uwsgi config "lazy-apps: yes" "should_start_http_server" return False

[xiecang]

"UWsgiPrometheusMetrics().should_start_http_server()" returns False when I set uwsgi's startup parameter "lazy-apps:yes".

How can I use it when setting "lazy-apps:yes" ?
uwsgi config:

uwsgi:
  wsgi: wsgi:app
  http-socket: 0.0.0.0:11010
  processes: 4
  threads: 16
  master: yes
  ignore-write-errors: yes
  ignore-sigpipe: yes
  die-on-term: yes
  wsgi-disable-file-wrapper: yes
  max-requests: 65535
  max-requests-delta: 1024
  log-prefix: uWSGI
  log-date: yes
  log-slow: 10000
  disable-logging: yes
  need-app: true
  reload-mercy: 1
#  lazy-apps: yes

code:

    from prometheus_flask_exporter.multiprocess import UWsgiPrometheusMetrics
    metrics = UWsgiPrometheusMetrics(app)
    metrics.start_http_server(9200, )

[rycus86]

Hm, I'm not familiar with the setting, let me have a quick look.

[rycus86]

One workaround that seems to work is exposing the metrics endpoint on the main application:

app = Flask(__name__)
metrics = UWsgiPrometheusMetrics(app)
metrics.register_endpoint('/metrics')

I'm still testing if the endpoint can be started on a new port with lazy-apps.

[rycus86]

See an example in https://github.com/rycus86/prometheus_flask_exporter/blob/master/examples/uwsgi-lazy-apps/server.py and the setup in https://github.com/rycus86/prometheus_flask_exporter/blob/master/examples/uwsgi-lazy-apps/Dockerfile#L14-L16

[rycus86]

I haven't found a way to start the endpoint on a separate HTTP server - the master process doesn't seem to run the module. If you know a way, please let me know!
Otherwise if you're happy with the workaround, please close the issue.

Thanks!

[xiecang]

It is very effective！
Thanks!

[tlinhart]

I'm a bit late here but anyway. Exposing the metrics endpoint on the main app is fine with lazy-apps. However, when for some reason you need to server the metrics on a different port (e.g. you don't want to expose them to the world), you can't do that. That's exactly my case. In my setup, the app is deployed as a Docker image (nginx -> uWSGI -> Flask app) and I need to run the metrics on a different port so that it's not exposed to the internet but only to Prometheus running on internal network. The solution I took is to use multiprocess.MultiProcessCollector from official Python client in my original Flask app. The other app that exposes the metrics uses the same setup for the prometheus_multiproc_dir environmental variable and hence accesses the metrics from my app.

[rycus86]

That sounds like a good setup. How's the second app running to expose the metrics endpoint? Is it running in the same container?

[tlinhart]

Yeah. I use Supervisord to run nginx and uWSGI. So basically I added another program to the game -- Flask app exposing the metrics. Another advantage is that this way, you can scale your app and the metrics app independently.

[rycus86]

Thanks for sharing!