[Obs AI Assistant] Error shown in knowledge base at startup

[viduni94]

An error is shown at the start on AI Assistant with regards to the knowledge base and shows the Inspect issues link. When Inspect issues is clicked, the error says that the model is not deployed.

• Initially we shouldn't show this error because at startup the only error is about the inference endpoint because the knowledge base is not installed already

"resource_not_found_exception
	Root causes:
		resource_not_found_exception: Inference endpoint not found [obs_ai_assistant_kb_inference]"

• In the shown error modelId is not populated (refer screenshot)

Steps to reproduce:

1. Start ES and Kibana
2. Open the AI Assistant flyout
3. The error will be shown within Inspect issues

Expected behavior:

• There shouldn't be an error at startup
• Wherever this error is relevant and shown because the model is not deployed, the correct modelId should be populated.

Screenshots (if relevant):

[elasticmachine]

Pinging @elastic/obs-ai-assistant (Team:Obs AI Assistant)

[neptunian]

#205970 will not address this.

I don't know that this is not intentional. I believe this only happens when you have a pre configured connector. I guess it's not really an error, just giving you some info that your model is not deployed. Not the best UX, but doesn't seem harmful either. I might be wrong but I think the main user journey is to setup a connector first in which case we auto install the knowledge base and they won't see this.

CC @teknogeek0

[viduni94]

I think the main user journey is to setup a connector first in which case we auto install the knowledge base and they won't see this.

Sounds good, thanks for the info @neptunian

[viduni94]

Once the PR for #205970 is merged, I'll review this issue and see whether we would need any changes to improve the UX here.

[sorenlouv]

@viduni94 One more thing you might want to take a look at, is how we handle timeouts. I've seen errors when the /setup endpoint takes more than 2 minutes to respond. I think this happens when there are no ML nodes or when we need to scale from 0. It might be hard to reproduce locally though.

[viduni94]

@viduni94 One more thing you might want to take a look at, is how we handle timeouts. I've seen errors when the /setup endpoint takes more than 2 minutes to respond. I think this happens when there are no ML nodes or when we need to scale from 0. It might be hard to reproduce locally though.

Sure @sorenlouv
I'll take a look at it.