Wire up DISTINCT aggregate

[pauldix]

Query looks like this:

select distinct id from user_events
where time > now() - 24h
group by time(1h)

You'd get back the distinct user ids for each hour bucket of time for the last 24 hours.

With distinct you can either pass in a field or a tag. The query engine will have to do wildly different things to get the result depending on what they do.

If it's a tag and no time limitation was specified, then the metastore can answer the query directly. If a time limit was specified, you'll have to run a query.

If it's a field, then you'll have to run a query and run through the whole engine.

[corylanou]

For this current iteration (and based on internal discussions), I'm proposing the following:

Supported

#SELECT DISTINCT <field key> FROM <measurement>
SELECT DISTINCT value FROM cpu

# SELECT DISTINCT <tag key> FROM <measurement>
SELECT DISTINCT host FROM cpu

#SELECT DISTINCT <tag key> FROM <measurement> WHERE time > now() - 24h
SELECT DISTINCT host FROM cpu WHERE time > now() - 24h

# SELECT DISTINCT <field key> FROM <measurement> GROUP BY time(1h) where time > now() - 1d
SELECT DISTINCT value FROM cpu GROUP BY time(1h) where time > now() - 1d

Not currently implemented (might in the future, but no promises)

#SELECT DISTINCT <field key>, <field key> FROM <measurement>
SELECT DISTINCT id, value FROM cpu

#SELECT DISTINCT <tag key>, <tag key> FROM <measurement>
SELECT DISTINCT host, region FROM cpu

# No aggregate functions with a select distinct (sum, count, mean, max, etc.)

#SELECT DISTINCT <field key>, count(<field key>) FROM <measurement>
SELECT DISTINCT id, sum(bytes) FROM network

#SELECT DISTINCT <tag key>, count(<field key>) FROM <measurement>
SELECT DISTINCT host, sum(bytes) FROM network

#SELECT DISTINCT <tag key>, <tag key> FROM <measurement>
SELECT DISTINCT id, value FROM cpu

SELECT DISTINCT <tag values> FROM series WHERE KEY=<tag key>

[andyxning]

@pauldix any plan to support distinct on tag key, see #3880 for more info. We do really need that functionality.

# SELECT DISTINCT <tag key> FROM <measurement> GROUP BY time(1h) where time > now() - 1d
SELECT DISTINCT tag_key FROM cpu GROUP BY time(1h) where time > now() - 1d