Apache Cassandra

Apache Cassandra is an open source NoSQL database that stores data across many machines at once. It is a project of the Apache Software Foundation, released under the Apache License 2.0, and it has no single owner who sells it. Every node in a cluster does the same job, so there is no primary server to lose, and the project says you can double capacity by doubling the number of nodes.

Data is spread by a partition key, copied to as many nodes as the replication factor says, and can be copied between data centres too. Applications talk to it in Cassandra Query Language (CQL), which looks like SQL but has no joins. The vendor's pages describe it as suited to applications that cannot afford to lose data, even when a whole data centre goes down. The current general availability release on the download page is 5.0, with 5.0.9 dated 7 August 2026.

• Masterless ring. All nodes are identical and talk to each other by gossip, so any node can take a request and act as coordinator. The project docs say this removes single points of failure and network bottlenecks.
• Per-query consistency. You choose a consistency level for each read or write, such as QUORUM, which means a majority of replicas must acknowledge. The docs call Cassandra an AP database by default and say tuning this takes some skill.
• Multi-data-centre replication. Replication across data centres is built in. The docs give the example of writing on the US west coast and reading the same data from nodes in Asia and Europe, and note that failed nodes can be replaced without downtime.
• Storage Attached Indexes. Version 5.0 adds SAI, which lets you filter on columns that are not part of the primary key. The release notes list it next to trie memtables, trie SSTables and the Unified Compaction Strategy.
• Vector search. Cassandra 5.0 has a vector data type and similarity functions. The documentation describes vector search as a way to find relevant content in large document collections, aimed at AI applications, on any CQL table.
• Operations tooling in the box. The tarball ships cqlsh, nodetool, cassandra-stress and SSTable tools. Audit logging records DML, DDL and DCL activity, and fqltool captures and replays production workloads for analysis.
• Dynamic Data Masking. Also new in 5.0, masking functions can hide column values from users without the right permission. A CIDR authorizer is listed in the same release notes for limiting access by network range.

• A platform team writing event data from several regions. The project's own case-study quotes come from companies that need steady write throughput. Backblaze says it needed to keep scaling writes and chose Cassandra for that. A team in that position can add nodes as write volume grows.
• An operations group that must survive a data-centre loss. The homepage says Cassandra can withstand an entire data centre outage with no data loss. A group running two or three sites can set replication per data centre and test failover with nodetool and the stress tool.
• A company already running Kubernetes. The ecosystem page lists community Docker images, an Ansible collection and a Prometheus exporter. A platform engineer can run Cassandra with the container and configuration tools the team already uses, and feed node metrics into existing dashboards.

The project's quickstart uses Docker. You pull the cassandra image, start a container on a Docker network, save a CQL file that creates a keyspace and a shopping_cart table, load it with a cqlsh container, and then read and write rows interactively. The page warns that the server needs a few seconds to finish starting.

Tarball, Debian and RPM installs are also documented, and the download page gives apt and yum repository steps for the 5.0 series. After install you check the node with nodetool status. The docs say little is gained from a single node in production, and that it is helpful mainly for learning. Client drivers exist for Java, Python, Ruby, C#, Node.js, PHP, C++, Scala, Go, Rust and others.

Cassandra spreads data over nodes that all behave alike, and you double capacity by doubling nodes. The price is in the query model: CQL runs no joins or sub-queries, and each select reads a single table, so you design tables around your queries.

The project says there is little point in running one node. The Apache Software Foundation, a US non-profit, publishes the code and sells nothing.

The database management systems list places Apache Cassandra 7 of 22: it follows Redis and comes before Couchbase. No numeric score is attached; established products follow the challengers in their ranked order.

Apache Cassandra is based in the United States; 15 of the 22 products here share its region, North America.

The project states you can add or remove nodes without downtime, so growth is a hardware and staffing decision.

Pricing as recorded: Free to run under Apache Licence 2.0; no vendor sells it.

The software costs nothing to download or run: the project footer states it is under the Apache License 2.0, and the download page lists no paid tier. What you pay for is the hardware or cloud instances and the staff who run the cluster. The vendor does not publish operating cost figures. The ecosystem page lists third-party managed services, including Azure Managed Instance for Apache Cassandra (instance-based pricing, per that page), DataStax Astra (described as having a free tier) and Instaclustr. Their prices are set by those companies, not by the Cassandra project, and were not checked here.

We took this from cassandra.apache.org, the vendor’s own site, when we reviewed Best Database Management Systems in 2026; that review is dated 7 November 2026.

Cassandra is software you host yourself, so the region and data location are whatever you choose. The docs say TLS encryption, client authentication and authorization are all off by default, and that a default install presents a large attack surface.

Encryption can be set separately for client-to-node and node-to-node traffic. FIPS settings are applied at the JVM level. The project publishes no compliance certificates; it points to a security page on apache.org for vulnerability reporting.

Integrations the vendor names (12):

• Docker community images
• Ansible Cassandra Collection
• AxonOps
• Beekeeper Studio
• Cassandra Prometheus Exporter
• Cassandra Lucene Index
• Apache Ignite
• Elassandra
• DataStax Astra
• Instaclustr
• Azure Managed Instance for Apache Cassandra
• Cassandra Migration

4 products in this guide come closest to Apache Cassandra on tags, pricing and region.

• PostgreSQL (#1 of 22): Aimed at: Teams that want a relational engine with no vendor able to reprice it.
• SQLite (#3 of 22): Aimed at: Applications, devices and small sites that need storage inside the app. (United States)
• Redis (#6 of 22): Not for: buyers who need cheap storage for very large vector collections. Memory-first data store used for caching, queues and sessions, now tri-licensed. (United States)
• MySQL (#9 of 22): Oracle-owned relational server, free under GPLv2, paid editions on request. Suits: Teams running widely supported web stacks who accept Oracle as owner.

Read from the vendor’s own pages on 5 November 2026.

• cassandra.apache.org/_/index.html: Project description, masterless claim, audit logging, case-study quotes, Apache License 2.0 footer
• cassandra.apache.org/_/cassandra-basics.html: Nodes, gossip, partitions, replication factor, consistency levels, scaling by adding nodes
• cassandra.apache.org/doc/latest/cassandra/getting-started/index.html: Install options, client driver list, production recommendations link
• cassandra.apache.org/doc/latest/cassandra/getting-started/cassandra-quickstart.html: Docker quickstart steps with cqlsh
• cassandra.apache.org/_/download.html: Release lines and dates, Debian and RPM install steps
• cassandra.apache.org/doc/latest/cassandra/new/index.html: Cassandra 5.0 feature list (SAI, vector type, DDM, CIDR authorizer, UCS)
• cassandra.apache.org/doc/latest/cassandra/vector-search/overview.html: Vector search description
• cassandra.apache.org/doc/latest/cassandra/managing/operating/security.html: Security features off by default, TLS options, FIPS at JVM level
• cassandra.apache.org/doc/latest/cassandra/getting-started/production.html: Token count and read ahead tuning guidance
• cassandra.apache.org/_/ecosystem.html: Third-party tools and managed services

We test for four things rather than for features: setup time, price in practice, exit and audience. Here that gives Apache Cassandra the #7 position in a list of 22. Method and commercial relationships are in our editorial process.