This is a bug in our infrastructure. Your VM got stopped when a particular host had capacity issues. Since your volume was on that one exact host, you couldn’t boot a new VM.
Switching to a dedicated CPU actually evicts other VMs on that host running shared-cpus. It seemed like a good idea when we initially built this, but now I believe we’re better off making new VMs fail. It’ll take us some time to redo this plumbing, but it’s a high priority.
The “real” answer in our infrastructure is to run >=2 VMs for max redundancy. Which obviously doesn’t work (yet) with sqlite.