How to get additional profile types such as allocated memory and heap profiling in Parca UI

[Adib234]

Hello, I came across the tutorial and read the note regarding how Parca Agent's automatic profiling with zero-instrumentation currently only supports CPU profiling. Does this mean I would need a pull-based approach of collecting profiles in pprof format from HTTP endpoints and how would this be done? I checked out the example https://github.com/polarsignals/pprof-example-app-go and then went on Parca UI but I still see only CPU samples being the only available profile type. I've had the example deployed in a cluster and deployed a Prometheus-based collector.

[brancz]

I've had the example deployed in a cluster and deployed a Prometheus-based collector.

Can you share what you tried here? This is the correct approach, the scraper is the Parca server itself.

[Adib234]

I first tried deploying the example using kubectl apply -f and then setting up parca exactly how it's done in the tutorial for setting up Parca with Kubernetes https://www.parca.dev/docs/kubernetes/, then I deployed my Prometheus-based collector however for some reason I only see CPU samples as the only profile type

[brancz]

Can you share what you see on the /targets page?

[Adib234]

I was able to resolve this by fixing parca.yaml

[sojinss4u]

I am also facing the same issue. @Adib234 Can you please share what changes you made in parca.yaml to get the memory profiles?
Thanks

[brancz]

@sojinss4u can you describe what you’ve done and what you expected to see?

[Raviteja-Salendra]

I first deployed parca server and parca agent using kubernetes following the tutorial (https://www.parca.dev/docs/kubernetes),but in parca UI Iam able to see only CPU profiles but not memory profiling.When I installed parca server and parca agent using in another instance Iam able to see both CPU and memory profiling using kubernetes .Iam also facing the same problem @Adib234 has solved it seems.

[brancz]

For memory profiling you need to 1) instrument your application and 2) set up scraping as described here: https://www.parca.dev/docs/ingestion#combining-push-and-pull-based-ingestion

[Raviteja-Salendra]

Is it possible to get memory profiling directly through the Parca agent without instrumenting the application -just like how I can get CPU profiles for all running pods -so that memory profiling also works without adding instrumentation to the pods?