Allow DISTINCT function to operate on tags

[TechniclabErdmann]

I would like to have following feature:

Since the new release 0.9.3 tags are resulting as own columns if you use SELECT * FROM measurement

Currently, it's not possible to use commands on this columns. An example:

SELECT * FROM measurements

returns:

time tagA tagB value
xxx M N 0.3
xxx M O 0.4
xxx P R 0.2

I want to do a query like:

SELECT count(distinct(tagA)) FROM measurements

The result is

2 (M+P)

Anyone else need this feature?

[TechniclabErdmann]

Idea also exist in #1815 at "Not currently implemented (might in the future, but no promises)"

[beckettsean]

@The-Nik are you asking for DISTINCT to support tags, or are you asking for the same functionality that SELECT * used to do? To get the similar SELECT * behavior, just include a GROUP BY *

[TechniclabErdmann]

I ask for DISTINCT to support tags for counting the different tag values 👍

[beckettsean]

@The-Nik you can use SHOW TAG VALUES plus some shell to get what you want:

$ influx -execute 'show tag values with key = a' -database mydb will print a list of all tag values associated with the key a on the database mydb. The output has two header lines, so if you pass it to wc -l just subtract 2 for the actual count:

$ influx -execute 'show tag values with key = a' -database mydb | wc -l and then subtract 2 from the output.

[TechniclabErdmann]

Yeah, this is a good way. But in my case, I need the number in Grafana in a single stat panel. In Grafana, there are some aggregate fuctions but no "count". So InfluxDB has to serve the exact value or I build something in my Grafana like a count-function ;-)

[yee379]

+1; being able to quickly summarise the distinct number of datasets / tags directly from the influx SQL would be very handy; e.g. a grafana panel of the number of sensors I have reporting data over time.

[edennis-sge]

+1: I have a similar use case to yee379 in mind.

[jakefoster]

+1 on being able to count distinct tags.

I also feel like this speaks to the deeper issue of providing guidance on what should be a tag vs. a value. For a schema-less DB there's sure a lot of subtlety around defining your schema! :)

[morganda]

I would also like to see this. We use the cpu and load plugins which themselves don't explicitly provide the cpu count. They do provide each cpu as an "instance" tag e.g. a box with 32 cpus will collect metrics on 32 individual cpus tagging them with with their instance number. If I could get the total count of cpus from the tags, then the load numbers would have a little more context in our graphs Grafana and Chronograph.