I recently interviewed there and they said they're still using Postgres master-slave setup most likely on some cloud provider. Could be catastrophic failure with multiple shards going down at the same time. This might also explain why they mentioned their "backups" not working