I think backpressure duration should only be computed like this if we're in the 
`suspendWatermark >= 0 && resumeWatermark >= suspendWatermark` block above. Two 
reasons:

1. Access to backPressureStartTimeNs must be synchronized, since it's being 
accessed from multiple threads and it's not a volatile or atomic.
2. If we didn't actually resume anything, then we don't know if backpressure 
was actually happening (maybe, maybe not) so we shouldn't report a time. In 
this case, we should return 0.

[ Full content available at: 
https://github.com/apache/incubator-druid/pull/6335 ]
This message was relayed via gitbox.apache.org for [email protected]

Reply via email to