Semantic Convention for Deployments

[thisthat]

Context

I would like to propose a semantic convention for Cloud Native Deployments. This semantic convention will support the observability of deployments by enriching logs and DORA metrics with information about the application and/or microservices under deployment. Furthermore, traces could also be augmented with this data gaining visibility in all the steps the deployment goes through and pinpointing root causes of failure.

Proposed Attributes

Deployments can have one or more applications. Each application comprises one or more processes, called Workload in K8s terminology. Furthermore, deployments usually also carry additional tasks that set the stage up to be able to serve the newly deployed application(s).
This issue proposes to track name, version, and status attributes for Application(s), Workload(s), and Task(s)

Please take a look at #3169 for a more comprehensive description.

Additional Info

We started implementing this in Keptn, an incubating CNCF project. An example of the usage of this data is available here: https://github.com/keptn-sandbox/klt-on-k3s-with-argocd

[joshgav]

Nice idea @thisthat! Do you think we might inject emitters for these events into say Jenkins (via the Java instrumentation repo) and Tekton (via the Go repo)? That would help with adoption.

[thisthat]

Hey @joshgav, I am not an expert on how these repos work, but I think auto-instrumentation for these attributes is feasible. deployment.application.name is the only attribute for which there is no 100% fit since both tools do not have the Application concept. Nevertheless, we could use the pipeline name for it.

[joshgav]

Hi folks - I believe this may in fact be the same or at least closely overlap this issue: open-telemetry/oteps#223, PTAL

@AloisReitbauer 👀

[dyladan]

I think this is being handled by the CI/CD group now in the semconv repo