Amazon Aurora is a relational database offered as a service in Amazon's AWS. Based on the open source version of MySQL, it is a commercial database that claims to be compatible with MySQL and PostgreSQL while providing superior throughput. Being provided as a cloud service, Aurora promises high availability by using distributed replications of backend storage. The system is being actively maintained and updated by Amazon.
Aurora was announced on Nov. 12, 2014 in Amazon's re:Invent conference in Las Vegas. It was officially released and ready to use as a service in AWS on July 27, 2015 by being added into the Amazon Relational Database Service. A major patch of Aurora was added in October 24, 2017, where Aurora was extended with PostgreSQL compatibility.
As claimed in the Aurora paper, the system doesn't need checkpointing to do fast recovery. The system has the following features to make that possible: + Using a distributed replication of backend storage. This makes it possible to use quorum based read for recovery. + The database engine only propagates logs to the backend storage, where the backend storage keeps processing these logs asynchronously to bring the database to its latest state. These designs make fast logging recovery very easy in Aurora, since the system can easily be brought back to a consistent state that is not far from the time of crushing, by utilizing read quorums among all replicated backend storage.
Aurora decouples the storage engine from its database engine, and the concurrency control protocol is entirely decided by the database engine it used. In the paper that introduced Aurora, the concurrency control model was stated to be exactly the same as the database engine it inherited from. So Aurora has the same concurrency control protocol, MVCC, as MySQL/InnoDB does. InnoDB's MVCC protocol stores a separate data structure for "rollback segments", which are actually undo logs. In the situation of a consistent read (for isolation levels beyond read committed), the logs will be applied in place to reconstruct the requested earlier versions of a row.
Aurora is stated to be a relational database engine. That can also be inferred from its full inheritance of MySQL/InnoDB's database engine and storage layout.
Aurora supports foreign keys, just as MySQL/InnoDB does. Best practice guide on Aurora's documentation provided some insights on how to properly use foreign keys, please refer to the citation for more details.
Aurora uses the same indexes as MySQL/InnoDB. In MySQL, both b-tree and hash indexes are used. The default index choice for MySQL is b-tree unless explicitly specified. Aurora also supports spatial indexes, and its implementation utilizes a b-tree.
In the paper that introduced Aurora, the authors stated that Aurora has exactly the same isolation levels as MySQL. The supported isolation levels includes the standard ANSI levels and Snapshot Isolation.
Aurora has different join algorithm compared to MySQL. It supports hash join in addition to the already-existing nested loop join in MySQL. When the hash join option is enabled, Aurora's optimizer will automatically choose a join method as it evaluates the query plan. However, there are several restrictions for hash join in Aurora. To be more specific, left-right outer joins, semijoins such as subqueries and multiple-table updates or deletes are not supported.
Aurora's logging design is logical logging. The system separates the database engine from the backend storage, where the database engine propagates logs continuously to the backend storage. Such logs are then asynchronously processed by the distributed storage servers to bring the database to its latest state.
Similar to MySQL, Aurora doesn't support query compilation.
Aurora uses the same query execution engine as MySQL.
Aurora supports the standard SQL query interface. It inherits the SQL compatibility from MySQL, as well as the extensions MySQL made to the SQL standard.
Aurora is a disk-oriented database.
Same as MySQL/InnoDB, Aurora is a row-storage DBMS. The tuples are stored row-by-row in the distributed storage servers.
Same as MySQL/InnoDB, Aurora supports stored procedures.
Aurora decouples its database engine from the storage backend. The database engine is a modified version with MySQL/InnoDB, where the storage backend consists of distributed replicas that span across different availability zones in AWS. The database engine will propagates logs to the backend storage, and the backend storage utilizes a quorum based synchronization scheme to ensure the consistency of the database.
Same as MySQL, Aurora supports materialized views.