Grafana table from Prometheus metric with key-value labels

Question

I have a metadata metric in Prometheus that looks like this

# HELP metadata Process metadata
# TYPE metadata counter
metadata{hostname="server-a",key="version",value="v1.1.0"} 1
metadata{hostname="server-a",key="feature1",value="true"} 1
metadata{hostname="server-a",key="feature2",value="false"} 1
metadata{hostname="server-b",key="version",value="v1.0.0"} 1
metadata{hostname="server-b",key="feature1",value="false"} 1

where the key label has the name of the field and the value label has a string value. We don't care about the value of the metric, but it'll always be 1.

Is it possible to have Grafana format this as a table? i.e. like this

Hostname	version	feature1	feature2
server-a	v1.1.0	true	false
server-b	v1.0.0	false

If I have a separate query (with Table format) for each possible key then I'm able to make it look like that with two transformations: an Outer join by hostname, then a Organize fields to remove the excess fields.

However, I'd like to do that without a separate query for each key, or even hardcoding key if possible. But despite all my fiddling with transforms I can't figure out how. There's also a couple of questions here that ask about similar things, but none of them do exactly what I'm looking for:

• Table-like visualization with label values as row values and column names which uses the actual metric value for the column values
• Grafana - create table with column values with Prometheus (dynamic) property/label data has each instance as a single metric with the right labels, instead of spread out over a number of metrics

And a few things that don't work

• A single query for metadata{} in table format has a column for the key and a column for the value, rather than column with the name of the key
• Closest I've gotten is with a Time series query then a Label to fields transformation, which has a column of 1s whose name is the value of key.

So is what I want even possible in Grafana?

Olympian

Answer 1

With the new Grouping to Matrix transform introduced by this PR in Grafana v8.5.0, this is now possible!

• Set the Prometheus query to have Format: Table and Type: Instant.

• Add a Grouping to Matrix Transform, with:

  • Column: key ("I want a different column for each key")
  • Row: hostname ("I want a different row for each hostname")
  • Value: value ("The cell should take the value from value")
  • Empty Value: both Null or Empty will produce the correct output

• Add a Rename by regex Transform.

  • Match: (.*)\\.*
  • Replace: $1

  The Grouping to Matrix transform produces a column with name like <row>\<column> (e.g. hostname/\key), and this fixes that to be just the row name (e.g. hostname)

• Finally, set the Visualisation (top right) to Table.

Olympian

Answer 2

Need to select in Options -> Format -> Table

Pigeontoed

This only gives a list of key-value pairs, not a table in the format the question asks. – Olympian

Answer 3

I've written down a tiny note regarding this problem. From my side, I can suggest using basic transformations:

1. Organize fields: remove the timestamp column;
2. Rename by regex: convert strings on labels to a JSON object like ["server-a", "v1.1.0"];
3. Reduce: move column name to value; Calculation can be Max;
4. Organize field: delete the Max column;
5. Extract fields: convert the JSON object to a list of columns;
6. Organize fields: hide and rename required columns.

All figures are presented in the story. That's not so quick approach, but it works.

Pokorny