My 10-minute homelab health check (and what it caught)

Dashboards are great, but they show what you configured them to show. Every now and then I run a manual check that looks for the things nobody is watching. The first time I did this properly, it found three real problems on a server I thought was fine.

Here’s the checklist, in the order I run it.

1. Containers that are quietly crash-looping

docker ps showing “Up” isn’t enough. A container stuck in a restart loop is “Up” for a few seconds at a time.

docker ps -a --format '{{.Names}}\t{{.Status}}' | sort
for c in $(docker ps -q); do
  docker inspect $c --format '{{.RestartCount}} {{.Name}}'
done | sort -rn | head

Look for: “Up 1 second”, “Restarting”, unexpected “Exited”, and any restart count above zero.

Found: Zigbee2MQTT at 22,112 restarts, crash-looping since the last reboot. Full story.

2. Failed system services

systemctl --failed
systemctl list-timers --all | head -20

Found: a certbot timer that had been failing every 12 hours for months. Why.

3. Monitoring alarms

If you run Netdata, its API lists active alarms without opening the UI:

curl -s localhost:19999/api/v1/alarms | python3 -c \
  'import sys,json; a=json.load(sys.stdin)["alarms"]; print(len(a), "active"); [print(k) for k in a]'

4. Temperatures

sensors | grep -E 'Tctl|Composite'   # CPU and NVMe
nvidia-smi --query-gpu=temperature.gpu --format=csv,noheader

My normal idle values: CPU ~38 °C, NVMe ~40 °C, GPU ~33 °C. Write yours down once, so you can tell when something’s off.

5. Disks and what’s eating them

df -h / /mnt/data
docker system df
journalctl --disk-usage

Found: 2.4 GB of container logs with no size limit. How I capped them. Also, docker system df claimed 32 GB reclaimable, which turned out to be ~3.

6. Did last night’s backup run?

tail -3 ~/backup-all.log
ls -lt /mnt/data/backups | head

The last “complete” line should be from last night. My backup script.

7. Error counts per container

for c in $(docker ps --format '{{.Names}}'); do
  n=$(docker logs --since 24h $c 2>&1 | grep -ciE '\berror\b|fatal|exception|panic')
  [ "$n" -gt 0 ] && echo "$n $c"
done | sort -rn

Most of what shows up here is noise, and that’s the point: you learn your baseline. On my server, Home Assistant logs ~2,000 harmless Bluetooth scanner errors a day. I know that’s normal, so a different error stands out.

Found: Prowlarr failing to sync an indexer several times a day. The fix.

8. Write the baseline down

The most useful output of the whole exercise is a short note: normal temperatures, normal disk usage, which log errors are expected. Next time, comparing against it takes minutes instead of an hour.

How often?

After every reboot (network attachment problems show up at boot), and otherwise once a month. Real monitoring catches the loud failures. This catches the quiet ones.