Make node name a default resource on spans

[fcevado]

For applications with multiple instances it's a pretty useful info to know where a trace was generated.
I'd suggest to set it as a default resource or allowing to pass mfa to resources configuration.

[bryannaegele]

Hey @fcevado, you can set these values for the resource via env var which is the easiest way to go about it imo. The env var is OTEL_RESOURCE_ATTRIBUTES and you can set any of the values listed in these docs. https://github.com/open-telemetry/opentelemetry-specification/tree/main/specification/resource/semantic_conventions

For example:

OTEL_RESOURCE_ATTRIBUTES: >
    service.name={{ .Release.Namespace }},service.version={{ .Values.image.tag }},
    container.name={{ .Chart.Name }}, container.image.name={{ .Values.image.repository }},container.image.tag={{ .Values.image.tag }},
    k8s.cluster.name={{ .Values.environment }},k8s.deployment.name={{ .Release.Name }},k8s.namespace.name={{ .Release.Namespace }}

[bryannaegele]

Building on your earlier comment, it could definitely be added under a different attribute by default if that's helpful to folks.

[tsloughter]

Yea, I'm down with adding a new attribute. I think it'd still be useful to also set service.instance.id, but if we are adding a node name attribute I think it would be preferable if we could set service.instance.id to the nodename if we can detect it is going to be unique automatically. It won't be perfect but simply not using it if its nonode or @localhost may be good enough to make the value useful to most people.

And I mean useful in the sense that APMs will use this attribute and not our node name attribute, so trying our best to make it something meaningful goes a long way I think.

[fcevado]

Do you know what the default is for like a generated Phoenix release?

@tsloughter for releases using mix release it gonna be <appname>@<hostname> for sure. on k8s <hostname> would be the podname. iirc on simple containers network <hostname> would be the container name, but it's been a while that i don't work with those. i'll try to run some scenarios here just to make sure.
for releases using distillery iirc the rules are the same.

if it's ok i can start working on a PR with a initial approach on that and make adjustments as we figure out the proper way to work with those. what are your thoughts?

[fcevado]

Based on the docs I guess if we detect there is no nodename we should make a uuid

@tsloughter thinking on not adding new dependencies and since uuid in the spec is just an example, wouldn't it be better to use rand:uniform just as it's already using for trace id and span id?

[tsloughter]

A PR would be great. And yea, I think we can probably get away with just a random number.