Check volumes and integrity of "first-shutdown" pings after bug 1667798
Categories
(Toolkit :: Telemetry, task, P1)
Tracking
()
People
(Reporter: chutten, Assigned: chutten)
References
Details
We're deliberately reducing the frequency and changing the expected values within "first-shutdown" pings in bug 1667798 (to try and correct them). This bug is about following-up with some analysis to check that it's all okay. Things to check:
- Proportion of "first-shutdown" pings coming from not-first sessions should drop to essentially 0%
- Proportion of "first-shutdown" pings coming from first sessions should remain unchanged.
|
Assignee | |
Comment 1•5 years ago
|
||
Rerun this after we get some more data and before Oct 14 or so (so we can back it out before it hits beta): https://sql.telemetry.mozilla.org/queries/75313/source#188071
|
|
Assignee | |
Comment 2•5 years ago
|
||
Preliminary results of the Oct 2 builds show that the number of "first-shutdown" pings sent from non-first sessions is much reduced... but still very present.
It occurs to me that the solution doesn't stop someone whose profile is "first according to the pref" from maintaining that it's still the first profile so log as sessionstore-window-restored isn't triggered. (Since the value of false (ie, that the first session has been detected and future sessions aren't the first) isn't recorded except in that observer).
If this is the case we should see that proportion rise when new populations come awake (on Monday) but decline over time as those mistaken profiles gradually stay awake long enough to hit sessionstore-windows-restored. Once is all it would take. We should be able to see this over the course of this week, and if so I think we can just leave the code as written.
If the numbers don't go down, then we might have to look into dispatching an idle task to record the false value to the pref. An idle task is probably the best perf-conscious method of ensuring the code'll run before shutdown.
|
|
Assignee | |
Comment 3•5 years ago
|
||
The fix is verified by the data to remove most but not all erroneously-sent "first-shutdown" pings, leaving the volume of correct "first-shutdown" pings intact. It is safe to continue riding the trains, and we might consider closing the last hole of it as described in Comment 2 if, in the long run, the volumes of the erroneous "first-shutdown" pings don't diminish as expected.
Description
•