Understanding MySQL Internals: Discovering and Improving a Great Database

Author: Alexander Sasha Pachev, Sasha Pachev
This Year Stack Overflow 3
This Month Stack Overflow 2


by anonymous   2017-08-20

Not PHP specific (as others have mentioned, "advanced" MySQL knowledge should be language-independent), but here you go (from this question and this question):

Understanding MySQL Internals :

Learn how data structures and convenience functions operate, how to add new storage engines and configuration options, and much more

High Performance MySQL:

Learn how to design schemas, indexes, queries and advanced MySQL features for maximum performance, and get detailed guidance for tuning your MySQL server, operating system, and hardware to their fullest potential. You'll also learn practical, safe, high-performance ways to scale your applications with replication, load balancing, high availability, and failover.

Pro MySQL:

Topics include transaction processing and indexing theory, benchmarking and profiling, and advanced coverage of storage engines, data types, subqueries, derived tables, and joins. Also covers MySQL 5's new enterprise features like stored procedures, triggers, and views.

(partial descriptions from Amazon included, see respective product page for more detailed info).

by anonymous   2017-08-20

Seems strange

[root@*****]# perror 28
OS error code  28:  No space left on device
[root@*****]# perror 32
OS error code  32:  Broken pipe

Since the mysqldump keeps breaking at random places, it is space-related, and no disk full condition, I would suspect the problem at a deeper layer : the MySQL Packet. What is a MySQL Packet?

According to the page 99 of the Book


here are paragraphs 1-3 explaining it:

MySQL network communication code was written under the assumption that queries are always reasonably short, and therefore can be sent to and processed by the server in one chunk, which is called a packet in MySQL terminology. The server allocates the memory for a temporary buffer to store the packet, and it requests enough to fit it entirely. This architecture requires a precaution to avoid having the server run out of memory---a cap on the size of the packet, which this option accomplishes.

The code of interest in relation to this option is found in sql/net_serv.cc. Take a look at my_net_read(), then follow the call to my_real_read() and pay particular attention to net_realloc().

This variable also limits the length of a result of many string functons. See sql/field.cc and sql/intem_strfunc.cc for details.

Given this explanation, making bulk INSERTs will load/unload a MySQL Packet rather quickly. This is especially true when max_allowed_packet is too small for the given load of data coming at it.

I wrote about this before : MySQL server has gone away obstructing import of large dumps

Try raising max_allowed_packet for the mysqldump to 1G as follows:

mysqldump --max-allowed-packet=1073741824 ...

and try the mysqldump.

If this does not do it, then do this:

Added this to my.cnf

max_allowed_packet = 1G

Then, login to MySQL as root@localhost and run this

mysql> SET GLOBAL max_allowed_packet = 1024 * 1024 * 1024;

and try the mysqldump.

Give it a Try !!!

by anonymous   2017-08-20

The information you have in the question concerning MyISAM is right on target. However, I would like to address your two additional questions:


What if users update existed data with longer data? Will MyISAM marked the record as deleted and find place that fits the new data or simply use overflow pointer to point to unfitted data?

According to the Book


Chapter 10 : "Storage Engines" Page 196 Paragraph 7 says

For records with variable length, the format is more complicated. The first byte contains a special code describing the subtype of the record. The meaning of the subsequent bytes varies with each subtype, but the common theme is that there is a sequence of bytes that contains the length of the record, the number of unused bytes in the block, NULL value indicator flags, and possibly a pointer to the continuation of the record if the record did not fit into the previously created space and had to be split up. This can happen when one record gets deleted, and a new one to be inserted into its place exceeds the original one is size. You can get the details of the meanings of different codes by studying the switch statement in_mi_get_block_info() in storage/myisam/mi_dynrec.c.

Based on that paragraph, the old record gets overwritten with linkage data only if the new data to insert cannot fit in the previously allocated block. This can result in many bloated rows.


Would it be very inefficient if the table has been deleted and inserted for many times since the record structure could potentially full of overflow pointers and unused space?

From my previous answer, there would be lots of blocks that have

  • block of space
  • the length of the record
  • the number of unused bytes in the block
  • NULL value indicator flags
  • possibly a pointer to the continuation of the record if the record did not fit into the previously created space and had to be split up

Such record links would start in the front of every row that have oversized data being inserted. This can bloat a MyISAM tables .MYD file very quickly.


The default row format of a MyISAM is Dynamic. When a table is Dynamic and experiences lots of INSERTs, UPDATEs, and DELETEs, such a table would need to optimized with


There is an alternative: switch the table's row format to Fixed. That way, all rows are the same size. This is how you make the row format Fixed:


Even with a Fixed Row Format, time must be taken to locate an available record but the time would be O(1) search time (In layman's terms, it would take the same amount of time to locate an available record no matter how many rows the table has or how many deleted rows there are). You could bypass that step by enabling concurrent_insert as follows:

Add this to my.cnf

concurrent_insert = 2

MySQL restart not required. Just run

mysql> SET GLOBAL concurrent_insert = 2;

This would cause all INSERTs to go to the back of the table without looking for free space.

Advantage of Fixed Row tables

  • INSERTs, UPDATEs, and DELETEs would be somewhat faster
  • SELECT are 20-25% faster

Here are some of my posts on SELECT being faster for Row Formats being Fixed

  • May 03, 2012 : Which is faster, InnoDB or MyISAM?
  • Sep 20, 2011 : Best of MyISAM and InnoDB
  • May 10, 2011 : What is the performance impact of using CHAR vs VARCHAR on a fixed-size field?

Disadvantage of Fixed Row tables

In most cases, when you run ALTER TABLE mytable ROW_FORMAT=Fixed;, the table may grow 80-100%. The .MYI file (index pages for the MyISAM table) would also grow at the same rate.


If you want speed for MyISAM tables and can live with bigger tables, my alternate suggestions would be needed. If you want to conserve space for each MyISAM table, leave the row format as is (Dynamic). You will have to compress the table with OPTIMIZE TABLE mytable; more frequent with Dynamic tables.