14.2.4.3. InnoDB INFORMATION_SCHEMA tables

The INNODB_CMP and INNODB_CMP_RESET tables contain status information on the operations related to compressed tables, which are covered in Section 5.4.6, "Working with InnoDB Compressed Tables". The compressed page size is in the column PAGE_SIZE.

These two tables have identical contents, but reading from INNODB_CMP_RESET resets the statistics on compression and uncompression operations. For example, if you archive the output of INNODB_CMP_RESET every 60 minutes, you see the statistics for each hourly period. If you monitor the output of INNODB_CMP (making sure never to read INNODB_CMP_RESET), you see the cumulated statistics since InnoDB was started.

For the table definition, see Table 20.1, "Columns of INNODB_CMP and INNODB_CMP_RESET".

The INNODB_CMPMEM and INNODB_CMPMEM_RESET tables contain status information on the compressed pages that reside in the buffer pool. Please consult Section 5.4.6, "Working with InnoDB Compressed Tables" for further information on compressed tables and the use of the buffer pool. The INNODB_CMP and INNODB_CMP_RESET tables should provide more useful statistics on compression.

InnoDB uses a buddy allocator system to manage memory allocated to pages of various sizes, from 1KB to 16KB. Each row of the two tables described here corresponds to a single page size.

These two tables have identical contents, but reading from INNODB_CMPMEM_RESET resets the statistics on relocation operations. For example, if every 60 minutes you archived the output of INNODB_CMPMEM_RESET, it would show the hourly statistics. If you never read INNODB_CMPMEM_RESET and monitored the output of INNODB_CMPMEM instead, it would show the cumulated statistics since InnoDB was started.

For the table definition, see Table 20.3, "Columns of INNODB_CMPMEM and INNODB_CMPMEM_RESET".

When a transaction updates a row in a table, or locks it with SELECT FOR UPDATE, InnoDB establishes a list or queue of locks on that row. Similarly, InnoDB maintains a list of locks on a table for table-level locks transactions hold. If a second transaction wants to update a row or lock a table already locked by a prior transaction in an incompatible mode, InnoDB adds a lock request for the row to the corresponding queue. For a lock to be acquired by a transaction, all incompatible lock requests previously entered into the lock queue for that row or table must be removed (the transactions holding or requesting those locks either commit or roll back).

A transaction may have any number of lock requests for different rows or tables. At any given time, a transaction may be requesting a lock that is held by another transaction, in which case it is blocked by that other transaction. The requesting transaction must wait for the transaction that holds the blocking lock to commit or rollback. If a transaction is not waiting for a a lock, it is in the 'RUNNING' state. If a transaction is waiting for a lock, it is in the 'LOCK WAIT' state.

The INNODB_LOCKS table holds one or more row for each 'LOCK WAIT' transaction, indicating any lock requests that are preventing its progress. This table also contains one row describing each lock in a queue of locks pending for a given row or table. The INNODB_LOCK_WAITS table shows which locks already held by a transaction are blocking locks requested by other transactions.

The data exposed by the transaction and locking tables represent a glimpse into fast-changing data. This is not like other (user) tables, where the data changes only when application-initiated updates occur. The underlying data is internal system-managed data, and can change very quickly.

For performance reasons, and to minimize the chance of misleading JOINs between the INFORMATION_SCHEMA tables, InnoDB collects the required transaction and locking information into an intermediate buffer whenever a SELECT on any of the tables is issued. This buffer is refreshed only if more than 0.1 seconds has elapsed since the last time the buffer was read. The data needed to fill the three tables is fetched atomically and consistently and is saved in this global internal buffer, forming a point-in-time "snapshot". If multiple table accesses occur within 0.1 seconds (as they almost certainly do when MySQL processes a join among these tables), then the same snapshot is used to satisfy the query.

A correct result is returned when you JOIN any of these tables together in a single query, because the data for the three tables comes from the same snapshot. Because the buffer is not refreshed with every query of any of these tables, if you issue separate queries against these tables within a tenth of a second, the results are the same from query to query. On the other hand, two separate queries of the same or different tables issued more than a tenth of a second apart may see different results, since the data come from different snapshots.

Because InnoDB must temporarily stall while the transaction and locking data is collected, too frequent queries of these tables can negatively impact performance as seen by other users.

As these tables contain sensitive information (at least INNODB_LOCKS.LOCK_DATA and INNODB_TRX.TRX_QUERY), for security reasons, only the users with the PROCESS privilege are allowed to SELECT from them.

As just described, while the transaction and locking data is correct and consistent when these INFORMATION_SCHEMA tables are populated. For example, the query in INNODB_TRX is always consistent with the rest of INNODB_TRX, INNODB_LOCKS and INNODB_LOCK_WAITS when the data comes from the same snapshot. However, the underlying data changes so fast that similar glimpses at other, similarly fast-changing data, may not be in synchrony. Thus, you should be careful in comparing the data in the InnoDB transaction and locking tables with that in the PROCESSLIST table. The data from the PROCESSLIST table does not come from the same snapshot as the data about locking and transactions. Even if you issue a single SELECT (joining INNODB_TRX and PROCESSLIST, for example), the content of those tables is generally not consistent. INNODB_TRX may reference rows that are not present in PROCESSLIST or the currently executing SQL query of a transaction, shown in INNODB_TRX.TRX_QUERY may differ from the one in PROCESSLIST.INFO.