New in Cassandra 3.0: Materialized Views

[ivank]

http://www.datastax.com/dev/blog/new-in-cassandra-3-0-materialized-views

[Sphax]

Could be useful, it's a shame they're quite expensive to use on high-write tables.

[arthursilva → Sphax]

I think it's a great usability improvement nonetheless. Sometimes you just need C* easy scalability (at least compared to the alternatives)

[dwenzek → Sphax]

Yes this will be definitely useful, due to how eager is Cassandra of denormalized data !

See the comments a month ago about Cassandra data modeling (https://news.ycombinator.com/item?id=9728160).

Even this example shows a materialized view per request !

[_benedict → Sphax]

Performance will be improved in later versions

[cowardlydragon → _benedict]

I'm guessing that materialized views can be sped up if the distribution of the component row keys can be mirrored across tables, so that the materialization can be treated as local and no internode communication is necessary.

Having primary keys with identical node distribution keys is a recent set of commits...

[_benedict → cowardlydragon]

Yes, although that is probably useful for only a very limited subset of use-cases for materialized views. There are, however, more general purpose improvements to the approach that are just more involved to implement (and have their own negative performance risks, that would need to be balanced)

[tacone]

"Materialized views are not supported through Thrift"

[threeseed → tacone]

Thrift has been essentially deprecated for quite a while now.

Either use CQL or the Binary Drivers.

[assface]

Awesome. Welcome to 1998.

[cowardlydragon → assface]

Where they had arbitrary linear cloud scaling ability?

Oh right, where you only had one 486 server to do 1 query / sec.