The name Redis means REmote DIctionary Server. The Redis project began when Salvatore Sanfilippo, nicknamed antirez, the original developer of Redis, was trying to improve the scalability of his Italian startup, developing a real-time web log analyzer. After encountering significant problems in scaling some types of workloads using traditional database systems, Sanfilippo began to prototype a first proof of concept version of Redis in Tcl. Later Sanfilippo translated that prototype to the C language and implemented the first data type, the list. After a few weeks of using the project internally with success, Sanfilippo decided to open source it, announcing the project on Hacker News. The project began to get traction, more so among the Ruby community, with GitHub and Instagram being among the first companies adopting it. Sanfilippo was hired by VMware in March, 2010. In May, 2013, Redis was sponsored by Pivotal Software. In June 2015, development became sponsored by Redis Labs. In October 2018 Redis 5.0 was released, introducing Redis Stream - a new data structure that allows storage of multiple fields and string values with an automatic, time-based sequence at a single key.
Differences with other database systems
Redis popularized the idea of a system that can be considered at the same time a store and a cache, using a design where data is always modified and read from the main computer memory, but also stored on disk in a format that is unsuitable for random access of data, but only to reconstruct the data back in memory once the system restarts. At the same time, Redis provides a data model that is very unusual compared to a relational database management system. User commands do not describe a query to be executed by the database engine but rather specific operations that are performed on given abstract data types. Hence, data must be stored in a way which is suitable later for fast retrieval, without help from the database system in form of secondary indexes, aggregations or other common features of traditional RDBMS. The Redis implementation makes heavy use of the fork system call, to duplicate the process holding the data, so that the parent process continues to serve clients, while the child process creates a copy of the data on disk.
Popularity
According to monthly DB-Engines rankings, Redis is often the most popular key-value database. Redis has also been ranked the #4 NoSQL database in user satisfaction and market presence based on user reviews, the most popular NoSQL database in containers, and the #4 Data store of 2019 by ranking website stackshare.io. It was voted most loved database in the Stack Overflow Developer Survey in 2017, 2018, 2019, and 2020.
Redis maps keys to types of values. An important difference between Redis and other structured storage systems is that Redis supports not only strings, but also abstract data types:
HyperLogLogs used for approximated set cardinality size estimation, available since Redis 2.8.9 in April 2014.
Stream of entries with consumer groups, allows you to store multiple fields and string values with an automatic, time-based sequence at a single key, available since Redis 5.0 in October 2018
Geospatial data through the implementation of the geohash technique, available since Redis 3.2.
The type of a value determines what operations are available for the value. Redis supports high-level, atomic, server-side operations like intersection, union, and difference between sets and sorting of lists, sets and sorted sets. More data types are supported based on Redis Modules API. The Redis module RedisJSON implements ECMA-404 as a native data type.
Redis typically holds the whole dataset in memory. Versions up to 2.4 could be configured to use what they refer to as virtual memory in which some of the dataset is stored on disk, but this feature is deprecated. Persistence in Redis can be achieved through two different methods. First by snapshotting, where the dataset is asynchronously transferred from memory to disk at regular intervals as a binary dump, using the Redis RDB Dump File Format. Alternatively by journaling, where a record of each operation that modifies the dataset is added to an append-only file in a background process. Redis can rewrite the append-only file in the background to avoid an indefinite growth of the journal. Journaling was introduced in version 1.1 and is generally considered the safer approach. By default, Redis writes data to a file system at least every 2 seconds, with more or less robust options available if needed. In the case of a complete system failure on default settings, only a few seconds of data would be lost.
Replication
Redis supports master-replica replication. Data from any Redis server can replicate to any number of replicas. A replica may be a master to another replica. This allows Redis to implement a single-rooted replication tree. Redis replicas can be configured to accept writes, permitting intentional and unintentional inconsistency between instances. The publish/subscribe feature is fully implemented, so a client of a replica may subscribe to a channel and receive a full feed of messages published to the master, anywhere up the replication tree. Replication is useful for read scalability or data redundancy.
Performance
When the durability of data is not needed, the in-memory nature of Redis allows it to perform well compared to database systems that write every change to disk before considering a transaction committed. Redis operates as a single process and is single-threaded or double-threaded when it rewrites the AOF. Thus, a single Redis instance cannot use parallel execution of tasks such as stored procedures.
Clustering
Redis introduced clustering in April 2015 with the release of version 3.0. The cluster specification implements a subset of Redis commands: all single-key commands are available, multi-key operations are restricted to keys belonging to the same node, and commands related to database selection operations are unavailable. A Redis cluster can scale up to 1,000 nodes, achieve "acceptable" write safety and to continue operations when some nodes fail.
Use cases
Due to the nature of the database design, typical use cases are session caching, full page cache, message queue applications, leaderboards and counting among others. Large companies such as Twitter are using Redis, Amazon Web Services is offering Redis in its portfolio, Microsoft is offering the Redis Cache in Azure, and Alibaba is offering ApsaraDB for Redis in Alibaba Cloud.
