It's because of the constant churn in Google's shared infrastructure. I've been the maintainer for production services left orphaned by departing ex-Googlers.

It's because of the constant churn in Google's shared infrastructure. I've been the maintainer for production services left orphaned by departing ex-Googlers. They usually go down within 3-6 months of their previous maintainer departing - either there's some PCR because the cluster they're homed in is going down for maintenance or transferring to another datacenter, or their dependencies migrate to a new RPC format, or somebody sticks a new quota on internal requests for a service, or your PA loses machine budgets, or your RPC deadlines are now more than the average latency for that service. It's stupid stuff like that - assumptions that worked a few years ago don't work anymore when the infrastructure is a moving target. Google's code is immensely complex, so once the institutional knowledge goes out the door, debugging and fixing problems becomes very difficult. Your maintainer, once the original author has left, is probably doing this as an unfunded 20% project; she will get zero credit on her performance review for making sure that a service her VP doesn't care about stays up. So it's usually easier to do nothing and just let it start failing. Announcing a turndown is actually a courtesy in this case.