1. InnoDB Dictionary

InnoDB Dictionary

Why Does InnoDB Need A Dictionary

An InnoDB dictionary is a set of internal tables InnoDB uses to maintain various information about user tables. It serves as API between a human and the database. While the humans refer to tables by their names, InnoDB works with integer identifiers. The dictionary stores correspondence between table name and index_id.

The dictionary tables are normal InnoDB tables, but they’re not visible to a user. However, some versions of MySQL provide read-only access to the dictionary in information_schema database.

The dictionary is stored in ibdata1. Root page of SYS_TABLES, for example, has id 8, so it’s the eighth page from the beginning of ibdata1.

The dictionary pages are in REDUNDANT format even if you use MySQL 5.6. I hope to write more about record formats in future posts. For now, it’s enough to mention that REDUNDANT is the oldest record format. It was available since 4.0 and was the only format by that time.


All dictionary tables have predefined hard-coded index_ids. No surprise, as long as InnoDB doesn’t have a dictionary for the dictionary :). SYS_TABLES has index_id 0-1. Root page of SYS_TABLES is page id 8.

NAME is the human readable table name. Well, the database name and the table name. For example, for table sakila.actor NAME will be “sakila/actor”.

ID is the table identifier. Just an integer number. The table is referenced to by this table_id in other dictionary tables.

N_COLS is the number of columns in the table.

I will pretend TYPE, MIX_ID, MIX_LEN and CLUSTER_NAME are not important. Actually, I have no idea what these fields are. Anyway, they were never used for data recovery, so let’s leave them for future investigations.

SPACE is an identifier of a tablespace. Thus ibdata1 is SPACE 0, ibdata2 is SPACE 1, Every ibd file has its own SPACE identifier. SPACE tells in what tablespace the table’s indexes are stored.

Sample record for table sakila.actor:

“sakila/actor” is the table name obviously. 61 is the table id. 4 is the number of fields in the table:

47 is the SPACE identifier. It would be 0 if innodb_file_per_table was disabled. It’s not in my environment:


Another important dictionary table is SYS_INDEXES. It stores information about InnoDB indexes.

Its index_id is 0-3 and the root page is stored in page 11.

Fields of the table:

TABLE_ID is the table identifier. It’s one from SYS_TABLES.

ID is the index identifier. This is a very important field. The matter is, every InnoDB page contains the index_id of the index it belongs to in the header. So, if we want to recover a table, we need to find all pages that belong to the PRIMARY index of the table, which means we need all pages with a specific index_id.

NAME is the name of an index. Table sakila.actor has two indexes: PRIMARY and idx_actor_last_name. So, these are names in SYS_INDEXES.

N_FIELDS is the number of fields in an index. idx_actor_last_name has one field, so is it in SYS_INDEXES.

TYPE is not an important field 🙂

SPACE is an InnoDB tablespace identifier showing where the index is stored. Probably, InnoDB developers foresaw the possibility to store indexes in different files, but for current MySQL versions it’s the same as in SYS_TABLES.

PAGE_NO is an identifier of the root page of the index. As you know, InnoDB stores indexes in B+tree data structure. Thus, root node of the B+tree is the root page identified by this field.



SYS_COLUMNS describes individual fields in a table. It is stored as index_id 0-2.

TABLE_ID is the table identifier, it comes from SYS_TABLES

POS is a relative position of the field in the table.

NAME is the name of the field.

MTYPE and PRTYPE encode the MySQL type of the field. InnoDB started its history as a separate product, not related to MySQL. These two fields are legacy from that time. There’s no strict correspondence between MySQL and the InnoDB type. InnoDB uses its own type that suits a particular MySQL type the best. Encoding and collation are stored here as well.

LEN is the length of the field.

PREC is precision for some MySQL types.

Isn’t it interesting to see an example of the table for sakila.actor? Here it goes. Remember, table_id is 61? We know that from SYS_TABLES.

As we see, smallint is two bytes, timestamp is four bytes. And varchar(45) is 135 bytes. Apparently, this is the maximum size, not the actual size of a field which is different for every record. But why 135, not 45? Because the field is utf8, where each character may use up to three bytes. 3 x 45 adds up to 135.


The last table for today (but not least) is SYS_FIELDS. It’s stored in index_id 0-4.
It lists individual fields on all indexes.

The table is quite simple.

INDEX_ID is the index identifier that we can find in SYS_INDEXES.

POS is the position of the field in the index.

And COL_NAME is the field name.

Why Do We Care?

Why do we need an InnoDB dictionary anyway?

There are two reasons. First, if we need to recover table A, we need to know index_id of its PRIMARY index. The easiest way to learn it is the InnoDB dictionary.

Second, as you see, the InnoDB dictionary stores some information about field names and types. Which means, in case we don’t know the table structure we can recover it from the dictionary.

Have a question? Ask the experts!

Previous Post Next Post