Skip to Content

Home Assistant: from Raspberry Pi to Docker

A migration of under an hour, a revised verdict, a six-month fault finally understood, and monitoring that was creating the fault.
November 15, 2026 by
Home Assistant: from Raspberry Pi to Docker
Federico Solutions, Myrko Federico

Home Assistant: from Raspberry Pi to Docker

Field: home automation, migration, operations

Period: August 2026

Client: internal project, my own home (Lab 77, R&D)

My home automation had long been running on a Raspberry Pi, with the official Home Assistant OS image. In August 2026, I moved it to the house server, as the Docker version. The migration itself took less than an hour. What came before and after taught me more: a revised verdict, a six-month fault finally understood, and monitoring that was creating the very fault it was meant to detect.

Context

I described my whole setup, open source home automation and AI, in a dedicated article. I will not repeat it here. This story covers a single project: taking Home Assistant off its small dedicated computer to run it alongside the other household services, on a more powerful machine that already hosted speech recognition and speech synthesis.

Problem

  • Everything on one small machine: the dedicated machine carried the home automation, but also the DNS filtering for the whole house. A single restart of the Pi had been enough to cut the whole house off from the internet.
  • A temperamental voice chain: for six months, the voice assistant had answered badly or not at all, with no cause found.
  • An installation choice: keep the official image in a virtual machine, or move to the Docker version, lighter but without the add-on store.
  • A hard constraint: reconfigure no device in the house.

Solution

1. Revise the verdict when the facts change

My first study concluded in favour of the virtual machine, on one decisive argument: a Home Assistant OS backup could supposedly only be restored into Home Assistant OS. When I checked the official documentation, that argument no longer held: the backup integration restores across all installation types, and each add-on is in fact a published Docker image. So I revised the verdict: Docker removed the two riskiest steps of the plan, and used far fewer resources.

2. Take DNS out of the home automation

DNS filtering is a service whose failure is felt throughout the house. I took it out of Home Assistant and made it a standalone service, which no longer depends on the home automation to start.

3. Keep the same address, and an intact way back

The new server took over the network address of the old Pi: no device needed reconfiguring. The Pi itself was switched off without being modified. Turning it back on returns the house to its previous state. A fallback script stops the home automation while keeping DNS running.

Result

The switchover took less than an hour, with a DNS outage of about three minutes; Home Assistant was responding ten seconds after starting. Six fixes had to be found straight away; the system then ran with 52 loaded integrations and 387 entities. Home Assistant uses about 320 MB of memory, where a virtual machine would have reserved about 4 GB.

What I take away

  • What the platform protected is no longer protected once you leave it. Several add-ons relied on the Home Assistant OS supervisor for authentication. Once on Docker, one service restarted with no data, because it had not been restored, and some access settings had to be reconfigured. I went through everything one by one.
  • The six-month fault came from the Pi's network setup. Connected to two networks at once, it received a request through one and replied through the other, so the reply got lost. It was neither the firewall nor the routing. Reproducing that setup on the new server would have moved the bug. A single network connection was enough.
  • The monitoring that created the fault. After the migration, a watchdog checked DNS every minute and restarted it if it did not answer. It fired 24 times in three days, always at the same minute of the hour. No real failure: every hour, the DNS filter reloaded its lists and stopped answering for a few seconds, the watchdog thought it was dead, restarted it, and caused a real outage for the whole house. The restart even reset the next reload to the same minute. Requiring two consecutive failures and a slightly longer wait fixed everything.
  • Regularity to the second points to a clock. No overload always comes back at the same minute. Before fixing a periodic fault, I first look for whatever has the same period.
  • A first cause is not necessarily the right one. I had announced a cause, and fixed it, before the next event disproved it. The fix was useful, but it was not the fault. A single observation window is not enough to conclude.

Going further

Do your home automation, your network or your small servers rely on setups nobody understands any more? I diagnose, migrate with minimal downtime, and document what has been done. See your tools and your data, or book a 20-minute discovery call.

Sources

Tags: Home Assistant, home automation, Docker, Raspberry Pi, migration, DNS, monitoring, lessons learned

What your premises tell the street: Wi-Fi and Bluetooth
An open source, offline Android app that passively records what nearby devices broadcast: a first surface audit, without intercepting anything.