Function aware predicate pushdown

When we cast the values or pass them to functions, predicate pushdown doesn’t work. It would be great if the system could understand that the following queries can also take advantage of predicate pushdown.

select count(*) from table where cast(created_at as date) > date '2017-4-1'

select count(*) from table where date_trunc('day', created_at) > date '2017-4-1'

[buremba]

Yes, it but the problem is that most of our queries are auto-generated and since our backend doesn’t know about the PrestoDB functions and casts, we can’t generate the appropriate query.

I’m aware that most of the functions are irreversible but they don’t always need to be one-to-one reversible. Here is an example query that we execute:

SELECT *
FROM
  (
   SELECT
     "date_trunc"('week', CAST("_time" AS DATE)) "week"
   , "collection"
   , "count"(*) "total"
   FROM table
   GROUP BY 1, 2
) 
WHERE ("week" IN (date '2016-10-03', date '2016-11-21')) 

date_trunc is also not reversible but if we can re-write the query as follows:

SELECT *
FROM
  (
   SELECT
     "date_trunc"('week', CAST("_time" AS DATE)) "week"
   , "collection"
   , "count"(*) "total"
   FROM table 
   WHERE _time BETWEEN date '2016-10-03' and date '2016-10-10' 
           OR _time BETWEEN date '2016-11-21' and date '2016-11-28'
   GROUP BY 1, 2
) 

The query planner can make the predicate pushdown work. It requires a reverse-expression implementation of date_trunc function so an extra interface for functions. It would also solve #6595.

It still requires a lot of work because now we need an interface for optional reverse-expression operation of functions. However; I believe that predicate pushdown is a killer feature because it saves lots of unnecessary IO and has a big impact on typical queries.

[buremba]

Actually, that bot closed 7 issues that I created but if there’s no interest in implementing them in the future, feel free to close. However; I believe that it’s not helpful for the community.