Zigbee2MQTT restarted 22,112 times, and Docker never told me

During a routine check of my server, one line in docker ps looked off:

zigbee2mqtt   koenkk/zigbee2mqtt:latest   Up 1 second

“Up 1 second” on a server that has been running for days means the container is dying and restarting. The restart counter made it obvious:

$ docker inspect zigbee2mqtt --format '{{.RestartCount}}'
22112

Twenty-two thousand restarts in about three days, one every ~14 seconds, ever since the last reboot. My Zigbee temperature sensor hadn’t reported anything to Home Assistant the whole time, and nothing had warned me.

The error

The logs repeated the same loop:

z2m: Connecting to MQTT server at mqtt://mosquitto:1883
z2m: MQTT failed to connect, exiting... (getaddrinfo EAI_AGAIN mosquitto)

EAI_AGAIN is a DNS failure: the container couldn’t resolve the name mosquitto. Containers in the same Compose project normally find each other by service name through Docker’s built-in DNS. That only works if they’re attached to the same network.

The cause: a container with no network

$ docker inspect zigbee2mqtt --format '{{range $k,$v := .NetworkSettings.Networks}}{{$k}} {{end}}'

$ docker inspect mosquitto --format '{{range $k,$v := .NetworkSettings.Networks}}{{$k}} {{end}}'
homeassistant-stack_default

Mosquitto was on the stack’s network. Zigbee2MQTT was attached to nothing. Its config still said it belonged to homeassistant-stack_default, but after the host rebooted, the network attachment was gone. Every restart reused the same broken container, so restarting could never fix it.

The fix

Recreate the container from the Compose file:

cd ~/dockerapps/homeassistant-stack
docker compose up -d --force-recreate zigbee2mqtt

--force-recreate throws away the broken container and builds a fresh one with the network attached properly. Data is safe because it lives in the mounted ./zigbee2mqtt folder. Within seconds:

z2m: Zigbee2MQTT started!
z2m:mqtt: MQTT publish: topic 'zigbee2mqtt/0xa4c1...', payload '{"humidity":61.7,"temperature":25.4,...}'

The sensor reported immediately.

How to catch this earlier

The real problem was that nothing noticed for three days. A few cheap checks:

• Look at restart counts, not just “Up”. This one-liner lists any container that has restarted at all:

  for c in $(docker ps -q); do docker inspect $c --format '{{.RestartCount}} {{.Name}}'; done | sort -rn | head

• Monitor the thing you care about. I now watch the sensor itself: if Home Assistant hasn’t seen a new temperature in an hour, something upstream is broken.
• Check after every reboot. Network attachment problems show up at boot, so a quick docker ps after a reboot catches them.
• Cap your logs. A crash loop also writes the same startup log thousands of times. This one had grown to 1.1 GB. Here’s how I limited Docker logs.

Takeaway

restart: unless-stopped is great at keeping a service running. It’s also great at hiding a service that’s broken. If a container’s state looks fine but the restart count keeps climbing, recreate it before you start debugging the app itself.