At 2021-10-04Z21:38 the fix for the above issue was released to production. Right at that moment you can see a pretty much linear increase in the number of alive IPv6 probes as the rollout progressed:
The number of pods with functional IPv6 went from around 3425 to 4055, a difference of around 630.
This is being noted here in data-annotations because this could have potentially caused a noticeable shift in traffic from IPv4 to IPv6.
Because this issue was caused by a race condition, it is probably not deterministic which sites/pods could have been affected at any given time, and the list probably shifted over time. That said, on the morning of October 4 (before the fix was deployed) this is the list of sites where anywhere from one to all pods were affected:
Summary of problem: https://github.com/m-lab/epoxy-images/issues/209
At 2021-10-04Z21:38 the fix for the above issue was released to production. Right at that moment you can see a pretty much linear increase in the number of alive IPv6 probes as the rollout progressed:
The number of pods with functional IPv6 went from around 3425 to 4055, a difference of around 630.
This is being noted here in data-annotations because this could have potentially caused a noticeable shift in traffic from IPv4 to IPv6.
Because this issue was caused by a race condition, it is probably not deterministic which sites/pods could have been affected at any given time, and the list probably shifted over time. That said, on the morning of October 4 (before the fix was deployed) this is the list of sites where anywhere from one to all pods were affected:
The list was generated with this query: