Transactions and optimistic concurrency control

Database transactions provide a safe and predictable programming model to deal with concurrent changes to the data. Traditional relational databases, like SQL Server, allow you to write the business logic using stored-procedures and/or triggers, send it to the server for execution directly within the database engine. With traditional relational databases, you are required to deal with two different programming languages the (non-transactional) application programming language such as JavaScript, Python, C#, Java, etc. and the transactional programming language (such as T-SQL) that is natively executed by the database.

The database engine in Azure Cosmos DB supports full ACID (Atomicity, Consistency, Isolation, Durability) compliant transactions with snapshot isolation. All the database operations within the scope of a container's logical partition are transactionally executed within the database engine that is hosted by the replica of the partition. These operations include both write (updating one or more items within the logical partition) and read operations. The following table illustrates different operations and transaction types:

Operation Operation Type Single or Multi Item Transaction
Insert (without a pre/post trigger) Write Single item transaction
Insert (with a pre/post trigger) Write and Read Multi-item transaction
Replace (without a pre/post trigger) Write Single item transaction
Replace (with a pre/post trigger) Write and Read Multi-item transaction
Upsert (without a pre/post trigger) Write Single item transaction
Upsert (with a pre/post trigger) Write and Read Multi-item transaction
Delete (without a pre/post trigger) Write Single item transaction
Delete (with a pre/post trigger) Write and Read Multi-item transaction
Execute stored procedure Write and Read Multi-item transaction
System initiated execution of a merge procedure Write Multi-item transaction
System initiated execution of deleting items based on expiration (TTL) of an item Write Multi-item transaction
Read Read Single-item transaction
Change Feed Read Multi-item transaction
Paginated Read Read Multi-item transaction
Paginated Query Read Multi-item transaction
Execute UDF as part of the paginated query Read Multi-item transaction

Multi-item transactions

Azure Cosmos DB allows you to write stored procedures, pre/post triggers, user-defined-functions (UDFs) and merge procedures in JavaScript. Azure Cosmos DB natively supports JavaScript execution inside its database engine. You can register stored procedures, pre/post triggers, user-defined-functions (UDFs) and merge procedures on a container and later execute them transactionally within the Azure Cosmos database engine. Writing application logic in JavaScript allows natural expression of control flow, variable scoping, assignment, and integration of exception handling primitives within the database transactions directly in the JavaScript language.

The JavaScript-based stored procedures, triggers, UDFs, and merge procedures are wrapped within an ambient ACID transaction with snapshot isolation across all items within the logical partition. During the course of its execution, if the JavaScript program throws an exception, the entire transaction is aborted and rolled-back. The resulting programming model is simple yet powerful. JavaScript developers get a durable programming model while still using their familiar language constructs and library primitives.

The ability to execute JavaScript directly within the database engine provides performance and transactional execution of database operations against the items of a container. Furthermore, since Azure Cosmos database engine natively supports JSON and JavaScript, there is no impedance mismatch between the type systems of an application and the database.

Optimistic concurrency control

Optimistic concurrency control allows you to prevent lost updates and deletes. Concurrent, conflicting operations are subjected to the regular pessimistic locking of the database engine hosted by the logical partition that owns the item. When two concurrent operations attempt to update the latest version of an item within a logical partition, one of them will win and the other will fail. However, if one or two operations attempting to concurrently update the same item had previously read an older value of the item, the database doesn’t know if the previously read value by either or both the conflicting operations was indeed the latest value of the item. Fortunately, this situation can be detected with the Optimistic Concurrency Control (OCC) before letting the two operations enter the transaction boundary inside the database engine. OCC protects your data from accidentally overwriting changes that were made by others. It also prevents others from accidentally overwriting your own changes.

The concurrent updates of an item are subjected to the OCC by Azure Cosmos DB’s communication protocol layer. Azure Cosmos database ensures that the client-side version of the item that you are updating (or deleting) is the same as the version of the item in the Azure Cosmos container. This guarantees that your writes are protected from being overwritten accidentally by the writes of others and vice versa. In a multi-user environment, the optimistic concurrency control protects you from accidentally deleting or updating wrong version of an item. As such, items are protected against the infamous “lost update” or “lost delete” problems.

Every item stored in an Azure Cosmos container has a system defined _etag property. The value of the _etag is automatically generated and updated by the server every time the item is updated. _etag can be used with the client supplied if-match request header to allow the server to decide whether an item can be conditionally updated. The value of the if-match header matches the value of the _etag at the server, the item is then updated. If the value of the if-match request header is no longer current, the server rejects the operation with an "HTTP 412 Precondition failure" response message. The client then can re-fetch the item to acquire the current version of the item on the server or override the version of item in the server with its own _etag value for the item. In addition, _etag can be used with the if-none-match header to determine whether a refetch of a resource is needed.

The item’s _etag value changes every time the item is updated. For replace item operations, if-match must be explicitly expressed as a part of the request options. For an example, see the sample code in GitHub. _etag values are implicitly checked for all written items touched by the stored procedure. If any conflict is detected, the stored procedure will roll back the transaction and throw an exception. With this method, either all or no writes within the stored procedure are applied atomically. This is a signal to the application to reapply updates and retry the original client request.

Next steps

Learn more about database transactions and optimistic concurrency control in the following articles: