leventov commented on issue #7087: Online aggregation / Interactive Query Session Protocol URL: https://github.com/apache/incubator-druid/issues/7087#issuecomment-498342943 An additional idea here (from [this paper](https://people.eecs.berkeley.edu/~brewer/Giant.pdf) by Eric Brewer, "Graceful degradation" section) is that as soon as there are first results ready for a query on a Broker, it can allow the next query to start in parallel. If the current limit of parallel queries on a Broker is X, we can make the limit of "parallel already partially computed queries" 3X or 5X, for example. This may help in case of Broker's overload so that it returns *something* useful to more clients faster (but they may wait for longer for complete results). This will also require different prioritization of per-segment queries on Historicals. The effectiveness of running some `NCPU * 5` processing threads may be reduced through unless we [move to async data processing](https://groups.google.com/d/msg/druid-development/WPfgzN3Y5i0/sgV2QoE4EQAJ) (at least in the part of the system where Broker aggregates partial results from Historicals).
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
