ACID

For other uses, see Acid (disambiguation).

In computer science, ACID (atomicity, consistency, isolation, durability) is a set of properties that guarantee database transactions are processed reliably. In the context of databases, a single logical operation on the data is called a transaction. For example, a transfer of funds from one bank account to another, even though that might involve multiple changes (such as debiting one account and crediting another), is a single transaction.

Jim Gray defined these properties of a reliable transaction system in the late 1970s and developed technologies to automatically achieve them. In 1983, Andreas Reuter and Theo Haerder coined the acronym ACID to describe them.

Characteristics

Atomicity

Main article: Atomicity (database systems)

Atomicity requires that database modifications must follow an "all or nothing" rule. Each transaction is said to be atomic. If one part of the transaction fails, the entire transaction fails and the database state is left unchanged. It is critical that the database management system maintain the atomic nature of transactions in spite of any application, DBMS (Database Management System), operating system or hardware failure.

An atomic transfer cannot be subdivided and must be processed in its entirety or not at all. Atomicity means that users do not have to worry about the effect of incomplete transactions.

Transactions can fail for several kinds of reasons:

1. Hardware failure: A disk drive fails, preventing some of the transaction's database changes from taking effect.
2. System failure: The user loses their connection to the application before providing all necessary information.
3. Database failure: E.g., the database runs out of room to hold additional data.
4. Application failure: The application attempts to post data that violates a rule that the database itself enforces, such as attempting to insert a duplicate value in a unique column.

Consistency

Main article: Consistency (database systems)

It ensures the truthfulness of the database.

The consistency property ensures that any transaction the database performs will take it from one consistent state to another.

Consistency states that only valid data will be written to the database.

The consistency property does not say how the DBMS should handle an inconsistency other than ensure the database is clean at the end of the transaction. If, for some reason, a transaction is executed that violates the database’s consistency rules, the entire transaction could be rolled back to the pre-transactional state - or it would be equally valid for the DBMS to take some patch-up action to get the database in a consistent state. Thus, if the database schema says that a particular field is for holding integer numbers, the DBMS could decide to reject attempts to put fractional values there, or it could round the supplied values to the nearest whole number: both options maintain consistency.

The consistency rule applies only to integrity rules that are within its scope. Thus, if a DBMS allows fields of a record to act as references to another record, then consistency implies the DBMS must enforce referential integrity: by the time any transaction ends, each and every reference in the database must be valid. If a transaction consisted of an attempt to delete a record referenced by another, each of the following mechanisms would maintain consistency:

• abort the transaction, rolling back to the consistent, prior state;

• delete all records that reference the deleted record (this is known as cascade delete); or,

• nullify the relevant fields in all records that point to the deleted record.

These are examples of propagation constraints; some database systems allow the database designer to specify which option to choose when setting up the schema for a database.

Application developers are responsible for ensuring application level consistency, over and above that offered by the DBMS. Thus, if a user withdraws funds from an account and the new balance is lower than the account's minimum balance threshold, as far as the DBMS is concerned, the database is in a consistent state even though this rule (unknown to the DBMS) has been violated.

Isolation

Main article: Isolation (database systems)

Isolation refers to the requirement that other operations cannot access data that has been modified during a transaction that has not yet completed. The question of isolation occurs in case of concurrent transactions (multiple transactions occurring at the same time). Each transaction must remain unaware of other concurrently executing transactions, except that one transaction may be forced to wait for the completion of another transaction that has modified data that the waiting transaction requires. If the isolation system does not exist, then the data could be put into an inconsistent state. This could happen, if one transaction is in the process of modifying data but has not yet completed, and then a second transaction reads and modifies that uncommitted data from the first transaction. If the first transaction fails and the second one succeeds, that violation of transactional isolation will cause data inconsistency. Due to performance and deadlocking concerns with multiple competing transactions, some modern databases allow dirty reads which is a way to bypass some of the restrictions of the isolation system. A dirty read means that a transaction is allowed to read, but not modify, the uncommitted data from another transaction. Another way to provide isolation for read transactions is via MVCC which gets around the blocking lock issues of reads blocking writes. The read is done on a prior version of data and not on the data that is being locked for modification thus providing the necessary isolation between transactions.

Durability

Main article: Durability (database systems)

Durability is the ability of the DBMS to recover the committed transaction updates against any kind of system failure (hardware or software). Durability is the DBMS's guarantee that once the user has been notified of a transaction's success the transaction will not be lost, the transaction's data changes will survive system failure, and that all integrity constraints have been satisfied, so the DBMS won't need to reverse the transaction. Many DBMSs implement durability by writing transactions into a transaction log that can be reprocessed to recreate the system state right before any later failure. A transaction is deemed committed only after it is entered in the log.

Durability does not imply a permanent state of the database. A subsequent transaction may modify data changed by a prior transaction without violating the durability principle.

Examples

The following examples are used to further explain the ACID properties. In these examples, the database has two fields, A and B, in two records. An integrity constraint requires that the value in A and the value in B must sum to 100. The following SQL code creates a table as described above:

CREATE TABLE acidtest (A INTEGER, B INTEGER CHECK (A + B = 100));

Atomicity failure

The transaction subtracts 10 from A and adds 10 to B. If it succeeds, it would be valid, because the data continues to satisfy the constraint. However, assume that after removing 10 from A, the transaction is unable to modify B. If the database retains A's new value, atomicity and the constraint would both be violated. Atomicity requires that both parts of this transaction complete or neither.

