I finally put the durable service stack through a real two-worker failure test today. I had the web app and PostgreSQL running over HTTPS, then paused the owning worker while its Firecracker VM stayed alive. The controller pulled the route and refused recovery until that old VM was cleaned up. The stack came back on the other worker with the same database canary. I also turned on backup freshness alerts and per-project snapshot retention, and checked that pruning an old point leaves recovery points for other projects alone. Next I want to keep the ops path simple and test the unreachable-host case when I have a dedicated machine for it.
commit 559c7ec