Comments on: Some little known facts about Innodb Insert Buffer

By: mz

mz — Wed, 15 Dec 2010 00:07:29 +0000

Innodb locking problem with unique keys when inserting

CREATE TABLE `male_id` (
`entry_id` int(10) unsigned NOT NULL AUTO_INCREMENT,
`jy_id` int(10) unsigned NOT NULL,
PRIMARY KEY (`entry_id`),
UNIQUE KEY `jy_id` (`jy_id`)
) ENGINE=InnoDB AUTO_INCREMENT=1

When 10 processes concurrently inserting rows to the above table, only the rows from one process are actually recorded in the table. I tested with autocommit=1 and even when transaction_isolation_level=seriazable. It seems Innodb is not doing locking properly. If I remove the primary key (entry_id), the problem remains. But the problem is gone if I change the unique key (jy_id) to be index (jy_id).

By: Wagner Bianchi

Wagner Bianchi — Mon, 09 Nov 2009 17:30:07 +0000

OK, Peter…thank you…I was missunderstood.

By: peter

peter — Fri, 06 Nov 2009 16:41:27 +0000

Wagner,

It is not pages but records which are being inserted. Merged records are those which are in place on the pages where should be. Unmerged records are ones stored in insert buffer and so still to be merged.

By: Wagner Bianchi

Wagner Bianchi — Fri, 06 Nov 2009 15:55:53 +0000

Can you give me an explaination about `merged` and `unmerged` pages?

By: Dan Dascalescu

Dan Dascalescu — Thu, 09 Jul 2009 11:19:45 +0000

Solid state drives are not the big revolution. The big revolution is nano-wire memory: 1000 faster than Flash memory, terabyte capacity, and lower power consumption. This will render databases obsolete, along with silly tricks involving hard disk rotation cycles.

By: Morgan Tocker

Morgan Tocker — Mon, 25 May 2009 20:07:55 +0000

Pavel,

The 100k max writes refers to the cheaper MLC (Multi layer) Flash. SLC (aimed more at the server end) can sustain a higher number of writes.

By: Pavel

Pavel — Mon, 02 Feb 2009 12:21:31 +0000

Pat, will solid state disks survive database workload with their 100k possible writes only?

By: peter

peter — Fri, 16 Jan 2009 06:53:24 +0000

Baron,

I planned another post about it. As I remember this patch is still in development, is not it ?

By: Baron Schwartz

Baron Schwartz — Fri, 16 Jan 2009 05:07:06 +0000

You didn’t mention that the Percona patches provide control over the insert buffer.

By: peter

peter — Fri, 16 Jan 2009 03:17:31 +0000

Pat,

You’re right. The storage industry is changed and I expect the best performance will be reached with systems which have not been written yet. Storage engines as well as MySQL often has assumptions about disk being rotating disk.