The issue causing customer instances to restart has been fully resolved across all regions.

We identified the root cause as a misconfigured logging infrastructure that was keeping stale files open. This caused some VMs to slowly fill up and create conditions that caused our orchestration system to proactively reschedule timescaledb workloads. This rescheduling caused affected customer databases not configured with HA replication to experience 5-10 minutes of downtime. Since we were able to resolve the problem before all nodes were affected, this was isolated to a small fraction of our customers.

We have now restored storage on all nodes and deployed a fixed version of the logging infrastructure to prevent recurrence.