Prometheus metrics for Blender

Exposes Prometheus metrics on port 8000 when enabled.

In particular, it exposes a Gauge, Blender_Render, that is 1 when the Blender instance is rendering, and 0 when it is not. The standard metrics, like cpu usage are also exposed.

Building the add-on

Currently we use an embedded and slightly modified version of the prometheus_client for Python, which is in the subdirectory. That is not ideal and might change once I have figured out how to monkey path the server instead of hacking the source, but i don´t like to depend on external Python packages because that makes it difficult to distribute and add-on. Copying is also very far from ideal, so a git sub-repository is probably the way to go.

git clone https://github.com/varkenvarken/blenderaddons.git
cd blenderaddons
zip -urv prometheus.zip prometheus

Then install prometheus.zip in the usual way and enable the add-on.

You can inspect the published metrics on http://localhost:8000.

Scraping this with a Prometheus container and creating a Grafana dashboard is something you'll have to figure out yourself.

Source code

Available on Github.

The code is extremely simple: when the add-on is enabled we create a Gauge metric and start the prometheus http server. We then register a Blender timer that checks every 10 seconds if we are rendering or not with the is_app_running() function and sets the Gauge accordingly.

The timer is made persistent, so it will keep running even if we load another .blend file. The Prometheus server is running in a separate (daemon) thread and will only end if we exit Blender or if we invoke our custom stop_http_server() function, and that's where the ugly hack comes in: the start_http-server() function does not return the server it creates, so w e have no way to call its shutdown() function (which is present, because it is a subclass of http.server) or call close() on the socket it is listening on, and this would prevent us from disabling and then enabling the add-on again, because we would get an address in use exception.

Same goes for the Gauge: If we want o be able to reenable the add-on we have to make sure to remove it from the registry in the unregister() function.

import bpy  # type: ignore
from .prometheus_client import Gauge, start_http_server, stop_http_server, REGISTRY

def every_10_seconds():
    global g
    r = bpy.app.is_job_running("RENDER")
    if r:
        g.set(1.0)
    else:
        g.set(0.0)
    return 10.0

def register():
    global g
    g = Gauge("Blender_Render", "Rendering processes")
    start_http_server(8000)
    bpy.app.timers.register(every_10_seconds, persistent=True)

def unregister():
    global g
    bpy.app.timers.unregister(every_10_seconds)
    REGISTRY.unregister(g)
    stop_http_server()