RSDK-8595 Unhealthy resource state

[maximpertsov]

Expose an Unhealthy state along with the related error on resource.Status.

Details

• This change is a pre-requisite for ⚠️ RSDK-7903 Include unhealthy remotes in resource list with status #4273, which addresses the issue of remotes disappearing from the resource list when they disconnect.
• While not enforced yet, the expected graph node state machine looks like this as of this change:

stateDiagram-v2
[*] --> Unconfigured
Unconfigured --> Configuring
Unconfigured --> Removing
Configuring --> Ready
Configuring --> Unhealthy
Ready --> Configuring
Unhealthy --> Ready
Ready --> Removing
Unhealthy --> Removing
Removing --> Configuring
Removing --> [*]

Loading

Depends on

• RSDK-8595 Add unhealthy state and error field api#539

[benjirewis]

LGTM! Just a final question.

[benjirewis]

[q] Any idea when this happens? Just want to get a sense of when resources are logging in the absence of a logger. Seems like a violation of my mental model of graph node loggers...

[maximpertsov]

i forgot the specifics but there is a lag between when a node is first created and when it receives an initial logger. if we remove the w.logger == nil guards from this file then we will get a nil-access panic pretty quickly.

I suppose we should use this function through the file instead of relying on the guard to get more uniform behavior -- happy to do this as a follow-up.

EDIT: something tells me the lag is specific to modular resources, since we pass logging configs to them over a unix socket after startup.