1. Recover Table Structure From InnoDB Dictionary

Recover Table Structure From InnoDB Dictionary

When a table gets dropped, MySQL removes the respective .frm file. This post explains how to recover the table structure if the table was dropped.

You need the table structure to recover a dropped table from the InnoDB tablespace. The B+tree structure of the InnoDB index doesn’t contain any information about field types. MySQL needs to know that in order to access records of the InnoDB table. Normally, MySQL gets the table structure from the .frm file. But when MySQL drops a table the respective frm file removed too.

Fortunately, there’s one more place where MySQL keeps the table structure. It’s the InnoDB dictionary.

The InnoDB dictionary is a set of tables where InnoDB keeps information about the tables. I reviewed them in detail in a separate InnoDB Dictionary post earlier. After the DROP, InnoDB deletes records related to the dropped table from the dictionary. So, we need to recover deleted records from the dictionary and then get the table structure.

Compiling Data Recovery Tool

First, we need to get the source code. The code is hosted on GitHub.

To compile it, we need gcc, bison and flex.

Recover The InnoDB Dictionary

Now, let’s create dictionary tables in the sakila_recovered database. The data recovery tool comes with the structure of the dictionary tables.

The dictionary is stored in the ibdata1 file. So, let’s parse it.

Now, we need to extract the dictionary records from the InnoDB pages. Let’s create a directory for the table dumps.

And now we can generate table dumps and LOAD INFILE commands to load the dumps. We also need to specify the -D option to the c_parser because the records we need were deleted from the dictionary when the table was dropped.

SYS_TABLES

SYS_INDEXES

SYS_COLUMNS

and SYS_FIELDS

With the generated LOAD INFILE commands it’s easy to load the dumps.

Now we have the InnoDB dictionary loaded into normal InnoDB tables.

Compiling Sys_parser

Sys_parser is a tool that reads the dictionary from tables stored in MySQL and generates the CREATE TABLE structure for a table.

To compile it we need MySQL libraries and development files. Depending on distribution, they may be in -devel or -dev package. On RedHat based systems, you can check it with the command yum provides “*/mysql_config” . On my server it was the mysql-community-devel package.

If all necessary packages are installed, the compilation boils down to a simple command:

Recover Table Structure

Now sys_parser can do its magic. Just run it to get the CREATE statement in the standard output.

It will use root as the username to connect to MySQL, querty as the password. The dictionary is stored in SYS_* tables in the sakila_recovered database. What we want to recover is sakila.actor. InnoDB uses a slash ‘/’ as a separator between database name and table name, so does sys_parser.

There are few caveats though.

  1. InnoDB doesn’t store all information you can find in the frm file. For example, if a field is AUTO_INCREMENT, the InnoDB dictionary knows nothing about it. Therefore, sys_parser won’t recover that property. If there were any field or table level comments, they’ll be lost.
  2. Sys_parser generates the table structure eligible for further data recovery. It could but it doesn’t recover secondary indexes, or foreign keys.
  3. InnoDB doesn’t store the DECIMAL type as a binary string. It doesn’t store the precision of a DECIMAL field. So, that information will be lost.

For example, table payment uses DECIMAL to store money.

Fortunately, Oracle is planning to extend the InnoDB dictionary and finally get rid of .frm files. I salute that decision, having the structure in two places leads to inconsistencies.

Have a question? Ask the experts!

Previous Post Next Post