[MESOS-9237] Add http handler metrics in libprocess.

• Type: Improvement
• Status: Accepted
• Priority: Major
• Resolution: Unresolved
• Component/s: libprocess, metrics
• Assignee: Unassigned
• Reporter: Benjamin Mahler
• Created: 15/Sep/18 21:39
• Updated: 15/Sep/18 21:40

Description

In a similar spirit to MESOS-9235, we can introduce per-route http metrics in libprocess to get a better sense of how long responses take to complete:

• The biggest window of time that we can measure is: time from parsing a request, to finishing {{send()}}ing all of the data for its response.
• An interesting subset of this time is how long it took from executing the handler to the response future being ready.
• Other interesting subsets of the overall time is how long it took us before we could execute the handler (e.g. routing delay, actor queueing delay), and how long it took us to send it to the os (e.g. I/O saturation, backpressure from a slow client, etc).

These metrics would ideally have histogram like distributions (e.g. percentiles over a window).

Issue Links

• supercedes: MESOS-7819 Libprocess internal state is not monitored by metrics.