AtomicityThe atomicity feature states that for a transaction to be considered successful, all steps within the transaction must complete successfully. Either all the steps should be applied to the database, or none of them should. A transaction should not be allowed to complete partway.
To support atomicity, MySQL uses a system called commit and rollback. Database actions are only temporarily performed during a transaction. When it appears that all the actions in a transaction would complete successfully, the transaction is committed (the server applies all the actions to the database). If it appears that any one of the actions would fail, the entire transaction is rolled back (any previous steps that were successful are reversed). This ensures that the transaction is completed as a whole.
MySQL uses the two-phase commit approach to committing transactions. The two-phase commit performs the transaction using two steps (or phases):
- Prepare phase: A transaction is analyzed to determine if the database is able to commit the entire transaction.
- Commit phase: The transaction is physically committed to the database.
ConsistencyThe concept of consistency is a little more difficult than atomicity. The consistency feature states that every transaction should leave the database in a valid state. The tricky part here is what is considered a “valid state.”
Often, this feature is applied to how a database server handles unexpected crashes. If the database takes a power hit in the middle of the commit phase of a multi-action transaction, can it leave the tables in a state where the data makes sense?
MySQL utilizes two features to accomplish consistency:
- Double-write buffering: With double-write buffering, before MySQL writes data to the actual tables, it stores the data in a buffer area. Only after all the transaction data is written to the buffer area will MySQL write the buffer area data to the actual table data files.
- Crash recovery: If there is a system crash before the buffer area is completely written to the table files, MySQL can recover the buffer area using the crash recovery feature, which recovers submitted transactions from a transaction log file.
IsolationThe isolation feature is required for multiuser databases. When there is more than one person modifying data in a database, odd things can happen. If two people try to modify the same data value at the same time, who’s to say which value is the final value?
When more than one person tries to access the same data, the DBMS must act as the traffic cop, directing who gets access to the data first. Isolation ensures that each transaction in progress is invisible to any other transaction in progress. The DBMS must allow each transaction to complete and then decide which transaction value is the final value for the data. It accomplishes this task using a feature called locking.
Locking does what it says: It locks data while a transaction is being committed to the database. While the data is locked, other users can’t access the data, not even for queries. This prevents multiple users from querying or modifying the data while it’s in a locked mode.
There are two basic levels of locking that MySQL uses to support isolation:
- Table-level locking: With table-level locking, any time a user requires a modification to a data record in a table, the DBMS locks the entire table, preventing other users from even viewing data in the table. As you can guess, this has an adverse effect on database performance, especially in environments where there is a lot of change to the data in the database. Early DBMS implementations used table-level locking exclusively.
- Row-level locking: To solve the problems of table-level locking, many DBMS implementations (including the MySQL InnoDB storage engine) now incorporate row-level locking. With row-level locking, the DBMS locks only the data record that’s being modified. The rest of the table is available for other users to access.
DurabilityThe durability feature states that when a transaction is committed to the database, it must not be lost. This sounds like a simple concept, but in reality durability is often harder to ensure than it sounds.
Durability means being able to withstand both hardware and software failures. A database is useless if a power outage or server crash compromises the data stored in the database.
MySQL supports durability by incorporating multiple layers of protection. The same double-write buffer and crash recovery features mentioned for the consistency feature also apply to the durability feature. MySQL writes all transactions to a log file, writes the changes to the double-write buffer area, and then writes them to the actual database files. If the system crashes during this process, most of the time MySQL can recover the transaction within the process.
The onus of durability also rests on the database administrator. Having a good uninterruptable power supply (UPS) for your database server, as well as performing regular database backups, is crucial to ensuring your database tables are safe.