I have a question around druid metrics. I have also attached a screenshot of observed metrics during our load test 1. Why historical query time exceeds the broker query time? a. We observed that broker query time sometimes lower than that of historical time. If the broker query time is the time from receiving a query from a client and sending the response back, and the historical query time is the time from receiving the query from broker and sending a response back to broker, then I would expect broker query time always > historical query time 2. Why broker query node time can be > than broker query time ? I would also expect the broker query node time being a part of broker query time? 3. Does the query time includes also the time of the query being queued if the jetty is low on threads? We noticed our broker was low on jetty threads and the queue was growing, but not sure whether that time is reflected in broker query time (or any other time)

Are these averages of multiple queries running, over a one minute or larger time interval? If so then I guess I could see a case when P95 of one metric might be higher than P95 of another metric, even if the average is lower. I don't know about the Jetty thread time being included or not ... hopefully someone else can answer that.

Are these averages of multiple queries running, over a one minute or larger time interval? If so then I guess I could see a case when P95 of one metric might be higher than P95 of another metric, even if the average is lower.

We're aggregating these metrics by max values