Add a host resource detector

[pyohannes]

Fixes #5398

This PR adds a resource detector for physical hosts which sets values according to semantic conventions for host resources:

• host.arch
• host.id
• host.name
• host.ip
• host.mac

[codecov]

Codecov Report

Attention: Patch coverage is 93.84615% with 4 lines in your changes are missing coverage. Please review.

Project coverage is 62.5%. Comparing base (30ed923) to head (f585990).

Additional details and impacted files

@@           Coverage Diff           @@
##            main   #5399     +/-   ##
=======================================
+ Coverage   62.3%   62.5%   +0.1%     
=======================================
  Files        189     191      +2     
  Lines      11575   11640     +65     
=======================================
+ Hits        7221    7282     +61     
- Misses      4145    4147      +2     
- Partials     209     211      +2     

Files	Coverage Δ
detectors/host/host.go	96.6% <96.6%> (ø)
detectors/host/host_id_linux.go	60.0% <60.0%> (ø)

[MrAlias]

This seems duplicative of WithHost. Why not update that option?

[pyohannes]

This seems duplicative of WithHost. Why not update that option?

Indeed it is duplicative.

@MrAlias Do you think it a feasible option to populate host.arch via WithHost, and add additional WithHostIP and WithHostMAC to opt into host.ip and host.mac (all implemented in the SDK)?

[MrAlias]

Taking a look at the host semantic conventions, it looks like they are all experimental (even host.name 😬).

I think in the long term we want to make the WithHost option include all the recommended/required stable semantic conventions for a host. But in the interim, having a detector here is probably the correct approach.

I think this PR is the correct approach. I'll plan to review.

[MrAlias]

• Missing support for BSD: https://github.com/open-telemetry/opentelemetry-go/blob/main/sdk/resource/host_id_bsd.go
• Missing fallback when the host system is not supported for the host ID: https://github.com/open-telemetry/opentelemetry-go/blob/main/sdk/resource/host_id_unsupported.go

[MrAlias]

This error should not be ignored. The test should be failed.

[pyohannes]

I adapted the tests so that the host.name is only added to the list of expected attributes when os.Hostname() doesn't return an error.

This covers cases where os.Hostname() returns an error, but a host.id might still be obtained.

I'm happy to both make the tests and detector logic fail when os.Hostname() fails, if folks think that's better.

[pyohannes]

Missing support for BSD: https://github.com/open-telemetry/opentelemetry-go/blob/main/sdk/resource/host_id_bsd.go
Missing fallback when the host system is not supported for the host ID: https://github.com/open-telemetry/opentelemetry-go/blob/main/sdk/resource/host_id_unsupported.go

Thanks @MrAlias, this should be ready for another review.

[dashpole]

But in the interim, having a detector here is probably the correct approach.

@MrAlias Should we use feature environment variables instead of a new detector module?

[MrAlias]

But in the interim, having a detector here is probably the correct approach.

@MrAlias Should we use feature environment variables instead of a new detector module?

That's a good idea.

[MrAlias]

But in the interim, having a detector here is probably the correct approach.

@MrAlias Should we use feature environment variables instead of a new detector module?

That's a good idea.

I've added this to the SIG meeting agenda to discuss.

[MrAlias]

@pyohannes can this be closed? We plan to move this into this (behind the experimental envar), right?