Option to add new metrics or new lable

[BuSHari]

Hello,

I there an option to add a new metrics or new lable to the existing metrics?
for example, i want to add a lable with the requester ip to the default metric

or add a new metric that contain the location data of the requester ip.

I tried to use the group by, but it overwrite the path lable, and i cant add more than one lable.

Thanks,
Idan

[rycus86]

Hi Idan,

You can add new metrics easily, but you can't add labels to the default metrics.
You could do something like this:

@metrics.summary('requests_by_status', 'Request latencies by status',
                 labels={'requester_id': lambda: get_requester_ip()})
def some_endpoint():
    ...

In this case, the endpoint will be tracked with requests_by_status and also by all of the default ones, like flask_http_request_duration_seconds and flask_http_request_total.

Adding more labels to the built-in metrics is not currently supported.

Does this help?

[BuSHari]

Thanks for the response.
I tried that, but i need it to track all my endpoints that doesn't have the @metrics.do_not_track()
When i tried to create it and duplicate it to some endpoints i got the error

ValueError: Duplicated timeseries in CollectorRegistry:...

[rycus86]

Yeah, I think it might be nice for this library to support this use case, where you either want to extend the default metrics, or apply a metric collector to each of the endpoints.
I'll have a look when I have time, I don't think it should be too difficult to add this.

[BuSHari]

exactly, it would be great...

[rycus86]

I've added support for this now. You could either add the same metric to all endpoints:

app = Flask(__name__)
metrics = PrometheusMetrics(app)

@app.route('/simple')
def simple_get():
    pass
    
metrics.register_default(
    metrics.counter(
        'by_path_counter', 'Request count by request paths',
        labels={'path': lambda: request.path}
    )
)

Or just add the same metric to some of the endpoints (not necessarily all of them) :

app = Flask(__name__)
metrics = PrometheusMetrics(app)

by_path_counter = metrics.counter(
    'by_path_counter', 'Request count by request paths',
    labels={'path': lambda: request.path}
)

@app.route('/simple')
@by_path_counter
def simple_get():
    pass
    
@app.route('/plain')
@by_path_counter
def plain():
    pass
    
@app.route('/not/tracked/by/path')
def not_tracked_by_path():
    pass

Let me know if this helps!

[BuSHari]

Thanks for the fix, but i'm getting this error when trying your example:

File "C:\*****\****\******\*******\env\lib\site-packages\prometheus_flask_exporter\__init__.py", line 358, in register_default
    for endpoint, view_func in self.app.view_functions.items():
AttributeError: 'NoneType' object has no attribute 'view_functions'

[rycus86]

@BuSHari could you provide a small example that fails to test with?

[BuSHari]

Is this helps?

user_info = metrics.register_default(
    metrics.counter(
        'flask_http_user_info', 'Get mote user info',
        labels={'path': lambda: request.path}
    )
)


@api.route('/', methods=['GET', 'POST'])
@user_info
class Serial(Resource):
    @api.login_required(oauth_scopes=['serials:read'])
    def get(self):
        pass

[BuSHari]

And this is an example if i try it for all endpoints:

from flask import Flask, request
from prometheus_flask_exporter import PrometheusMetrics
from prometheus_flask_exporter.multiprocess import GunicornPrometheusMetrics

# Check if app run trough gunicorn
is_gunicorn = "gunicorn" in os.environ.get("SERVER_SOFTWARE", "")

if is_gunicorn:
    metrics = GunicornPrometheusMetrics(app=None)
else:
    metrics = PrometheusMetrics(app=None)


def create_app(flask_config_name=None, **kwargs):
    import threading
    threading.stack_size(2 * 1024 * 1024)
    app = Flask(__name__, **kwargs)

    # Activate metrics '/metrics'
    metrics.init_app(app)
    # static information as metric
    metrics.info('app_info', 'Application info', version='1.0.0')

    gunicorn_error_logger = logging.getLogger('gunicorn.error')
    app.logger.handlers.extend(gunicorn_error_logger.handlers)

    env_flask_config_name = os.getenv('FLASK_CONFIG')

    with app.app_context():

        from . import extensions
        extensions.init_app(app)

        from .templates.db_admin import add_db_admin_cmds
        add_db_admin_cmds(app)

        from . import modules
        modules.init_app(app)

        metrics.register_default(
            metrics.counter(
                'by_path_counter', 'Request count by request paths',
                labels={'path': lambda: request.path}
                )
            )

    return app

[rycus86]

Yep, so the register_default does not return anything, that should just register a default metric, that's why it complains about the None I believe.

user_info = metrics.register_default(...

Your second example with Gunicorn should work though.
I'll try to have a look later why it doesn't.

[BuSHari]

ok thank you.
The first example is working for individuals endpoints, please let me know if there is a fix for all endpoints.

Thank you...

[rycus86]

All right, I'll have a look later if I can.
By the looks of it, that should work.
Do you use plain Flask, or some other extension like Restplus or Restful?

[BuSHari]

thank you, I use restplus

[rycus86]

OK, it should be fixed in version 0.11.0.
There was a problem, that the default registration only worked if the Flask app was passed in at the time of creating PrometheusMetrics object, not when it's init_app is used.
Now changed it so that it works under inside a with app.app_context(): context, otherwise the register_default(..., app=None) function now accepts the Flask app to look for endpoints on.

See https://github.com/rycus86/prometheus_flask_exporter/blob/master/examples/restplus-default-metrics/server.py for an example, though it's pretty much what you had above, your example should now work without any changes.

Let me know how this works out for you!

[BuSHari]

Great!
Thank you its working now... I'll test it further and let you know for any issues