During auto-patching of the infrastructure, the another service had locked the migration state, causing the newly patched services to hang. Lock was released 40 minutes after the incident started, and the patching completed.
We have taken measures to make sure this will not happen again.