異なる粒度での集計

PostgreSQLでクエリ性能を改善する

Amy McCarty

Instructor

データ粒度 — 詳細度

行を一意にする
1 列以上

Video_Games

id	game	first_yr
012	Grand Theft Auto	1997
234	Legend of Zelda	1986

Game_Platforms

game_id	platform	year
234	FCDS	1986
234	GameCube	2003
234	Wii	2006

異なる粒度の結合

Video_Games の「The Legend of Zelda」行が、Games_Platforms の3行すべてにリンクしている図。

SELECT g.id, g.game, g.first_yr
  , COUNT(p.platform) AS no_platforms
  , MAX(p.year) AS last_platform_yr
  , p.platform
FROM video_games g
INNER JOIN game_platforms p
  ON g.id = p.game_id
GROUP BY g.id, g.game, g.first_yr
, p.platform

異なる粒度の結合

出力

id	game	first_yr	no_platforms	last_platform_yr	platform
234	Legend of Zelda	1986	3	2006	FCDS
234	Legend of Zelda	1986	3	2006	GameCube
234	Legend of Zelda	1986	3	2006	Wii

粒度変更の準備

Game_Platforms

game_id	platform	year
234	FCDS	1986
234	GameCube	2003
234	Wii	2006

粒度変更の準備

SELECT game_id, platform, year
  , COUNT(platform) AS no_platforms
  , MAX(year) AS last_platform_yr
FROM game_platforms
GROUP BY game_id, platform, year

出力

game_id	platform	year	no_platforms	last_platform_yr
234	FCDS	1986	3	2006
234	GameCube	2003	3	2006
234	Wii	2006	3	2006

CTE のおさらい

一時結果を使う単独クエリ
WITH 句

WITH cte AS
  ( SELECT * FROM tableA )

SELECT * FROM cte

結合前に集計できる

CTE が解決

WITH platforms_cte AS 
(  SELECT game_id, platform, year
      , COUNT(platform) AS no_platforms
      , MAX(year) AS last_platform_yr
   FROM game_platforms
   GROUP BY game_id, platform, year )

SELECT g.id, g.game, cte.no_platforms
  , cte.platform AS last_platform
FROM video_games g
INNER JOIN platforms_cte cte 
  ON g.id = cte.game_id
  AND cte.last_platform_yr = cte.year

CTE が解決

WITH platforms_cte AS 
(  SELECT game_id, platform, year
      , COUNT(platform) AS no_platforms
      , MAX(year) AS last_platform_yr
   FROM game_platforms
   GROUP BY game_id, platform, year )

SELECT g.id, g.game, cte.no_platforms
  , cte.platform AS last_platform
FROM video_games g
INNER JOIN platforms_cte cte 
  ON g.id = cte.game_id
  AND cte.last_platform_yr = cte.year

出力

id	game	no_platforms	last_platform
234	Legend of Zelda	3	Wii

結合時に粒度を合わせる

繰り返し・重複なし
必要最小限の結果
二重計上なし

練習しましょう！

PostgreSQLでクエリ性能を改善する