Staff opening a cafe want every PC on without walking the floor. Wake-on-LAN does that: a “magic packet” containing a machine’s MAC address, broadcast on the local network, tells a powered-off PC to boot. The catch is in the word local.
The cloud can’t send a local broadcast
The CR Cafe backend lives in the cloud, and a magic packet doesn’t cross the internet into someone’s LAN. Something inside the cafe has to send it. Usually that’s another station agent that is already on — but first thing in the morning, when every PC is off, there is nothing on the network to send it.
The relay
So we built CR Cafe Relay: a tray-only Electron app for the always-on counter PC. It keeps a connection to the backend and broadcasts wake packets when asked. It isn’t a station — no lock screen, no licence, never counted against a plan’s station limit.
The bug that looked like success
The first design chose one sender for each wake. Enrollment codes are cafe-wide, so a relay could be installed from a different building. When it won the pick, it broadcast every wake onto its own network, the PCs never woke — and the dashboard still said “Wake signal sent”, because the server reported success as soon as it had emitted the message.
Fan out, then confirm
- A wake now goes to every relay and every online agent in the cafe. Magic packets are idempotent, so extra senders cost nothing
- Each sender acknowledges with a result. The first success wins; a failure only shows once every sender has answered and all failed
- A timeout reports that the signal went out but wasn’t confirmed, rather than failing — older relays that don’t acknowledge still work
- The dashboard lists every connected relay and flags one that appears to be on a different network
Setting up Wake-on-LAN on cafe PCs
Software is only half of it. For wakes to work, each PC needs Wake-on-LAN enabled in the BIOS or UEFI and on the network adapter, and should be connected by Ethernet — most Wi-Fi adapters can’t wake from a full shutdown. The relay machine should stay on and on the same LAN as the stations.
The same “confirm, don’t assume” principle runs through the rest of the real-time layer; see building real-time station control with Socket.IO.