From: Zardosht Kasheff
Date: March 2 2010 2:54pm
Subject: unique_checks and InnoDB primary key
List-Archive: http://lists.mysql.com/internals/37761
Message-Id: <2f9663ba1003020654o78bae449nffa321c3cf7ebf3f@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1

Hello,

I have a question about the variable unique_checks and InnoDB. I see
that if unique_checks is 0, the storage engine is allowed to ignore
duplicates of secondary unique keys. InnoDB uses this setting to allow
the insert buffer to increase insertion speed.

My question is this: if unique_checks is 0, is InnoDB allowed to
ignore duplicates of the primary key? Looking at code, the answer
seems to be no. And as a result, the insert buffer is not used for the
primary key, correct?

Assuming my understanding above is correct, here is a follow up
question. Suppose I have a table schema:
create table foo (a int, b int, primary key (a,b))engine=InnoDB;
Because the insert buffer cannot be used for the primary key, can I
get better performance by changing the schema to the following?
create table foo (a int, b int, unique key (a,b))engine=InnoDB;

Or will this not work because InnoDB selects the unique key as the
clustered key?

Thanks
-Zardosht