Architecture in brief

Essential information for understanding and using Cassandra.

Cassandra is designed to handle big data workloads across multiple nodes with no single point of failure. Its architecture is based on the understanding that system and hardware failures can and do occur. Cassandra addresses the problem of failures by employing a peer-to-peer distributed system across homogeneous nodes where data is distributed among all nodes in the cluster. Each node exchanges information across the cluster every second. A sequentially written commit log on each node captures write activity to ensure data durability. Data is then indexed and written to an in-memory structure, called a memtable, which resembles a write-back cache. Once the memory structure is full, the data is written to disk in an SSTable data file. All writes are automatically partitioned and replicated throughout the cluster. Using a process called compaction Cassandra periodically consolidates SSTables, discarding obsolete data and tombstone (an indicator that data was deleted).

Cassandra is a row-oriented database. Cassandra's architecture allows any authorized user to connect to any node in any datacenter and access data using the CQL language. For ease of use, CQL uses a similar syntax to SQL. From the CQL perspective the database consists of tables. Typically, a cluster has one keyspace per application. Developers can access CQL through cqlsh as well as via drivers for application languages.

Client read or write requests can be sent to any node in the cluster. When a client connects to a node with a request, that node serves as the coordinator for that particular client operation. The coordinator acts as a proxy between the client application and the nodes that own the data being requested. The coordinator determines which nodes in the ring should get the request based on how the cluster is configured. For more information, see Client requests.

Key structures

  • Node

    Where you store your data. It is the basic infrastructure component of Cassandra.

  • datacenter

    A collection of related nodes. A datacenter can be a physical data center or virtual datacenter. Different workloads should use separate datacenters, either physical or virtual. Replication is set by datacenter. Using separate data centers prevents Cassandra transactions from being impacted by other workloads and keeps requests close to each other for lower latency. Depending on the replication factor, data can be written to multiple datacenters. However, datacenters should never span physical locations.

  • Cluster

    A cluster contains one or more datacenters. It can span physical locations.

  • Commit log

    All data is written first to the commit log for durability. After all its data has been flushed to SSTables, it can be archived, deleted, or recycled.

  • Table

    A collection of ordered columns fetched by row. A row consists of columns and have a primary key. The first part of the key is a column name.

  • SSTable

    A sorted string table (SSTable) is an immutable data file to which Cassandra writes memtables periodically. SSTables are append only and stored on disk sequentially and maintained for each Cassandra table.

Key components for configuring Cassandra 

  • Gossip

    A peer-to-peer communication protocol to discover and share location and state information about the other nodes in a Cassandra cluster. Gossip information is also persisted locally by each node to use immediately when a node restarts.

  • Partitioner

    A partitioner determines which node will receive the first replica of a piece of data, and how to distribute other replicas across other nodes in the cluster. Each row of data is uniquely identified by a primary key, which may be the same as its partition key, but which may also include other clustering columns. A partitioner is a hash function that derives a token from the primary key of a row. The partitioner uses the token value to determine which nodes in the cluster receive the replicas of that row. The Murmur3Partitioner is the default partitioning strategy for new Cassandra clusters and the right choice for new clusters in almost all cases.

    You must set the partitioner and assign the node a num_tokens value for each node. The number of tokens you assign depends on the hardware capabilities of the system. If not using virtual nodes (vnodes), use the initial_token setting instead.

  • Replication factor

    The total number of replicas across the cluster. A replication factor of 1 means that there is only one copy of each row on one node. A replication factor of 2 means two copies of each row, where each copy is on a different node. All replicas are equally important; there is no primary or master replica. You define the replication factor for each datacenter. Generally you should set the replication strategy greater than one, but no more than the number of nodes in the cluster.

  • Replica placement strategy

    Cassandra stores copies (replicas) of data on multiple nodes to ensure reliability and fault tolerance. A replication strategy determines which nodes to place replicas on. The first replica of data is simply the first copy; it is not unique in any sense. The NetworkTopologyStrategy is highly recommended for most deployments because it is much easier to expand to multiple datacenters when required by future expansion.

    When creating a keyspace, you must define the replica placement strategy and the number of replicas you want.

  • Snitch

    A snitch defines groups of machines into datacenters and racks (the topology) that the replication strategy uses to place replicas.

    You must configure a snitch when you create a cluster. All snitches use a dynamic snitch layer, which monitors performance and chooses the best replica for reading. It is enabled by default and recommended for use in most deployments. Configure dynamic snitch thresholds for each node in the cassandra.yaml configuration file.

    The default SimpleSnitch does not recognize datacenter or rack information. Use it for single-datacenter deployments or single-zone in public clouds. The GossipingPropertyFileSnitch is recommended for production. It defines a node's datacenter and rack and uses gossip for propagating this information to other nodes.

  • The cassandra.yaml configuration file

    The main configuration file for setting the initialization properties for a cluster, caching parameters for tables, properties for tuning and resource utilization, timeout settings, client connections, backups, and security.

    By default, a node is configured to store the data it manages in a directory set in the cassandra.yaml file.
    • Package installations: /var/lib/cassandra
    • Tarball installations: install_location/data/data

    In a production cluster deployment, you can change the commitlog-directory to a different disk drive from the data_file_directories.

  • System keyspace table properties

    You set storage configuration attributes on a per-keyspace or per-table basis programmatically or using a client application, such as CQL.