Add User-Defined metrics to UDF

[KirkBuah]

I’m running a pipeline with UDF vertices. Each UDF vertex contains a Python project where we define custom metrics using the prometheus_client library.

Normally, to expose these metrics, we’d run:

from prometheus_client import start_http_server

start_http_server(server_port)

However, I’ve run into a problem: The default port 2469 is already used by Numaflow metrics.

I can’t bind to another port because, in the UDF container, only 2469 and 2470 are open (and already in use).

My question: What’s the right way to expose custom Prometheus metrics from a UDF vertex in this setup?
Should I reuse the existing metrics endpoint somehow, or is there a recommended way to make additional ports available for UDF-defined metrics?

[vigith]

You can provide custom port in ports section under the container in the udf section.

udf:
  container:
    image: docker.io/some-image:latest
    env:
      - name: foo
        value: bar
    ports:
      - name: metrics
        containerPort: 8493

[KirkBuah]

Thank you, that's what I ended up doing and it worked.