Bugzilla

With the rollout of Fx 103 at 100% we're seeing a marked increase in http "0"s that's being tracked in https://mozilla-hub.atlassian.net/browse/DSRE-928. These represent submissions where a client initiated an HTTP request but from GCP's perspective the client disconnected before the server could fully respond and e.g. a 200 response could be returned to the client. These appear to be, just as in bug #1776201, mostly newtab pings.

I suspect this is related to the work in this bug and other followup bugs from the incident investigation. My assumption is that we'd ideally like to minimize 0s via client behavior changes and that work on this bug might decrease the number of 0s we're seeing, but that any pings that are interrupted in this way will eventually be resent and so it's not particularly concerning that we're seeing so many disconnects. If this doesn't sound like an accurate characterization of the current state please let me know and I will prioritize server side investigation of the issue.

With the rollout of Fx 103 at 100% we're seeing a marked increase in http "0"s that's being tracked in https://mozilla-hub.atlassian.net/browse/DSRE-928. These represent submissions where a client initiated an HTTP request but from GCP's perspective the client disconnected before the server could fully respond and e.g. a 200 response could be returned to the client. These appear to be, just as in bug #1776201, mostly newtab pings.

I suspect this is related to the work in this bug and other followup bugs from the incident investigation. My assumption is that we'd ideally like to minimize 0s via client behavior changes and that work on this bug might decrease the number of 0s we're seeing, but that any pings that are interrupted in this way will eventually be resent and so it's not particularly concerning that we're seeing so many disconnects. If this doesn't sound like an accurate characterization of the current state please let me know and I will prioritize server side investigation of the issue.

EDIT: :chutten has confirmed this is likely not a server side issue and the corresponding Jira ticket has been closed