Python prometheus_client，CollectorRegistry 中的重复时间序列

Question

Using the Python Prometheus client, I'm trying add a metric with differing labels to the same collector registry.

My goal is to essentially collect together multiple metrics in one CollectorRegistry() object and push them all at once to push gateway. Ex:

my_metric_name{name="foo",tag="1"} 1.0
my_metric_name{name="foo",tag="2"} 5.0
my_metric_name{name="foo",tag="3"} 7.0

From that above set I want to push them all at once from one registry, rather than creating a new registry for each one. Can I do that?

Background:

Consider this working snippet:

#!/usr/bin/env python3

from prometheus_client import CollectorRegistry, Gauge

registry = CollectorRegistry()

g = Gauge("test", "some gauge",
            labelnames=["name", "tag"], 
            registry=registry).labels("foo", 1)
g.set(1)

Which creates a Gauge metric like this:

test{name="foo",tag="1"} 1.0

However, this does not work:

#!/usr/bin/env python3

from prometheus_client import CollectorRegistry, Gauge

title = "foo"
dataset = [1, 2, 3]

registry = CollectorRegistry()

for item in dataset:
    g = Gauge("test", "some gauge",
             labelnames=["name", "tag"], 
             registry=registry).labels(title, item)
    g.set(1)

I was expecting it to create something like this, with the labels making unique items (not duplicates):

test{name="foo",tag="1"} 1.0
test{name="foo",tag="2"} 1.0
test{name="foo",tag="3"} 1.0

Instead it fails on the second loop with this error:

ValueError: Duplicated timeseries in CollectorRegistry

Wouldn't the different label names make them not duplicates?

I can work around this by pushing the metrics to push gateway more often (on every loop iteration) and creating a new registry. But it would really be helpful to package up a bunch of metrics and send them all at once instead of making many calls over the network.

Answer 1

I came to the realization that you only need to declare the Gauge object once, then each time you call set() on it you'll get a new metric added to the registry:

#!/usr/bin/env python3

from prometheus_client import CollectorRegistry, Gauge

title = "foo"
dataset = [1, 2, 3]

registry = CollectorRegistry()

g = Gauge("test", "some gauge",
         labelnames=["name", "tag"], 
         registry=registry)

for item in dataset:
    g.labels(title, item).set(1)

Results in:

test{name="foo",tag="1"} 1.0
test{name="foo",tag="2"} 1.0
test{name="foo",tag="3"} 1.0