells which class should be used to retrieve text (it can be
http, database, file system, really easy to plug in), as well as the
locator. In case of database it points to external storage cluster.
I see. So, if one were to create an "external storage" database instance,
what schema is needed? Is it the same as vanilla MW?
Are there any documents on how to create an external storage node or
cluster? Is it as simple as just installing MediaWiki on those nodes?
So if one were
to attempt to minimally imitate this setup (using
the fewest
possible nodes), one would need:
'main' instance, and 'text storage' instance. you can have that all
on single node :-) but yeah, if you want to have all replication and
sharding done the way we do, you start with 4 nodes :)
I'm happy to start with just a 'main' instance and a 'text storage'
instance
- probably as different databases in the same MySQL server on my localhost.
I just don't yet understand how to do the following:
* Set up the 'text storage' instance (is this just a plain MW install?)
* Point the 'main' instance to use the 'text storage' instance. (is this
just configuring $wgExternalServers in LocalSettings.php)
Thanks for all your help so far. If all of this is already documented
somewhere, I'll be happy to read that instead of continuing with these
questions. :)
-- Jim
On 7/9/07, Domas Mituzas <midom.lists(a)gmail.com> wrote:
>
> Hello Jim,
>
> > * The WMF has multiple clusters - a main cluster, and some replication
> > clusters.
>
> We don't call them 'replication clusters'. They're 'external
> storage'. Probably we could have other names too. Its simply append-
> only storage, that for convenience now is mysql with replication.
>
> > * Each cluster has a master node and some number of slave nodes.
>
> Yes!
>
> > * The main cluster's `text` table uses the `old_flags` parameter to
> > indicate
> > that `old_text` is really a "locator", not the real data.
>
> Right, otherwise text goes directly into 'text'.
>
* A
"locator" is a string which tells MW in which replication
cluster to
find the revision.
locator tells which class should be used to retrieve text (it can be
http, database, file system, really easy to plug in), as well as the
locator. In case of database it points to external storage cluster.
>
So if one were
to attempt to minimally imitate this setup (using
the fewest
possible nodes), one would need:
'main' instance, and 'text storage' instance. you can have that all
on single node :-) but yeah, if you want to have all replication and
sharding done the way we do, you start with 4 nodes :)
>
> BR,
> --
> Domas Mituzas --
http://dammit.lt/ -- [[user:midom]]
>
>
>
> _______________________________________________
> Wikitech-l mailing list
> Wikitech-l(a)lists.wikimedia.org
>
http://lists.wikimedia.org/mailman/listinfo/wikitech-l
>