Resolved
This incident has been resolved.
Monitoring
We've rolled back some infrastructure changes that may have allowed the registry to leak connections. Registry errors have returned to normal levels, and deploys are functioning properly in all regions.
Identified
The registry is rejecting connections from some hosts in some regions. We're restarting processes, which seems to be helping reduce the error rate.
Investigating
We are investigating an increased rate of image pull failures in a handful of regions. These are interfering with some customer deploys, and show as a failed VM in fly status.