A DSL for Physical Plans

What?

Compiles execution plans written in the DevelOPs execution plan DSL into SQL statements that force SQL Server to use the specified plan.

For example, given an execution plan like this (which represents a merge join of table1 and table2):

PMJOIN(attr1 = attr2, PTABLE(table1), PTABLE(table2))

Our compiler will output:

SELECT *  FROM table1 INNER MERGE JOIN table2 table_right ON (attr1 = attr2) OPTION(FORCE ORDER);

For a more complex execution plan, like this:

PNLJOIN(s_nationkey = n_nationkey, PMJOIN(ps_suppkey = s_suppkey, PTABLE(PS), PTABLE(S)), PSELECT(n_name = 'ASIA', PTABLE(N)))

Our compiler will generate SQL like this:

SELECT *  FROM (SELECT *  FROM PS  INNER MERGE JOIN S table_right ON (ps_suppkey = s_suppkey)) table_left INNER LOOP JOIN (SELECT *  FROM N table_inner WHERE n_name = 'ASIA') table_right ON (s_nationkey = n_nationkey) OPTION(FORCE ORDER);

Note that SQL Server will not allow you to use anything but a nested loop join if the join predicate involves an attribute that is also used within a selection. Our compiler automatically detects and decomposes these queries into sub-queries, so if you use an execution plan like this:

PMJOIN(attr1 = attr2, PSELECT(attr1 = "value", PTABLE(table1)), PTABLE(table2))

Our compiler will produce SQL like this:

SELECT * INTO #temp1 FROM table1 table_inner WHERE attr1 = "value" OPTION(FORCE ORDER);
SELECT * INTO #temp2 FROM #temp1  INNER MERGE JOIN table2 table_right ON (attr1 = attr2) OPTION(FORCE ORDER);

Why?

This compiler allows query optimizer developers to test the performance of queries created by their optimizers against the SQL Server execution engine.

How?

You can download a pre-built JAR file here. To run it, use:

java -jar eptosql.jar --help

To simply translate from standard in, use:

java -jar eptosql.jar

To translate from a file, use:

java -jar eptosql.jar -f FILE

Happy compiling. :)

Who?

Project members:

Brionne Godby < bgodby (at) brandeis (dot) edu >
Rachel Leeman-Munk < rmunk (at) brandeis (dot) edu >
Ryan Marcus < rcmarcus (at) brandeis (dot) edu >

With thanks to:

Zhibo Peng < docp (at) brandeis (dot) edu >

System Design

The following UML diagrams depict design decisions and implementation details.

Presentation

RyanMarcus / EPtoSQL