GV_PAGE_LOAD_MS reports times longer than 10 minutes
Categories
(GeckoView :: General, defect, P3)
Tracking
(firefox72 affected)
| Tracking | Status | |
|---|---|---|
| firefox72 | --- | affected |
People
(Reporter: Dexter, Unassigned)
References
Details
While looking at our Glean SDK error stream, we found that the GV_PAGE_LOAD_MS histogram, being reported through GeckoView->Glean in Fenix as geckoview.page_load_time, is reporting loading times longer than 10 minutes.
This could be either a bug in the GeckoView code measuring the page load time on Android, or we're really measuring > 10 minutes time on some loads.
See the comment below for context:
(In reply to Alessio Placitelli [:Dexter] from bug 1591153 comment #7)
As discussed over IRC, and seen in this query,
geckoview_page_load_timehas 59 samples in the599,512,966,122bucket, which is the bucket holding values for 10 minutes lengths. As documented, values greater than 10 minutes are cropped and reported in that bucket, in addition to generating an error.So, for the specific page load time metric, we're seeing legitimate page loads longer than 10 minutes...
Eugen, since you own this metric, is this something that should be fixed?
Comment 1•6 years ago
|
||
There might be two scenarios that can potentially lead to unreasonable high times for GV_PAGE_LOAD_MS:
- The app gets backgrounded and snoozed by Android during a page load.
- Some interplay between page load, reload and redirect.
We could prevent 2. by canceling the timer in the event of a reload or redirect.
I don't think 1. is a very realistic scenario, since we would most likely just get killed instead, which would reset the timer. I'm not really sure we have a way of working around this issue either.
Comment 2•6 years ago
|
||
Eugen suggests we compare GV_PAGE_LOAD_MS times with Gecko's other telemetry probes for page load times. Do they also show these 10+ minute page load times or is this problem specifically GV_PAGE_LOAD_MS?
Updated•3 years ago
|
Description
•