We have to limit the number of queries that Looker can run on our presto cluster, to prevent it getting overloaded.
However, this means that if someone cancels a query in looker, the query
CALL system.runtime.kill_query(query_id => '20180813_181253_01046_d77wf') will get queued behind the other queries, and so won’t cancel the query immediately and prevent it taking up resources.
Is there a way around this? We have to put all of the looker queries in the same queue at the moment, because they run as the same user, but could the cancel queries be run as a different user? Or could Looker kill queries by the Presto API
DELETE /v1/query/<query-id> instead?