We have concluded that a seemingly insignificant database modification caused unexpected locking resulting in timeouts from the API. We apologize for this incident and have started a process to strengthen the evaluation of infrastructure changes to detect potential negative side effects.