Support distinct projection queries?

[jgeewax]

Related to #304 (#304 (comment))

and

http://stackoverflow.com/questions/33283759/how-to-do-a-gql-query-using-gcloud-python

How do we do the equivalent of GQL: SELECT DISTINCT p FROM Kind ?

/cc @Fkawala

[jgeewax]

@pcostell : Can the API even allow this type of thing using a structured query object ? Or is this only supported via GQL ?

[tseaver]

Datastore queries take a projection parameter, which is a list of the fields to be returned. AFAICT, the API don't expose any way to do any "summary" operations (DISTINCT, COUNT, SUM, etc.) except to pull the values back and munge them client-side.

[pcostell]

Sorry OOO so short reply:
Yes it can. In v1beta2 you can use group_by, in v1beta3 you can use distinct_on.

SELECT DISTINCT p FROM Kind == SELECT p FROM KIND GROUP BY p.

Aggregation ops aren't supported because they would require full scans of the data to answer. DISTINCT however can be implemented efficiently.

[Fkawala]

@pcostell so the parameter group_by of object Query does what I expect, and there is not any interest (speed / efficiency ?) to use GQL SELECT DISTINCT ? Thanks.

[pcostell]

@Fkawala -- yes, using group_by on all your projection properties is the exact same as using SELECT DISTINCT IN GQL. (In fact that is how GQL DISTINCT is implemented.

[dhermes]

@pcostell I'm closing this out. It seems support already exists. LMK if I'm mistaken.