Add PerfMetrics to glean probes
Categories
(Core :: Disability Access APIs, task)
Tracking
()
People
(Reporter: morgan, Assigned: morgan)
References
(Depends on 1 open bug, Blocks 3 open bugs)
Details
Attachments
(1 obsolete file)
Tracking our perfmetrics in glean probes will allow us to understand the variety of performance experiences users of the a11y engine are facing.
This data will help us in a few ways:
- It'll help us prioritise existing performance work. We will prioritise bugs that are indicative of especially poor performance over those that fall close to or within "average" performance profiles for users.
- It'll help us write tests that better approximate "poor performance".
In the convo I had with the perf folks at the recent engineering summit, I was told we should target the 95th percentile of in-the-wild performance as our "poor performance" benchmark.
| Assignee | ||
Updated•17 days ago
|
| Assignee | ||
Comment 1•17 days ago
•
|
||
This patch adds Glean metrics for all of our existing PerfStats metrics, and passes them into the
existing PerfStats auto recorder constructors.
With D337305, the durations recorded by the PerfStats metrics are mirrored to Glean.
There's a small discrepancy in the units we use for recording: the Glean metrics use nanoseconds
while the PerfStats metrics use miliseconds. This is because Glean records into an integer while
PerfStats records into a double. Based on my local runs and try runs on the reference hardware,
we aren't overflowing these fields, but we do have a wide range (1ns for some probes, 10+ms for others).
Updated•17 days ago
|
Updated•3 days ago
|
Description
•