Skip to content

InfoMetricFamily’s add_metric() parameters are strangely named and should provide default values #1049

Open

Description

@calestyo

Hey.

I was looking into InfoMetricFamily’s add_metric() and I either I just don't get it or the API is a bit strange.

With an Info object I'd have done something like this:

m = prometheus_client.Info("logical_drive", "foobar", ("controller_name",
                                                       "array_name",
                                                       "logical_drive_name",
                                                       "caching",
                                                       "device",
                                                       "raid_level",
                                                       "logical_drive_label",
                                                       "multidomain_status",
                                                       "parity_initialization_status",
                                                       "status"
                                                      )
                      )

and then set it like:

m.labels(controller_name=controller_name, array_name=array_name, logical_drive_name=logical_drive_name,
         **{n: logical_drive_properties.get(n, "")  for n in (
                                                              "device",
                                                              "raid_level",
                                                              "logical_drive_label",
                                                              "multidomain_status",
                                                              "parity_initialization_status",
                                                              "status"
                                                             )
           },
         caching=bool_or_none_to_label_value( logical_drive_properties.get("caching") )
        )

(here an example where I set some of the properties directly, and some via dict unpacking)

  • It would fail if I forgot a label.
  • I could use label names as attribute names, like controller_name= rather than "controller_name"=... which is however not super important

With InfoMetricFamily seem quite a bit different:

  • add_metric() now has the parameters labels and value.
  • As far as I understand the code, labels are actually not labels, but values for these, namely the ones set with labels in the constructor of InfoMetricFamily.
  • It's no longer possible to give labels as dict, one really needs to give them in the right order as sequence. Why does Info allow that but not this?
  • AFAICS, there is no check if exactly those labels are set, that were given in the constructor. Why over at Info but not here? What sense does it then even make to set the labels in the constructor?
  • value is misleading in so many ways. It's not the value of the metric (that is 1) and even if it relates to the labels being the “value”, then it should be values.
  • It might have perhaps even been better to swap the two names... or rater use some completely different names.

Anyway... guess this can't be changed now without breaking the ABI.

However, as far as I understand the code:

dict(dict(zip(self._labelnames, labels)), **value),

the idea is one can give labels and/or value and both are merged in a final dict of label/value pairs.... but then it would be nice if labels and value were not required.

Why not simply give them default values () respectively {}?

I could provide a patch if nothing speaks against that particular change. But the above points are IMO still problematic. Especially also that the behaviour is considerable different from Info, which is not really obvious.

Cheers,
Chris.

Activity

  1. calestyo commented on Jul 11, 2024

    @calestyo
    Author

    It seems even worse... the other objects like GaugeMetricFamily do not allow to set their labels as dict... so one always must present them in the right order, which is error prone.

    Why don't they just offer the same means as Gauge.labels() :-(

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions