Do you have a gut feel for when the Cloudant clustering code will make the trunk? 1Q10? 2Q10?

-Brad

Any gut feel for when the Cloudant clustering code will hit the trunk? 1Q10?, 2Q10?

-Brad

Yes it’s possible to do it all in the http domain (and I think that’s really important).

But pragmatically I think a pure Erlang approach is simpler, as it is designed to handle exactly these sorts of problems.

Chris

There are plans to include a native-Erlang clustering feature (based on Dynomite and work done by Cloudant) which should scale up to thousands of machines.

@Jan Compaction sounds like a faster way to do the same. Probably the best way to do it like that indeed.

@Matt, with couchdb lounge, you could basically scale out by adding nodes and decreasing the amount of data per node. I’m kind of curious how well supported/stable their stuff is.

Just a couple of corrections and points of interest:

The _rev is not actually a uuid. It’s a predictable value (but treat it as opaque!) that is essentially a hash of the doc and its parent _rev.

As Jan L pointed out, you can also cleanup a database by compacting it. Compaction can take place while the database is in use.

It’s not quite true that you can “add nodes to scale” map-reduce. CouchDB does make use of multiple view servers (one per core, I think) to rattle through map-reduce but it doesn’t distribute load to additional Erlang nodes.

Views are actually updated on read, not write. They can take a signficant amount of time to build if a lot has changed so you need to be a little careful adding “shitloads” of docs if you want your UIs to stay responsive.

Have fun playing with CouchDB!

Add since you can use CouchDB as a basic HTML template server with Show and List functions, you can even use CouchDB to run your UI. It is a fantastic LAMP alternative.

One nit though: “Cleanup by replicating”: There is also compaction, the removes old data from a node without copying to a second node. You do need enough space for your working copy though.