Add gauge for max exporter queue capacity

[swar8080]

Is your feature request related to a problem? Please describe.
Hello, i'm looking to monitor when exporter queue size gets close to reaching max capacity. Currently i'm hardcoding the max queue size with a query like queueSize / 2048. We may increase that limit for some applications in the future, so it'd be nice to avoid hardcoding the size limit with a query like queueSize / maxQueueSize

Describe the solution you'd like
Add a gauge that reports that configured maximum export queue size

Describe alternatives you've considered
Keeping the configured maximum in-sync between applications and alerts

Additional context
I'm monitoring this on behalf of all application teams so it'd be nice to have a single alert configuration that works for all applications

[jkwatson]

Since this is a constant for the lifetime of a VM, could you simply create the gauge yourself at the moment when you're configuring the SDK?

[swar8080]

We could, but it would be nice if the teams I support don't have to make code changes. Implementing this within the SDK means we can use the OTEL operator + java agent to automatically roll this out to all our applications. I'm assuming we'd just have to update the version of the java agent that the operator injects

[jkwatson]

🤔 I don't know the state of whether the metrics that we're emitting currently are official, or more experimental. Does the specification require a set of metrics to be exported by the batch span processor? I wrote the original metrics several years ago, but I know there was a desire to have the metrics be "official" via specification, and I've lost track of whether that has been done yet.

[trask]

hi @swar8080, there's some ongoing work, see open-telemetry/semantic-conventions#598

assuming that proposal moves forward (it's stalled currently but expect it to be picked back up at some point), I think you could instead alert on the metric otelcol.processor.items where otel.outcome=queue_full if my understanding of your requirement is correct?

[swar8080]

Thanks @trask i'll keep an eye on the sdk metric convention

Ideally we could alert when the export queue is close to filling up so that we can hopefully increase the resources/limit before drops happen

Maybe we could also do this with an agent extension now that the OTEL operator allows injecting them without any code changes for service owners

[jack-berg]

This is solved in #7931, #7895 which have added:

• otel.sdk.processor.[span|log].queue.size
• otel.sdk.processor.[span|log].queue.capacity