Add experimental span attribute job.system

[trask]

Introduces experimental job.system span attribute from open-telemetry/opentelemetry-specification#1582

[motivation: we're mapping scheduled jobs to "top-level" operations (similar to SERVER/CONSUMER spans, but scheduled jobs are INTERNAL), and I need a way to identify them that doesn't involve checking the InstrumentationScope, since the scope isn't available in samplers]

[mateuszrzeszutek]

We should probably include Spring Batch too.

One more question: this kinda duplicates the instrumentation/scope name, is that not usable for you?

[trask]

We should probably include Spring Batch too.

will do

One more question: this kinda duplicates the instrumentation/scope name, is that not usable for you?

oh ya, I should have mentioned, I want to detect now in a sampler, which doesn't have scope (yet) 😞

[julealgon]

Question on this: shouldn't it be a resource attribute instead of a span attribute?

We already have these today for identifying the web Engine:

• https://opentelemetry.io/docs/specs/semconv/resource/webengine/

With attributes:

• webengine.description
• webengine.name
• webengine.version

It feels to me like the same structure would apply to a job scheduler/runner.

[trask]

hi @julealgon! I don't see webengine.* attributes in this PR, can you clarify the relation to this PR, or even better probably open a new issues with all of the context? thanks

[julealgon]

hi @julealgon! I don't see webengine.* attributes in this PR, can you clarify the relation to this PR, or even better probably open a new issues with all of the context? thanks

Hi @trask . Those webengine.* attributes are not in the PR at all. I just brought them up because they were standardized in OTEL Semantic Conventions as resource attributes and, to me, they model something very similar to job.system, but job.system was added as a span-level attribute here in the instrumentation instead.

I was arguing/asking whether it would not have made more sense to have job.system in a "Job resource" instead (in my mind it would).

[laurit]

@julealgon the scheduling library, such as quarts, is but a small part of the application, but the resource is shared for all telemetry from the same jvm.