Comments on: Better Primary Keys, a Benefit to TokuDB’s Auto Increment Semantics http://tokutek.com/2009/08/better_primary_keys_a_benefit_to_tokudbs_auto_increment_semantics/ Database Performance Sat, 28 Aug 2010 12:16:29 +0000 hourly 1 http://wordpress.org/?v=3.0 By: Kristian Nielsen http://tokutek.com/2009/08/better_primary_keys_a_benefit_to_tokudbs_auto_increment_semantics/comment-page-1/#comment-37 Kristian Nielsen Tue, 04 Aug 2009 15:39:18 +0000 http://better_primary_keys_a_benefit_to_tokudbs_auto_increment_semantics#comment-37 My standard trick to do this in InnoDB (or other databases) is this: CREATE TABLE t ( c1 int, c2 varchar(30), c3 date, idx smallint, PRIMARY KEY (c3, idx) ); Then you insert rows like this: INSERT INTO t (c1,c2,c3,idx) SELECT 42, "my c2 value", "2007-01-10", IFNULL(MAX(idx)+1, 0) FROM t WHERE c3 = "2007-01-10" (Don't get fooled by the INSERT ... SELECT; this inserts a single row with constant values except for the `idx` column, which gets the next number in sequence). This gets the same clustered index storage without the need for an autoincrement key. An advantage is that values of `idx` counts from 0 for each distinct date value, so data looks nicer and may be able to use a smaller integer size to save space and index btree size. A disadvantage is the need to compute the MAX() for inserts, so bulk insertion is more complicated. My standard trick to do this in InnoDB (or other databases) is this:

CREATE TABLE t (
c1 int,
c2 varchar(30),
c3 date,
idx smallint,
PRIMARY KEY (c3, idx)
);

Then you insert rows like this:

INSERT INTO t (c1,c2,c3,idx) SELECT 42, “my c2 value”, “2007-01-10″, IFNULL(MAX(idx)+1, 0) FROM t WHERE c3 = “2007-01-10″

(Don’t get fooled by the INSERT … SELECT; this inserts a single row with constant values except for the `idx` column, which gets the next number in sequence).

This gets the same clustered index storage without the need for an autoincrement key.

An advantage is that values of `idx` counts from 0 for each distinct date value, so data looks nicer and may be able to use a smaller integer size to save space and index btree size. A disadvantage is the need to compute the MAX() for inserts, so bulk insertion is more complicated.

]]>