custom collector with client_python successive collections

[Cory Robinson]

Hi,

When implementing a simple custom collector with the python client, refreshing the browser will generate 2 successive collections.

This tends to cause "duplicate" unneeded successive, say, REST API calls for example,  when collecting metrics from REST API endpoints real-world scenarios.

How can this be avoided?

Here is some sample case below that exhibits this behavior...

import time

from prometheus_client import start_http_server

from prometheus_client.core import GaugeMetricFamily, CounterMetricFamily, REGISTRY

class CustomCollector(object):

    def collect(self):

        c = GaugeMetricFamily('my_gauge', 'Help text', labels=['foo'])

        c.add_metric(['bar'], 1.7)

        print("{}, {}".format(time.time(), c))

        yield c

    

    def describe(self):

        return []

if __name__ == '__main__':

    start_http_server(8000)

    REGISTRY.register(CustomCollector())

    while True:

        time.sleep(1)

Console output at webpage refresh on localhost:8000:

> python test.py
1592667742.5491908, Metric(my_gauge, Help text, gauge, , [Sample(name='my_gauge', labels={'foo': 'bar'}, value=1.7, timestamp=None, exemplar=None)])
1592667742.6191902, Metric(my_gauge, Help text, gauge, , [Sample(name='my_gauge', labels={'foo': 'bar'}, value=1.7, timestamp=None, exemplar=None)])

Notice the times come in very quick succession. When collecting real metrics from a REST API endpoint, the metrics displayed on the webpage will always be the first metric, not the 2nd succession.

Thanks,

Cory

[Cory Robinson]

Hi folks, any ideas on this issue I am facing with custom collectors using the Python client?

Thanks,

Cory