Why do we count so many less clients sending events pings per day than other pings?
Categories
(Data Platform and Tools :: Glean: SDK, defect, P4)
Tracking
(Not tracked)
People
(Reporter: travis_, Unassigned)
Details
Reference: https://sql.telemetry.mozilla.org/queries/90374#223686
While doing some validation work on Glean Server Knobs I noticed I was getting less data than I expected and since the data was shipping on the events ping I decided to look at reasons why this might be when I discovered the referenced query above and saw that we count considerably less clients via the events ping than any other ping.
Some of this is due to sending less events pings because now we no longer send them if they are "empty" (no events). But, there's still some differences when taking that into consideration. For my experiment a test event is recorded at the same time that a newtab.opened event is recorded, but the newtab event goes in the newtab ping and the test event goes in the events ping. So, every time we got a newtab opened event from a client enrolled in my experiment, we should have also received an event ping with the test event, but we didn't: https://sql.telemetry.mozilla.org/queries/90413/source#223783
Looks like we may not be getting all of the events pings we should and we should investigate this to see what's going on.
|
Reporter | |
Updated•2 years ago
|
|
||
Comment 1•2 years ago
|
||
It would be interesting to verify (or perhaps :chutten can just confirm) that we don't see sequence holes in the events pings, a la https://bugzilla.mozilla.org/show_bug.cgi?id=1809275 and friends.
|
||
Comment 2•2 years ago
|
||
I haven't looked at this specific subpopulation regarding seq holes in "events" pings on Firefox Desktop, but when we were surveying for what a "typical" seq hole rate might be, we didn't find "events" pings in general in Firefox Desktop to be holely.
One thing that didn't occur to me at the time is we send "newtab" pings from more clients than those who open a newtab. There's also the component_init reason. If you exclude those clients, the lines do get closer... but not closer enough: https://sql.telemetry.mozilla.org/queries/91475/source#226457
Description
•