Spark replacement for EXISTS and IN

Solution 1:

SparkSQL doesn't currently have EXISTS & IN. "(Latest) Spark SQL / DataFrames and Datasets Guide / Supported Hive Features"

EXISTS & IN can always be rewritten using JOIN or LEFT SEMI JOIN. "Although Apache Spark SQL currently does not support IN or EXISTS subqueries, you can efficiently implement the semantics by rewriting queries to use LEFT SEMI JOIN." OR can always be rewritten using UNION. AND NOT can be rewritten using EXCEPT.

A table holds the rows that make some predicate (statement parameterized by column names) true:

• The DBA gives the predicates for each base table T with columns T.C,...: T(T.C,...)
• A JOIN holds the rows that make the AND of its arguments' predicates true; for a UNION, the OR; for an EXCEPT, the AND NOT.
• SELECT DISTINCTkept columnsFROMT holds the rows where EXISTS dropped columns [predicate of T].
• TLEFT SEMI JOINU holds the rows where EXISTS U-only columns [predicate of T AND predicate of U].
• TWHEREcondition holds the rows where predicate of T AND condition.

(Re querying generally see this answer.)

So by keeping in mind predicate expressions corresponding to SQL you can use straightforward logic rewrite rules to compose and/or reorganize queries. Eg using UNION here need not be "clumsy" either in terms of readability or execution.

Your original question indicated that you understood that you could use UNION and you have edited variants into your question that excise EXISTS and IN from your original queries. Here is another variant also excising OR.

    select <...>    
    from A, B, C, (select ID from ...) as e
    where
      A.FK_1 = B.PK and
      A.FK_2 = C.PK and
      A.ID = e.id
union
    select <...>    
    from A, B, C, (select ID from ...) as e
    where
      A.FK_1 = B.PK and
      A.FK_2 = C.PK and
      A.ID = e.ID

Your Solution 1 does not do what you think it does. If just one of the exists_clause tables are empty, ie even if there are ID matches available in the other, the FROM cross product of tables is empty and no rows are returned. ("An Unintuitive Consequence of SQL Semantics": Chapter 6 The Database Language SQL sidebar page 264 of Database Systems: The Complete Book 2nd Edition.) A FROM is not just introducing names for rows of tables, it is CROSS JOINing and/or OUTER JOINing them after which ON (for INNER JOINs) and WHERE filter some out.

Performance is typically different for different expressions returning the same rows. This depends on DBMS optimization. Many details, which the DBMS and/or programmer may be able to know and if so may or may not know and may or may not best balance, affect the best way to evaluate a query and the best way to write it. But executing two ORed subselects per row in a WHERE (as in your original queries but also your late Solution 2) is not necessarily better than running one UNION of two SELECTs (as in my query).