Optimize todos counters in usage data

What does this MR do?

Optimizes the todo counter query for batch counting in usage data

Issue #208891 (closed)

Query

explain SELECT COUNT(DISTINCT "todos"."author_id") FROM "todos" WHERE "todos"."created_at" BETWEEN '2020-02-07 06:22:09.650617' AND '2020-03-06 06:22:09.650886' AND "todos"."author_id"  BETWEEN 610000 AND 612000

Optimization

CREATE INDEX index_todos_on_author_id_and_created_at ON public.todos USING btree (author_id, created_at);
# The query has been executed. Duration: 6.503 min (edited) 

Before: Bad! https://explain.depesz.com/s/CCpc

After: Good: https://explain.depesz.com/s/QwPV

Timing

After the index for batch counting takes 55 seconds pessimistic

• 5.5 million users,
• with 10K batch sizes
• 5.5M/10K = 550 loops
• Time: < 100 ms ( cold cache )

Migration output

$ VERBOSE=true bundle exec rake db:migrate:up VERSION=20200306170531
== 20200306170531 AddIndexOnAuthorIdAndCreatedAtToTodos: migrating ============
-- transaction_open?()
   -> 0.0000s
-- index_exists?(:todos, [:author_id, :created_at], {:algorithm=>:concurrently})
   -> 0.0037s
-- execute("SET statement_timeout TO 0")
   -> 0.0003s
-- add_index(:todos, [:author_id, :created_at], {:algorithm=>:concurrently})
   -> 0.0196s
-- execute("RESET ALL")
   -> 0.0012s
== 20200306170531 AddIndexOnAuthorIdAndCreatedAtToTodos: migrated (0.0251s) ===

$ VERBOSE=true bundle exec rake db:migrate:down VERSION=20200306170531
== 20200306170531 AddIndexOnAuthorIdAndCreatedAtToTodos: reverting ============
-- transaction_open?()
   -> 0.0000s
-- index_exists?(:todos, [:author_id, :created_at], {:algorithm=>:concurrently})
   -> 0.0038s
-- execute("SET statement_timeout TO 0")
   -> 0.0003s
-- remove_index(:todos, {:algorithm=>:concurrently, :column=>[:author_id, :created_at]})
   -> 0.0259s
-- execute("RESET ALL")
   -> 0.0012s
== 20200306170531 AddIndexOnAuthorIdAndCreatedAtToTodos: reverted (0.0314s) ===