Hi Michael, I just sent an email talking about a project - Quark - we have a built on top of Calcite. We use lattices extensively. We define the tiles and do not depend on Calcite to select the tiles. Calcite optimizes queries correctly and chooses the right tiles etc. We've have a couple of issues but it shouldnt come in the way of starting to use it. 1. Lattices dont work if its used in a multi-threaded app. Calcite planner is not accessed concurrently but a different thread might own it and use it. For e.g. one thread opened the connection and another thread is creating and executing statements. 2. Each tile in a lattice gets parsed and setup for every single query. If you have 4-5 lattices with 6-7 dimensions, optimization becomes slow. We are still investigating if we are not using it correctly and if we are, how to solve. We will fire JIRAs once we have more concrete information.
On Thu, Jan 21, 2016 at 7:50 AM hongbin ma <[email protected]> wrote: > hi Michael, > > this is Hongbin Ma from Apache Kylin (http://kylin.apache.org/), like > James > mentioned, Kylin aimed to server OLAP queries by 1. pre-aggregating raw > records and save it into hbase 2. use Calcite to parse the query SQL and > translate it to HBase scans. > > AFAIK there're other projects doing the similar things except that they > store the pre-aggregated results in ElasticSearch. However it seems not > open sourced, and I believe it's not as mature and extendable as Kylin. > We're also very interested in trying out different storage engine rather > than HBase (like Cassandra). So if you're interested please throw a topic > in our dev list. > > > -- > Regards, > > *Bin Mahone | 马洪宾* > Apache Kylin: http://kylin.io > Github: https://github.com/binmahone >
