#Custom HTTP component issue
hasty dragon
To be clear I was thinking this would apply to an override of any native integration. Its obviously most egregious with something extremely core like http. But really overriding any native integration can create confusing issues like this that end up wasting developer time and cycles.
digital iron
I assumed (probably wrongly) that, given the issue was still present in safe mode, custom_components were innocent.
https://github.com/home-assistant/core/issues/167673#issuecomment-4224611318
If true, I think we should at minimum tell users during safe mode that there are custom integrations which override native integrations. Ideally, we would ignore custom integrations alltogether.
mystic sparrow
Hold up, I believe safe mode ignored all custom components
mystic sparrow
Nope, even overrides are ignored
or at least, for normal integrations they are afaik
light pawn
So I have recreated that problem here, using 2026.3.4 http compoent as a custom component on latest HA Core versions
Safe mode makes it come up
hasty dragon
🤔
light pawn
So yeah that is not true, they probably didn't trigger it properly.
We do have that frontend is reachable test in Supervisor since a couple of releases, maybe it makes sense to have this trigger automatic safe mode? 🤔
delicate silo
In these cases we should ask them for a diagnostics dump from any integration. They all include a list of all custom integrations.
light pawn
The way Core fails in this situation is kinda interesting, http works, it's simply WebSocket which doesn't work. Supervisor doesn't detect this since we check the API via http only. Maybe it would make sense to also check WebSocket 🤔
Anyhow, this brought two/three WebSocket related issues/oddities on Supervisor side to light: @hasty dragon https://github.com/home-assistant/supervisor/pull/6725 👀
light pawn
In these cases we should ask them for a diagnostics dump from any integration. T...
Also, do we have common instructions we share to trigger safe mode? Seems users didn't properly went into safe mode even though we asked.
delicate silo
hasty dragon
In these cases we should ask them for a diagnostics dump from any integration. T...
That's fair but the problem in this case is the frontend didn't come up. The users on that issue/thread didn't even know how to get us logs let alone a diagnostic dump
We were kind of stuck trying to puzzle it out from a screenshot someone took of the monitor they connected to their HA device showing the logs
But yea, we can at least point them to the safe mode link next time to ensure they are properly in safe mode. Then they should've been able to get us diagnostic data
I still think it might be worth some kind of repair or warning though alerting users to the dangers of what they're doing and why its likely to result in update issues
delicate silo
They were able to downgrade / restore to 2026.3.x. They can give us a diagnostics dump from that version too. 🙂
hasty dragon
good point lol
delicate silo
I think a repair issue for custom integrations that override integrations that have integration_type: "system", sounds good. Overriding those should not be done unless you're testing during development.
fallen sundial
I'm not strictly opposed to a repair, but also don't think it's that important here. We don't really run into issues like this often, and it would've been solved if the user in question enabled safe mode properly. So I would rather focus on making that both easier to enable and easier to see if it's actually enabled, as that helps with diagnosing all other custom integrations problems too.