You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Apr 3, 2019. It is now read-only.
From the discussion in mozilla/fxa-activity-metrics#16 (comment), although the skew-correction code for client-side flow event timestamps appears to work well in testing, it is not solid enough to drive our metrics pipeline. Instead the server should set timestamps on all flow events, not just begin.
Unfortunately, because of the way metrics payloads are batched, I suspect this means adding a dedicated endpoint for flow events, to which we send data as soon as events occur. I don't think we can reliably lean on speed-trap's offset property because it leaves us at the mercy of the client, which could literally be anyone.
The text was updated successfully, but these errors were encountered:
Unfortunately, because of the way metrics payloads are batched, I suspect this means adding a dedicated endpoint for flow events, to which we send data as soon as events occur.
I was wrong, just realised there is a flush method we can use when it's a flow event, duh.
@vladikoff, would it be any better if we sent the flow events to a dedicated endpoint? Or is your objection more about server load generally rather than the metrics endpoint specifically?
Just wondering whether I should persevere with this issue or kill it.
Had a chat about this in the meeting just now, we think the validation just added should be sufficient to rule out most cases of the issue. Killing this, but we can bring it back if there are problems in the future.
From the discussion in mozilla/fxa-activity-metrics#16 (comment), although the skew-correction code for client-side flow event timestamps appears to work well in testing, it is not solid enough to drive our metrics pipeline. Instead the server should set timestamps on all flow events, not just
begin
.Unfortunately, because of the way metrics payloads are batched, I suspect this means adding a dedicated endpoint for flow events, to which we send data as soon as events occur.I don't think we can reliably lean on speed-trap'soffset
property because it leaves us at the mercy of the client, which could literally be anyone.The text was updated successfully, but these errors were encountered: