Setting classpath for individual datasource is a good idea. However convenience for 80% cases must not be comprised. E.g. a meaningful default hive datasource is still very important in my opinion.
Yang On Sun, Jul 10, 2016 at 7:35 PM, Yiming Liu <[email protected]> wrote: > Hi Kylin developers, > > Currently, Kylin will load the hive configuration from the HIVE_DEPENDENCY > classpath. That means Kylin supports only one hive source. KYLIN-1826 aims > to support multiple hive data sources, but the design is a little > complicated by introducing the EXTERNAL HIVE concepts. > > The data source management becomes more tricky when multiple hive clusters > and multiple kafka clusters are needed. I just rise the question today > without specific solution yet, all suggestions are welcomed. I think it > could be very useful if Kylin could support data source management. > > It should have the following features: > 1. Defines Hive Cluster/Kafka Cluster as the data source factory under > Project. > 2. One Project could have more than one Hive/Kafka/SparkSQL Cluster > definitions. > 3. When "Load Table/Streaming", Kylin could load the TABLE(Hive) and > Topic(Kafka) definition from Hive/Kafka directly. > 4. The following model design and cube build are the same as before still. > > I know it's not a critical requirement, but maybe someone wants it too. > > Thank you a lot. > > -- > With Warm regards > > Yiming Liu (刘一鸣) >
