- The application process can only run one active node at a time.
- All nodes are running already running the application process.
- The application process only writes when receiving requests.
TODO
- Pause in flight requests incoming to proxy
- Wait the sync timeout to make sure nodes caught up
- Send SIGTERM to active node to flush
- Kill active node's sync
- Treat inactive as active, route the requests to it
- Configure tail node to sync to what was originally the active node
- Enqueue previous active node to the upstream hosts
- Finished