We should collect all the labels ever used on a labeled metric about glean-dictionary HOT 4 OPEN

Is it safe to assume that once a label has occupied a certain index in the array, no other label can occupy that same index?

Hypothetical example:

Version 70 has: labels = ["foo", "bar"]
Version 71 adds a new label: labels = ["foo", "bar", "baz"]
Version 72 removes a label, would that be: 1) labels = ["bar", "baz"] or 2) labels = [null, "bar", "baz"]?

On the currently in-development metrics dashboard, the aggregated categorical histograms have no labels and look something like:

    {
        "0": 0.123,
        "1": 0.234,
        "2", 0.345
    }

I'm using the key of the histogram data and matching it up with the index of the labels to assign the proper labels to the histograms.

If option (1) is the case, this will need to be smarter and know that the aggregated histogram was for a specific version and assign the labels that existed for that version.

If option (2) is the case, we could strip the k/v pairs that don't have a non-null value in the labels array.

Either way I think I've convinced myself that the naive approach I took on first attempt is not sufficient and will need to be aware of changing labels between versions.

from glean-dictionary.

This would indeed be a problem for the current desktop telemetry system. However, for Glean, labeled metrics don't really have a concept of an index. They are sent and stored in tables as full strings.

from glean-dictionary.

Guessing this would be an example of a labeled metric:

http://localhost:5000/#!/apps/firefox-reality/metrics/history_sync.failure_reason

from glean-dictionary.

Part of solving this would be figuring out how to make glean dictionary render historical data, which is something we haven't really scoped out yet.

from glean-dictionary.