Release version 113 spike in perf.page_load events
Categories
(Fenix :: Browser Engine, defect)
Tracking
(Not tracked)
People
(Reporter: travis_, Assigned: travis_)
Details
Noticed this in our weekly meeting, with the release of 113 there seems to be a noticeable spike in perf.page_load events.
ref: https://sql.telemetry.mozilla.org/queries/92095/source#227885
This may be expected, but it certainly seems like a lot of events for over the weekend. We noticed this because of an increase in an error we track in our SDK weekly meeting and I figured it was best to pass this along quickly.
Comment 1•1 year ago
|
||
113 has recently entered the release channel so the number of users has spiked dramatically and we have many more page load events. The overall number of events for 113 in release roughly matches the number for 112, so I don't think this spike in unusual. You can see a similar spike for 112 as well here https://sql.telemetry.mozilla.org/queries/92118/source#227929
Assignee | ||
Comment 2•1 year ago
•
|
||
I could understand that many pageload events in desktop, considering the population, but this is also spiking on the order of 2.5 billion events per day in Fenix. That's where I initially saw this. Fenix reports 3-4 times as many of these events as desktop.
Assignee | ||
Comment 3•1 year ago
|
||
In version 111 we fixed an issue that was forcing custom pings to be sent on startup, and then later in version 113 true support for sending custom gecko pings in Android was added.
Bug 1811872 - Investigate precipitous increase in glean.restarted
InvalidState errors starting in Fx111 (Glean v52)
Bug 1737520 - FOG/GeckoView: Enable custom ping support
Thanks to chutten for helping me identify this, but this seems likely to explain what we see with the data here. v110 was sending data, but it really wasn't supposed to be. That was fixed by Glean in v111 but the ability to send the custom pageload ping from Geckoview wasn't enabled until v113, when it would have flushed any events that had been accumulating in v111/v112
v110 was sending custom pings at startup but not intentionally. This was corrected in v111. Then in v113 the real ability to send custom pings in Android (from Gecko) was added, and thus we got all the backed-up pings from v111/v112...
This accounts for the events we got in 111 before the patch landed in m-c to update Glean, and no pings at all from 112 until the other patch landed in v113.
Action items from this will be to annotate this metric with this information in the Glean Dictionary, as well as adding a note in the historical analysis "gotchas" section of docs.telemetry.mozilla.org.
Closing this as FIXED
Description
•