Observability in AEMaaCS with OpenTelemetry

When operating in AEMaaCS, you cannot add a Java agent to instrument your application. To address this limitation, you can install an OSGi bundle that uses the OpenTelemetry SDK to provide telemetry for servlets and components. This bundle also allows developers to add custom telemetry, enhancing the observability of your application.

Krish

Is there any performance impact on the application?

(see answer in talk video)

Tomasz Sobczyk

Not a question - rather statement - there is an improved version of new relic ( supporting sling) that was prepared by new relic folks in the past. I hope we will see it added one day to aem cloud - as it would improve the data we see in apm. This will obviously not fix lack of logs issue (new relic ootb in theory can support it as well) Still it will be detached from rest of architecture that you can have

Michal Cukierman

Do you send all spans/traces or do you use samplers?

(see answer in talk video)

Michal Cukierman

Do you instrument Apache/Dispatcher and the CDN to preserve complete trace?

Tomasz Sobczyk

You cannot really instrument it. I presume you can just stream the logs with ootb exporters. Actually would be great to know how well that went. Are there any unique tracing request ids issued by cdn that flow down stream?

Rogier

You can on on-prem of course ;)

Tad

This is amazing work! What sort of delay are you typically seeing for log data for AEMaaCS with this framework?

(see answer in talk video)

Michal Cukierman

How to you link spans with exceptional log entries? Do you attach exception to the current span in case of errors/failures?

(see answer in talk video)
Does it also pick up any building errors from cloud manager? So when your build is failing instead of having to download the error files from CM and go through it yourself it immedialty shows the reason it failed?

Barry

telemetry data is only exposed for running instances so the cloudmanager logs are not there. You can always foresee something that is triggered on the adobe i/o events for the cloudmanager builds en push them towards your logging solution

Why complain about AEM as a walled garden if it's not even part of Sling?

Barry

What is exactly your question? The point that was being made is that as an implementor you are not the driver of what comes in the APM provided by Adobe and we want to be in control to determine what should be instrumented and what not and more important get it inside the available tooling in order to have distributed traces. It's not about "complaining" that it's closed, that's perfectly fine.

What conflicts can it have with new relic apm?

Barry

As far as I know there should be no conflicts with any other tooling or frameworks