Long document with notes: https://docs.google.com/document/d/1DAwLvF5F3cT1dG85VA2zC7ANPx2EZxPb9vL0RwaWIV0 I did another query last night on the number of pings with sketchy subsession lengths (between 24 and 48 hours and more than 48 hours): https://sql.telemetry.mozilla.org/queries/61420/source It hovers around 1% for most (up to 6% of pings, on release, on a 1% sample of all pings). I also took :wlach's query and adjusted it for actual client counts: https://sql.telemetry.mozilla.org/queries/61419/source It shows similar small numbers. While we certainly do have clients with invalid session lengths, IMO they are much less significant than initially thought. We might still need to be careful when basing analysis on session length/subsession hours sum (active ticks/active hours seems to be more accurate anyway) I went ahead and submitted the most minimal patch fixing the monotonicNow/Date.now/-1 bug. I left out any instrumentation, as I think if we want that we need to implement it much lower (in TimeStamp directly), as Telemetry is most likely not the first/only caller of this. :wlach, does this make sense? do you agree with my conclusions?
Bug 1514392 Comment 16 Edit History
Note: The actual edited comment in the bug view page will always show the original commenter’s name and original timestamp.
Long document with notes: https://docs.google.com/document/d/1DAwLvF5F3cT1dG85VA2zC7ANPx2EZxPb9vL0RwaWIV0 I did another query last night on the number of pings with sketchy subsession lengths (between 24 and 48 hours and more than 48 hours): https://sql.telemetry.mozilla.org/queries/61420/source It hovers around 1% for most (up to 6% of pings, on release, on a 1% sample of all pings). I also took :wlach's query and adjusted it for actual client counts: https://sql.telemetry.mozilla.org/queries/61419/source It shows similar small numbers. While we certainly do have clients with invalid session lengths, IMO they are much less significant than initially thought. We might still need to be careful when basing analysis on session length/subsession hours sum (active ticks/active hours seems to be more accurate anyway) I went ahead and submitted the most minimal patch fixing the monotonicNow/Date.now/-1 bug. I left out any instrumentation, as I think if we want that we need to implement it much lower (in TimeStamp directly), as Telemetry is most likely not the first/only caller of this. :wlach, does this make sense? do you agree with my conclusions? (Edit: Oh and if you do, you could abandon your patch :))