At a glance
| Data model | Wide-column: tables with partition keys and clustering columns |
| Query language | CQL (Cassandra Query Language); no joins |
| Consistency model | Tunable per request; eventual by default, strong with quorum reads and writes |
| Replication | Masterless, configurable replication factor per data centre |
| Partitioning or sharding | Automatic, by hashing the partition key onto a token ring |
| Transactions | Lightweight transactions (compare-and-set) per partition; no multi-partition ACID transactions in 5.0 |
| Deployment and cloud availability | Self-managed on Linux, Docker or Kubernetes; managed by third parties such as Astra DB and Amazon Keyspaces (compatible API) |
| Licensing | Apache License 2.0, free |
| Maintained versions | 5.0.9, 4.1.12, 4.0.21 (7 August 2026) |
Cassandra is a strong fit when you need high write throughput, linear horizontal scale and multi-data-centre availability with no single primary, and you can model tables around known queries. It is fully open source with no licence fees, but it has no joins and only lightweight transactions, so it rewards careful data modelling and an operations team that knows it.
Overview
Apache Cassandra is a distributed database from the Apache Software Foundation. Data lives in tables defined by a partition key, which decides which nodes hold a row, and clustering columns, which order rows within a partition. Every node can serve reads and writes, replicas are placed per data centre, and the client chooses a consistency level for each request. You query with CQL, which looks like SQL but has no joins and expects filters on the primary key or an index.
Cassandra 5.0 (GA on 5 September 2024) added Storage Attached Indexes (SAI), trie memtables and SSTables, the Unified Compaction Strategy, a vector data type with similarity functions, dynamic data masking, JDK 17 support and more guardrails. The project keeps three lines maintained: 5.0 (5.0.9), 4.1 (4.1.12) and 4.0 (4.0.21), all released on 7 August 2026. For modelling sensor data, see Cassandra for telemetry.
Key features
- Masterless architecture: any node accepts reads and writes, with no single point of failure
- Tunable consistency per request (for example ONE, QUORUM, LOCAL_QUORUM, ALL)
- Multi-data-centre replication with per-data-centre replication factors
- CQL query language with partition and clustering keys, TTLs and lightweight transactions
- Storage Attached Indexes and a vector type with similarity functions (5.0)
- Unified Compaction Strategy and trie-based memtables and SSTables (5.0)
- Dynamic data masking and a CIDR authorizer (5.0)
Integrations and ecosystem
Pros and cons
Pros
- Permissive Apache 2.0 licence with no paid tier or usage limits
- No primary node, so writes continue through node and data-centre failures
- Scales out by adding nodes, with automatic data distribution
- Large ecosystem of drivers, tools and managed offerings from several vendors
Cons
- No joins and limited ad-hoc querying: tables must be designed per query pattern
- Only lightweight, single-partition transactions
- Operating clusters (repair, compaction, capacity planning) needs specialist skills
- No official managed service from the project; managed options come from third parties
Pricing
Apache Cassandra is free open source software under the Apache License 2.0, with no paid edition from the project. Costs come from your own infrastructure and operations, or from third-party managed services and support (for example DataStax Astra DB, now part of IBM, or Cassandra-compatible Amazon Keyspaces), each priced by its own vendor.
Pricing checked on the vendor's official page on 7 November 2026. Prices change; confirm before buying.
Who it is best for
Teams with large, write-heavy workloads such as telemetry, messaging or activity feeds that must stay available across data centres and can model data around known queries.
Typical use cases
- IoT and telemetry ingestion with time-bucketed partitions and TTLs
- Messaging, activity feeds and event logs at high write rates
- Multi-region user and device data that must survive a data-centre outage
- Vector similarity search next to operational data using the 5.0 vector type and SAI
Who should look elsewhere
- Applications that need joins, ad-hoc analytics or multi-row ACID transactions: see SQL vs NoSQL.
- Small datasets that a single relational or document database handles comfortably: see Cassandra vs MongoDB.
Verdict
Apache Cassandra remains the reference open source wide-column database for always-on, high-write workloads, and 5.0 adds useful indexing, compaction and vector features. It costs nothing to license but a lot to run well, so choose it when availability and scale outweigh query flexibility, or use a managed service.
Frequently asked questions
What is the latest version of Apache Cassandra?
The downloads page lists 5.0.9 (7 August 2026) as the latest GA release, with 4.1.12 and 4.0.21 also maintained. 4.0 is maintained until the 6.0 release.
What is new in Cassandra 5.0?
Storage Attached Indexes, trie memtables and SSTables, the Unified Compaction Strategy, a vector data type and similarity functions, dynamic data masking, a CIDR authorizer, JDK 17 support, more guardrails and new math and collection functions.
Does Cassandra support transactions?
Cassandra 5.0 offers lightweight transactions, a compare-and-set mechanism within a single partition. It does not offer multi-partition ACID transactions, so design writes so that each operation touches one partition.
Is there a managed Cassandra service?
The Apache project does not run one. Third parties do: DataStax Astra DB (DataStax was acquired by IBM) and Amazon Keyspaces, which is CQL-compatible but not Apache Cassandra itself, among others. Check each vendor for compatibility and pricing.
Sources
- Apache Cassandra downloads
- Cassandra 5.0: new features
- Apache Cassandra 5.0 announcement
- Apache Cassandra 5.0 documentation
- apache/cassandra repository and licence (GitHub)
Checked November 2026.
Listed in: NoSQL & Distributed Databases.
How we research reviews and what star ratings mean: our editorial method.