Would druid interrupt the runners when query cancelled?


I’ve got a few questions about druid’s behaviours when query cancelled.

  1. From the broker’s side, will broker tell historical nodes’ stop running queries when the query gets timeout/cancelled?

  2. One historical node may fire several runners for a certain query. Some are running and others are pending. When the query gets timeout/cancelled, the pending runners should be cleared. My question is, would historical nodes interrupt the query running runners as well?


Queries are supposed to be interrupted upon a cancellation request. We use such a feature internally on a very regular basis.

Timeouts work sometimes and sometimes don’t. In general they are good enough, but see https://github.com/druid-io/druid/issues/1415 . The bits of work on the historicals tend to honor timeouts correctly, but it is possible to get into cases where the “total time to answer a query” is well beyond the timeout limit even though no historical parts timed out.