Consistency failure

Consistency is a very general term that demands the data meets all validation rules. In the previous example, the validation is a requirement that A + B = 100. Also, it may be implied that both A and B must be integers. A valid range for A and B may also be implied. All validation rules must be checked to ensure consistency.

Assume that a transaction attempts to subtract 10 from A without altering B. Because consistency is checked after each transaction, it is known that A + B = 100 before the transaction begins. If the transaction removes 10 from A successfully, atomicity will be achieved. However, a validation check will show that A + B = 90. That is not consistent according to the rules of the database. The entire transaction must be undone.

Isolation failure

To demonstrate isolation, we assume two transactions execute at the same time, each attempting to modify the same data. One of the two must wait until the other completes in order to maintain isolation.

Consider two transactions. T₁ transfers 10 from A to B. T₂ transfers 10 from B to A. Combined, there are four actions:

• subtract 10 from A
• add 10 to B.
• subtract 10 from B
• add 10 to A.

If these operations are performed in order, isolation is maintained, although T₂ must wait. Consider what happens, if T₁ fails half-way through. The database eliminates T₁'s effects, and T₂ sees only valid data.

By interleaving the transactions, the actual order of actions might be: $A-10$, $B-10$, $B+10$, $A+10$. Again consider what happens, if T₁ fails. T₁ still subtracts 10 from A. Now, T₂ adds 10 to A restoring it to its initial value. Now T₁ fails. What should A's value be? T₂ has already changed it. Also, T₁ never changed B. T₂ subtracts 10 from it. If T₂ is allowed to complete, B's value will be 10 too low, and A's value will be unchanged, leaving an invalid database. This is known as a write-write failure, because two transactions attempted to write to the same data field.

Durability failure

Assume that a transaction transfers 10 from A to B. It removes 10 from A. It then adds 10 to B. At this point, a "success" message is sent to the user. However, the changes are still queued in the disk buffer waiting to be committed to the disk. Power fails and the changes are lost. The user assumes that the changes have been made.

Implementation

Processing a transaction often requires a sequence of operations that is subject to failure for a number of reasons. For instance, the system may have no room left on its disk drives, or it may have used up its allocated CPU time.

There are two popular families of techniques: write ahead logging and shadow paging. In both cases, locks must be acquired on all information that is updated, and depending on the level of isolation, possibly on all data that is read as well. In write ahead logging, atomicity is guaranteed by copying the original (unchanged) data to a log before changing the database. That allows the database to return to a consistent state in the event of a crash.

In shadowing, updates are applied to a partial copy of the database, and the new copy is activated when the transaction commits.

Locking vs multiversioning

Many databases rely upon locking to provide ACID capabilities. Locking means that the transaction marks the data that it accesses so that the DBMS knows not to allow other transactions to modify it until the first transaction succeeds or fails. The lock must always be acquired before processing data, including data that are read but not modified. Non-trivial transactions typically require a large number of locks, resulting in substantial overhead as well as blocking other transactions. For example, if user A is running a transaction that has to read a row of data that user B wants to modify, user B must wait until user A's transaction completes. Two phase locking is often applied to guarantee full isolation.

An alternative to locking is multiversion concurrency control, in which the database provides each reading transaction the prior, unmodified version of data that is being modified by another active transaction. This allows readers to operate without acquiring locks. I.e., writing transactions do not block reading transactions, and readers do not block writers. Going back to the example, when user A's transaction requests data that user B is modifying, the database provides A with the version of that data that existed when user B started his transaction. User A gets a consistent view of the database even if other users are changing data. One implementation relaxes the isolation property, namely snapshot isolation.

Distributed transactions

Guaranteeing ACID properties in a distributed transaction across a distributed database where no single node is responsible for all data affecting a transaction presents additional complications. Network connections might fail, or one node might successfully complete its part of the transaction and then be required to roll back its changes, because of a failure on another node. The two-phase commit protocol (not to be confused with two-phase locking) provides atomicity for distributed transactions to ensure that each participant in the transaction agrees on whether the transaction should be committed or not. Briefly, in the first phase, one node (the coordinator) interrogates the other nodes (the participants) and only when all reply that they are prepared does the coordinator, in the second phase, formalize the transaction.

See also

• Open Systems Interconnection
• Transactional NTFS
• Basically Available, Soft state, Eventual consistency (BASE)
• Concurrency control

References

• Gray, Jim (September 1981). "The Transaction Concept: Virtues and Limitations" (PDF). Proceedings of the 7th International Conference on Very Large Databases. 19333 Vallco Parkway, Cupertino CA 95014: Tandem Computers. pp. 144–154. Retrieved 2006-11-09. {{cite conference}}: Unknown parameter |booktitle= ignored (|book-title= suggested) (help)CS1 maint: location (link)

• Gray, Jim, and Reuter, Andreas, Distributed Transaction Processing: Concepts and Techniques. Morgan Kaufmann, 1993. ISBN 1-55860-190-2.