Filter by an ad-hoc aggregated field

iex(51)> FiveWhys.Understand.Session |> Ash.Query.aggregate(:count_of_interactions, :count, :interactions) |> Ash.Query.filter(:count_of_interactions > 4) |> Ash.read!(authorize?: false) |> length

:count_of_interactions > 4 is comparing an atom to an integer, and the literal comparison is actually expanded before querying to just filter: true.

You can’t reference those dynamic aggregates by name (back in the day you could, but no longer, it had ambiguity/compositional errors).

Here is an example trying to aggregate on a different field. This query returns only 9 records with more than 4 interactions, which agrees with the sql:

iex(50)> FiveWhys.Understand.Session |> Ash.Query.filter(count(interactions) > 4) |>  Ash.read!(authorize?: false) |> length
[debug] QUERY OK source="sessions" db=37.6ms idle=511.7ms
SELECT s0."id", s0."title", s0."shared", s0."updated_at", s0."inserted_at", s0."user_id", s0."problem_statement", s0."prior_session_id" FROM "sessions" AS s0 LEFT OUTER JOIN LATERAL (SELECT si0."session_id" AS "session_id", coalesce(count(*), $1::bigint)::bigint AS "aggregate_0" FROM "public"."interactions" AS si0 WHERE (s0."id" = si0."session_id") GROUP BY si0."session_id") AS s1 ON TRUE LEFT OUTER JOIN "public"."summaries" AS s2 ON s0."id" = s2."session_id" LEFT OUTER JOIN LATERAL (SELECT si0."session_id" AS "session_id", any_value(si0."inserted_at"::timestamp ORDER BY  si0."inserted_at" DESC ) FILTER (WHERE si0."inserted_at"::timestamp IS NOT NULL)::timestamp AS "latest_interaction_date" FROM "public"."interactions" AS si0 WHERE (s0."id" = si0."session_id") GROUP BY si0."session_id") AS s3 ON TRUE WHERE (coalesce(s1."aggregate_0", $2::bigint)::bigint::bigint > $3::bigint::bigint) ORDER BY (CASE WHEN NOT (s2."id"::uuid IS NULL) AND (s2."inserted_at"::timestamp::timestamp > s3."latest_interaction_date"::timestamp::timestamp) THEN s2."inserted_at"::timestamp ELSE s3."latest_interaction_date"::timestamp END)::timestamp DESC [0, 0, 4]
↳ anonymous fn/3 in AshPostgres.DataLayer.run_query/2, at: lib/data_layer.ex:785
9

select count(*) from (select s.id, count(*) from sessions s inner join interactions i on s.id = i.session_id group by s.id having count(*) > 4
);
 count
-------
     9
(1 row)

This query returns 15 (all of the sessions):

iex(51)> FiveWhys.Understand.Session |> Ash.Query.aggregate(:count_of_interactions, :count, :interactions) |> Ash.Query.filter(:count_of_interactions > 4) |>  Ash.read!(authorize?: false) |> length
[debug] QUERY OK source="sessions" db=19.7ms queue=1.0ms idle=868.8ms
SELECT s0."id", s0."title", s0."shared", s0."updated_at", s0."inserted_at", s0."user_id", s0."problem_statement", s0."prior_session_id", coalesce(s1."count_of_interactions"::bigint, $1::bigint)::bigint FROM "sessions" AS s0 LEFT OUTER JOIN LATERAL (SELECT si0."session_id" AS "session_id", coalesce(count(*), $2::bigint)::bigint AS "count_of_interactions" FROM "public"."interactions" AS si0 WHERE (s0."id" = si0."session_id") GROUP BY si0."session_id") AS s1 ON TRUE LEFT OUTER JOIN "public"."summaries" AS s2 ON s0."id" = s2."session_id" LEFT OUTER JOIN LATERAL (SELECT si0."session_id" AS "session_id", any_value(si0."inserted_at"::timestamp ORDER BY  si0."inserted_at" DESC ) FILTER (WHERE si0."inserted_at"::timestamp IS NOT NULL)::timestamp AS "latest_interaction_date" FROM "public"."interactions" AS si0 WHERE (s0."id" = si0."session_id") GROUP BY si0."session_id") AS s3 ON TRUE ORDER BY (CASE WHEN NOT (s2."id"::uuid IS NULL) AND (s2."inserted_at"::timestamp::timestamp > s3."latest_interaction_date"::timestamp::timestamp) THEN s2."inserted_at"::timestamp ELSE s3."latest_interaction_date"::timestamp END)::timestamp DESC [0, 0]
↳ anonymous fn/3 in AshPostgres.DataLayer.run_query/2, at: lib/data_layer.ex:785
15

select count(*) from sessions;
 count
-------
    15
(1 row)

My first example (I assume you are talking about the very first post in the thread) simply doesn’t filter by count_of_summaries - it returns all entries, even those where count of summaries is 0 or 1. (There are no errors or warnings, though). Is it considered a bug? If yes, I can file a ticket in ash_postgres.

Sorry for being the source of all these edge cases :).

The following does work:

Session
|> Ash.Query.filter(count(summary) > 1)
|> Ash.read!()

That is actually an out of date example, sorry. You would pass the result of Ash.Query.Aggregate.new into the sort, i.e Ash.Query.sort({aggregate, :desc}). However, your filter example should work AFAIK. In what way is it not working?

Ok (thanks to ChatGPT o3 w/ search for the idea), is seems that

Session
|> Ash.Query.filter(count(summary) > 1)
|> Ash.read!()

should work as expected! However, I’m still a bit puzzled why the above (admittedly, naïve) solution doesn’t work, as e.g. the following sorting by an aggregate field is explicitly mentioned in the docs for Ash.Query:

MyApp.Author
|> Ash.Query.aggregate(:published_post_count, :posts, query: [filter: [published: true]])
|> Ash.Query.sort(published_post_count: :desc)
|> Ash.Query.limit(10)
|> Ash.read!()

and one would expect that if a sort by an ad-hoc aggregate field works, a filter should work as well.

There is also a parameter :filterable in Ash.Query.Aggregate.new, which is defined like this:

• :filterable? - Whether or not this aggregate may be used in filters. The default value is true .

and it says that Aggregate.new is used with Query.aggregate or Ash.aggregate.