You updated Home Assistant, or an integration, and now half a room doesn't work. On Settings → Devices & services one of the integration cards is red or orange and says Failed to set up, Retrying setup, or Not loaded. Every device from that integration shows as unavailable, which is why things seem to have broken "randomly": they all belong to the one integration that didn't start.
Why it happens
An integration is the piece of Home Assistant that talks to a brand or a protocol: your inverter, your Zigbee stick, your cloud-connected air conditioner. When Home Assistant starts, each one runs its setup. If setup fails, the integration and all of its entities are left unavailable.
The common causes:
The device or service it talks to was unreachable at startup. The device changed IP address, was off, or the manufacturer's cloud was down. Home Assistant shows Retrying setup and keeps trying on its own, waiting longer each time.
An update changed something. A new version of Home Assistant changed how an integration stores its settings (a migration) and it failed, or removed an option the integration was relying on.
A custom integration (from HACS) isn't compatible with the new Home Assistant version. Custom integrations update on their own schedule, and they often lag a release behind.
The login expired. Cloud integrations need re-authenticating from time to time; after an update this often surfaces all at once.
A Python dependency failed to install, usually on an unusual install type or with no internet at the moment of startup.
Here is how our check reported one on a demonstration instance, and the detail it recorded:
The finding as it appears in the report
1 integration isn't loading properly
Affected: uptime_kuma. Each one takes its devices down with it, so this is often the cause of things 'randomly' not working.
Demonstration instance: a disposable Home Assistant with made-up entities, run on our own test bench. Not a customer's house. The text in the box is exactly what the service's check and safety rules produced on it.
The integration the check found not loaded
uptime_kuma: state setup_retry, reason: Connection to Uptime Kuma failed
Demonstration instance: a disposable Home Assistant with made-up entities, run on our own test bench. Not a customer's house. The text in the box is exactly what the service's check and safety rules produced on it.
How to fix it yourself
Read the actual reason. On the integration's card, the message under the name is often enough ("Connection refused", "Invalid authentication"). For more, open Settings → System → Logs, search for the integration's name, and read the first error after startup, not the last. The first one is usually the cause; the rest are consequences.
Check Repairs.Settings → System → Repairs often has a specific instruction for exactly this: "re-authenticate", "this option was removed", "this integration is no longer supported". If there's a repair, follow it first.
Check the device is reachable. If the reason mentions a connection, confirm the device is on and has the same IP address it had yesterday. A router that hands out a new address after a power cut is a very common cause. Give important devices a fixed (reserved) address in your router.
Reload it. On the integration's card, three-dot menu, Reload. If the cause was temporary (the cloud was down, the device was booting), this fixes it. It's cheap and safe: a reload only restarts that one integration. If the integration is still starting up, give it a minute first: reloading in the middle of setup just starts setup again.
Re-authenticate if asked. If the card says Reconfigure or a repair asks you to sign in again, do it. Your settings and entity names are kept.
Turn on debug logging if the reason isn't clear. Three-dot menu on the card, Enable debug logging, then Reload, reproduce the problem, and Disable debug logging again. The log file downloads automatically and is far more specific.
For a custom integration, update it or roll it back. In HACS, check for an update to the integration; the author has often already fixed it. If not, use Redownload and choose the previous version, then restart.
Roll back Home Assistant itself as the last resort. If a core update broke something important and there's no fix yet, restore the backup that was taken before the update. On Home Assistant OS, the update dialog offers to take one; if you ticked it, a partial restore of just the Home Assistant part puts you back. Read our backup and restore guide first, because a restore discards everything done since the backup.
Report it. If a core integration broke on an update and none of the above helps, check the Home Assistant issue tracker for that integration: someone has usually reported it within a day of a release, with a workaround.
When a reload won't help
A reload fixes a temporary failure. It can't fix a wrong password, a device on a new IP address, an incompatible custom integration, or an integration you've disabled or told Home Assistant to ignore. Those last two aren't faults at all: a disabled integration or an ignored discovery isn't running on purpose, and there's nothing to fix. Our check lists them separately so they aren't mistaken for failures.
Do it properly
Take a backup before any Core or OS update, and read the release notes' "backward-incompatible changes" section for the integrations you use. The recovery path for a bad update is the backup taken before it.
Update custom integrations before updating Home Assistant, not after.
Most integrations are set up in the UI under Settings → Devices & services, not in configuration.yaml. If a guide tells you to add YAML for an integration that has a UI, it's out of date.
The check this service applies before a reload: only for an integration Home Assistant reports as failed or retrying; never for one still starting up, one you disabled, or one you told Home Assistant to ignore; and never for an integration that owns a lock, alarm, camera, cover, valve or water heater, or whose name mentions a garage, door, gate, alarm, siren or lock, because a reload restarts every device it owns. On that demonstration instance the rule came back like this:
The safety rule for reloading it
Reload uptime_kuma -> ACCEPTED: the check can prove this change is justified
Demonstration instance: a disposable Home Assistant with made-up entities, run on our own test bench. Not a customer's house. The text in the box is exactly what the service's check and safety rules produced on it.
Or we'll do it for you — A$29 Check-up, no subscription: we reload an integration that has failed, where that's safe, and tell you plainly when it needs a new login, a new address or an update instead.
