llamadb/docs/Query Planner.md

Query Planner

The query plan is a program that iterates, transforms and yields data from database tables.

Because the query plan is basically code, it can be represented as a Lisp-style series of S-expressions. Lisp notation is what this document will use to illustrate both the internal data structure of the query plan and how it's executed.

The top-level expression of a query plan must evaluate to a function that yields the rows returned to the DBMS.

scan

Iterates through all the rows from a database table.

Syntax: (scan TABLE SOURCE-ID YIELD-FN)

If YIELD-FN is not provided, all columns from the table are yielded.

search

Iterates through a range of rows from a database index.

Syntax: (search TABLE SOURCE-ID TABLE-INDEX MIN MAX YIELD-FN)

If YIELD-FN is not provided, all columns from the table are yielded.

temp-group-by

Syntax: (temp-group-by YIELD-IN-FN YIELD-GROUP-FN YIELD-OUT-FN)

map

Iterates a function that yields rows, and transforms them. Useful for subqueries.

Syntax: (map YIELD-IN-FN SOURCE-ID YIELD-OUT-FN)

source_id.column_offset

Shorthand for (column-field source_id column_offset). Returns the matching column from the first row of the source_id.

---

SELECT name FROM employee;

(scan employee A
  (yield A.name))

SELECT name, albumtitle FROM album, artist WHERE album.artistId = artist.artistId;

; Nested scan. Bad for performance, but requires no optimization rules.
(scan album A
  (scan artist B
    (if (= A.artistId B.artistId)
      (yield B.name A.albumtitle))))

; Nested search. More efficient than a nested scan.
(scan album A
  (search artist artist_idx B
    (inclusive A.artistId)      ; minimum key
    (inclusive A.artistId)      ; maximum key
    (yield B.name A.albumtitle)))

SELECT country, count(*) AS population FROM people GROUP BY country;

; Uses a temporary, in-memory table to perform the GROUP BY
(temp-group-by A
  (scan people B)
  (yield B.country)
  (yield A.country (count A)))

; Group by index.
(search-group-by people_country_idx A
  (yield A.country (count A)))

SELECT avg(stats.population) FROM (SELECT count(*) AS population FROM people GROUP BY country) stats;

(aggregate A
  (temp-group-by B
    (scan people C)                ; ungrouped rows
    (yield C.country)              ; what to group by
    (yield B.country (count B)))   ; output of temp-group-by
  (yield (avg A population)))