@gianm 
Parallelizing union queries with the existing design doesn't work even using 
AsyncQueryRunner as the merging for all the queries except GroupByV1 are 
performed lazily.
I think what would help fix this and union queries overall would be to modify 
the `Query` interface to accept a list of datasources instead of a single 
datasource. In this case, the `BrokerServerView` would have to merge the 
timelines for all the datasources in the list before the segments to be queried 
are identified. This would essentially condense the queries for each datasource 
into a single query.
What do you think of this proposal?


[ Full content available at: 
https://github.com/apache/incubator-druid/issues/6057 ]
This message was relayed via gitbox.apache.org for [email protected]

Reply via email to