Yes — Prefect self-hosted 3.6.12 supports metrics export, but not Kubernetes-native metrics. What it supports - A Prometheus/OpenMetrics endpoint on the Prefect server API - A separate client/worker metrics endpoint for workers and other Prefect client processes What it does not support - No built-in Kubernetes-specific metrics like pod status, node usage, scheduler state, or kube API object metrics - For those, you’d still use standard Kubernetes observability tools like

kube-state-metrics

, node exporter, cAdvisor, etc. Server/API metrics - Enable with

PREFECT_API_ENABLE_METRICS=true

- Alias also exists:

PREFECT_SERVER_METRICS_ENABLED=true

- Endpoint:

/api/metrics

- Served in Prometheus/OpenMetrics text format So if your self-hosted API is at:

<http://prefect-server:4200>

You’d scrape:

<http://prefect-server:4200/api/metrics>

What the server exposes Prefect 3.6.12’s API metrics endpoint mainly exposes the default metrics provided by

prometheus_client

, including: -

process_*

metrics - CPU time - resident/virtual memory - file descriptors, etc. -

python_*

metrics - GC activity - interpreter/runtime stats - threads, etc. I did not find evidence of a rich set of Prefect-server-specific business metrics on that endpoint in 3.6.12. Worker/client metrics Workers can also expose metrics from a separate HTTP server. Enable with: -

PREFECT_CLIENT_METRICS_ENABLED=true

- Optional port override:

PREFECT_CLIENT_METRICS_PORT=4201

Default port: -

4201

Worker/client metrics exposed These Prefect-specific metrics are present: -

prefect_logs_observed

- Labels:

client

-

prefect_log_websocket_connections

- Labels:

client

,

direction

,

connection

-

prefect_events_emitted

- Labels:

client

-

prefect_events_observed

- Labels:

client

-

prefect_event_websocket_connections

- Labels:

client

,

direction

,

connection

-

prefect_event_websocket_checkpoints

- Labels:

client

Where label values include things like: - `direction`:

in

,

out

- `connection`:

initial

,

reconnect

In Kubernetes This means: - You can scrape Prefect server and worker pods with Prometheus - But the metrics are application/process metrics, not Kubernetes control-plane or pod lifecycle metrics Typical setup: - Scrape Prefect API at

/api/metrics

- Scrape worker pods on port

4201

- Use a

ServiceMonitor

or PodMonitor if you’re using Prometheus Operator Example env vars for the API pod:

- name: PREFECT_API_ENABLE_METRICS
  value: "true"

Example env vars for a worker pod:

- name: PREFECT_CLIENT_METRICS_ENABLED
  value: "true"
- name: PREFECT_CLIENT_METRICS_PORT
  value: "4201"

Bottom line - Yes: self-hosted Prefect 3.6.12 supports Prometheus metrics in Kubernetes - No: it does not expose Kubernetes-specific metrics out of the box - Exposed metrics: mostly

process_*

,

python_*

, plus a small set of Prefect client/worker event/log websocket counters If you want, I can also give you a ready-to-apply: - Kubernetes

Service

+

ServiceMonitor

for the API -

Service

+

ServiceMonitor

for workers - sample Prometheus scrape config