Pushing to PushGateway is not pushing any metrics

[ricardoatsouza]

Hello,

I have a set of metrics in my batch job, all of them being created in the following way:

    r = CollectorRegistry()
    ...
    c = Counter(<name>, <description>, registry=r, labelnames=<global_label_names>, labelvalues=<global_label_values>)

In my use case, I have a small set of global labels that I define to every metric when they are created. Then, at the end of the batch job, the metrics are pushed to PushGateway through the push_to_gateway function. The command runs successfully. However, in PushGateway, none of the metrics is actually there.

    push_to_gateway(<url>, job=<job_name>, registry=r)

While trying to understand why, I crossed with this line in the code. Basically, the created metrics are only added to the given registry r when there are no labels given to it. Consequently, the registry r is empty when push_to_gateway runs.

I wonder if this is the intended behaviour. In case positive, is there a way to use the global prometheus registry in push_to_gateway (registry is a required parameter in the function call)?

[roidelapluie]

labelvalues=<global_label_values> is an internal parameter (it is not mentioned in the documentation). Please look at the documentation for proper usage of counters. In particular, I think you need to call Counter.labels() to set the label values.

It makes more sense to ask questions like this on the prometheus-users mailing list or our forums rather than in a GitHub issue. In our community channels, more people are available to potentially respond to your question, and the whole community can benefit from the answers provided.

[roidelapluie]

It also looks like grouping_key is not documented. Perhaps that is what you want?

[csmarchbanks]

Hello, this is a duplicate of #398, I have also created #660 to close that issue and make it clear that labelvalues is for internal use only.

[ricardoatsouza]

@roidelapluie Thanks for the quick answer. Sorry if I couldn't reply earlier. It would be indeed beneficial if the documentation explicitly say that this is intended for internal use.

What I would like to achieve is to define a set of fixed values for my counter (or histogram, etc). Those values do change on every execution of the batch process and are not dimensionally big. I saw in the documentation that I can create the labels using label_names, but then I would have to assign their values using labels every time I would call inc.

That's a small use case, and I am not sure whether it happens often enough to justify its implementation. 🤷‍♂️

[csmarchbanks]

Hello, that is certainly a reasonable use case, and I wonder if this library covers it well enough as is. If not, we can create a new issue that is an enhancement request. Currently the way to create a metric with a fixed set of labels would be like this:

c = Counter(
    "my_requests_total",
    "HTTP Failures",
    ["method", "endpoint"],
).labels("get", "/")
c.inc()

Does a workflow like that enable what you require?

[ricardoatsouza]

Hello @csmarchbanks. Naive if me not to think of your proposal. It seems to work just fine. If I encounter issues, I might come back, but for now the case is closed. Thanks!

[roidelapluie]

Awesome 🥳