I am looking for a way to effectively do partitioning in a DB that doesn't support it and we don't need to scale to multiple DB's. Does shards provide that ability? From reading the docs, it seems designed to partition horizontally across databases.

I haven't tried, but I see no problem doing that. Just create one SF config per shard, all SF pointing to the same database

I think the biggest problem you'd run into pointing multiple SessionFactories at the same database is that it violates one of the primary assumptions that Shards currently makes, namely that data is not replicated across your shards. If you point multiple Session Factories at the same database then your data is effectively replicated across all shards, and a query that executes across all shards will return duplicate results. Even attempting to look up a single object would yield strange results because that object actually lives on all shards. Your strategies would probably need to do something strange to make this work.

If your goal is to get single database horizontal partitioning in a db that doesn't support it, I think you're going to need to codify the different partitions in your db schema. That might mean appending a discriminator to the end of each table (Address_US, Address_CA) so that at least each record lives in a distinct shard. Then the issue is getting Shards to support shard-specific table names. At the moment we don't support this (we require each shard to have the same mappings, which includes the table name) but it's certainly something we could consider if people found it useful. Is this something you could maybe use?

Max

Oh year I forgot that the table names are the same between shards.
This is a constraint that could be dropped though.

Yes, it could. The trick is to do it in such a way that people don't need to declare a different mapping for each shard. I was looking for a way to dynamically swap out the value of an xml entity so I could do something like this:

table_name="Address&SHARD_ID;"

And then as I set up each SessionFactory provide a different value for SHARD_ID. I'd prefer to do this at the XML parsing layer rather than text replacement but I haven't cracked it yet. Any ideas?

inject a custom org.hibernate.cfg.NamingStrategy

(probably wrapping the existing one)

That would work great for table names, but I'd like to be able to do it in arbitrary places throughout the document (like for cache config attributes).

the cache regions can be prefixed through a configuration property, so we should be fine for this one too.

Stop making it so easy! :-)

What if you declare at shard master config level a naming policy? Or is this what Emmanuel already suggested?

./alex
--
.w( the_mindstorm )p.

That's pretty much what Emmanuel suggested, but I haven't tried to see how it works. Want to give it a shot?

Max

He he... I thought it was what Emmanuel suggested :-). I'll be in vacation for the next 2 weeks, but then I will join the dev discussion group and we will see if I can help.

bests,
./alex
--
.w( the_mindstorm )p.
_____________________________________
Alexandru Popescu, OSS Evangelist
TestNG/Groovy/AspectJ/WebWork/more...
Information Queue ~ www.InfoQ.com