@cap’n Jun 13, 2026

If your dashboard said Running while PocketHost was rebooting pieces of itself in the background, that was fair skepticism. Instance status used to be a best guess scattered across Mothership, the edge, and whatever happened to survive a partial restart.

We tightened that up. Status on the dashboard should track reality much more closely now, including when only Mothership or only the edge bounces.

What was wrong

power and status are different knobs. Power is your intent (on or off). Status is runtime truth (idle, starting, running).

After an edge reboot, containers were gone but Mothership might still say Running. After a Mothership reboot, the database reset to Sleeping while your instance kept serving traffic on a warm edge. Neither case felt great in the UI.

We already shipped power off that actually stops containers. This work finishes the other half: keeping status honest across restarts without fragile per-instance chatter.

The new contract

Think of it as a simple handshake:

Edge comes back — it wipes local containers, tells Mothership everything is idle, and reloads its mirror cache. One round trip.

Mothership comes back — it assumes idle until the edge speaks up. The edge reconnects over SSE, reports which instances are actually warm, and Mothership flips only those rows back to starting or running.

Edge stays up, Mothership recycles — your containers never stopped. On reconnect the edge sends its live instance list in a single POST /api/mirror call. Mothership applies the updates server-side and returns a fresh mirror dump. The dashboard hears each change over PocketBase realtime, same as any other record update.

We deliberately do not reload the edge mirror from a cold idle snapshot on reconnect. That snapshot would lie for a few hundred milliseconds while warm instances were still running. The edge keeps serving from memory and patches Mothership upward instead.

When status can still lie briefly

If the edge goes down hard, Mothership may show Running for a bit with nothing behind it. We accept that. There is no witness process inside every container yet. When the edge returns, it resets non-idle rows and the UI catches up.

That edge-down window is the main known gap. A heartbeat lease is on the icebox if it becomes annoying in production.

What you should notice

• Recycle Mothership during normal hosting: powered-on instances that never stopped should return to Running, not get stuck on Starting.
• Edge maintenance: instances should settle to Sleeping or Idle until traffic wakes them again.
• Fewer “refresh the dashboard to see the truth” moments after infra hiccups.

This also unblocks the next platform thread: Mothership PocketBase v0.39 and eventually splitting the control-plane app from the hosting CLI. Runtime status needed a clear owner story first.

Questions or weird status after a deploy? Discord as always.