Increase the number of buckets in the queue latency histogram

[nicktindall]

Previously the queue latency histogram had 18 buckets, which meant any percentile over ~65s would be calculated as Long.MAX_VALUE ms (292,471,208 years).

This change adds two more buckets to the histogram allowing us to represent values up to ~ 4 minutes 20 seconds, hopefully covering the vast majority of queue latencies we're likely to see.

[elasticsearchmachine]

Pinging @elastic/es-distributed-coordination (Team:Distributed Coordination)

[mhl-b]

Can we manually assign buckets? Current structure has very low granularity at milliseconds and coarse at seconds. While we mostly look at seconds latency. Something like [10ms,100ms,500ms,1s,2s,4s,8s,10s,15s,20s,40s,60s,60s+]?

[nicktindall]

Can we manually assign buckets? Current structure has very low granularity at milliseconds and coarse at seconds. While we mostly look at seconds latency. Something like [10ms,100ms,500ms,1s,2s,4s,8s,10s,15s,20s,40s,60s,60s+]?

Yeah we'd need to implement a different type of histogram

[mhl-b]

Ouch, right. I thought we can pass custom ranges. LGMT

[nicktindall]

Ouch, right. I thought we can pass custom ranges. LGMT

I did consider making a arbitrary bucket histogram at the time, but the exponential one already existed. I think it'd be good going forward to have both, but not as part of this change.