Make readiness timeout configurable about jhsingle-native-proxy HOT 9 CLOSED

Thanks for the update, but just to clarify do you mean:

Adding --ready-check-path=/healthz seems to work well for your dashboard starting OK. But that you no longer have the problem anyway, even if you drop --ready-check-path again.

?

Yes, that's it.

from jhsingle-native-proxy.

The bug is still happening even with --ready-check-path=/healthz

from jhsingle-native-proxy.

This idea probably makes sense, and I would like to add it as an option at some point.

However, it may be more efficient for you to use the --ready-check-path=/ready-check argument instead. This allows you to specify an alternative URL (e.g. /ready-check) that will tell jhsingle-native-proxy if the web server is ready. It is inefficient to request the main Streamlit app to do this really - maybe you can choose the URL to a smaller static asset from Streamlit that at least would indicate the Streamlit web server is ready to receive requests. This should be able to respond much quicker, and ultimately means that JupyterHub can redirect users to the dashboard sooner, so their request for the main Streamlit app can begin sooner too.

Please let me know if that helps.

from jhsingle-native-proxy.

I'll try to set --ready-check-path=/healthz for streamlit and what this what it gives

from jhsingle-native-proxy.

Unfortunately, I couldn't reproduce the bug again, even though --ready-check-path=/healthz is working for streamlit dashboards, we can't be sure it fixes the bug until reproducing it.

I'll get back whenever the bug is reproduced.

from jhsingle-native-proxy.

Thanks for the update, but just to clarify do you mean:

Adding --ready-check-path=/healthz seems to work well for your dashboard starting OK. But that you no longer have the problem anyway, even if you drop --ready-check-path again.

?

from jhsingle-native-proxy.

I reproduced the bug with the following flask app:

from flask import Flask
from time import sleep

app = Flask(__name__)

@app.route('/')
def hello_world():
    sleep(30)
    return 'Hello, World!'

and theses configurations:

c.CDSDashboardsConfig.presentation_types = ['flask']

c.VariableMixin.extra_presentation_launchers = {
    'flask': {
        'args': [
            'flask', 'run', '{--}port={port}'
        ],
        'env': {'FLASK_APP': 'app.py'},
    }
}

from jhsingle-native-proxy.

A better reproduction app:

from flask import Flask
from time import sleep, time

app = Flask(__name__)

started_at = time()
ready_after = 30

@app.route('/')
def hello_world():
    now = time()
    if now - started_at > ready_after:
        return 'Appliation ready!'
    else:
        # sleep for more that 1 seconds to make the _ready_check call fail: 
        # https://github.com/yuvipanda/simpervisor/blob/master/simpervisor/process.py#L196
        # Even status code 400 is considered as ready as per https://github.com/ideonate/jhsingle-native-proxy/blob/master/jhsingle_native_proxy/proxyhandlers.py#L707
        # return 'Application not yet ready! Please try later', 400
        sleep(5)

Why better?

The first app will always fail to start even if we increase the ready_timeout, it fails because the ready_func call is only waited for 1 second in https://github.com/yuvipanda/simpervisor/blob/master/simpervisor/process.py#L196

from jhsingle-native-proxy.

By the way, it would be great if you'd like to take a look at the experimental Flask framework:

https://cdsdashboards.readthedocs.io/en/stable/chapters/userguide/frameworks/flask.html

That should be more flexible for being able to specify a flask script to run, and it does it through gunicorn instead of the dev server. There may be some tweaks needed...

from jhsingle-native-proxy.