A Swarm spanning two buildings

When people hear “container orchestration at home” they assume the big-company stuff, and then they assume suffering. We run Docker Swarm — the modest, older alternative — and the honest pitch is: it’s the version you can hold in your head.

The idea in one sentence: several machines join up as a team that acts like one big computer. You tell the team “run this app,” and the team decides where. If a machine dies, its apps are simply restarted on a survivor — usually before anyone notices. Kill a machine mid-use and the service is back before you’ve finished feeling clever about it.

One front door that can’t be knocked down

Here’s my favorite mechanism. The whole lab answers at one internal address — but that address isn’t nailed to any single machine. It’s more like a crown: whichever healthy machine currently wears it answers the door, and if that machine goes down, the crown automatically passes to a survivor. Nothing to reconfigure, nothing to remember at 2 AM.

Standing at that front door is a doorman service that handles two chores for every app behind it: encrypted connections (the padlock in the browser, with certificates that renew themselves) and one shared login in front of everything. No app in the lab ever ships its own bolted-on password screen again — the door handles it once, for all of them.

Apps as recipe cards

Every service deploys from a short recipe file kept in version control: what to run, how many copies, what name it answers to. Adding an app to the lab is writing a dozen lines and handing them to the team. Rolling back is restoring yesterday’s file. The running lab is, quite literally, whatever the folder of recipes says it should be — which changed how the whole thing feels. There’s no mystery machine state anymore; there’s a folder you can read.

The rule: precious data doesn’t ride the machinery

The most important decision is what we kept out. The databases, the lab’s internal phone book, the family’s git server — the things that hold irreplaceable state — do not run on the swarm. They live on plain, boring, dedicated machines with real disks and real backups. The orchestra is great at shuffling replaceable things around; it is a terrible custodian for one-of-a-kind things. So: replaceable services get the acrobatics, precious data gets a permanent home, and the doorman fronts both so nobody using the lab can tell the difference.

And one piece of honest math we designed around instead of denying: two of the team’s three “managers” live in the main house, so if the whole house goes dark, the team pauses. That’s exactly why nothing you’d need during an outage — the phone book, the database — depends on the team at all. The next entry is about that.

Verdict

For a two-building, one-family data center, this hit the sweet spot: real resilience, recipes simple enough to read, and a front door that hands out encryption and logins like utilities. The lab stopped being a pile of containers and became a platform — one whose entire definition fits in a folder and whose entire mental model fits in a head.