Apache Cassandra Fundamentals

Apache Cassandra is a distributed NoSQL database designed for high write throughput and availability across multiple data centers. It uses a peer-to-peer architecture with tunable consistency, making it suited for workloads that prioritize uptime and horizontal scalability over strong transactional guarantees.

itDatabases and data storage

Don't Panic — Apache Cassandra Fundamentals

Apache Cassandra is a database that declines to appoint one machine as the important one. Data lives in partitions across a cluster, with replicas holding copies, so a request can still find a route when a node or site has a difficult day. This is not magic. It is a rather stern agreement between the application, the data model, and the operators.

The agreement begins with the partition key. It turns a query into a destination: Cassandra hashes that key to a token, then sends the work to the replicas for that token range. A table is therefore not a neutral container for business entities. It is a promise about the questions an application will ask. Put device events for one device and day together, for example, and the useful time-range query has somewhere definite to go. Put every event for a busy device together forever, and the partition becomes a small, private weather system.

The node a client contacts is the coordinator for that one request. It asks the appropriate replicas for responses and applies the selected consistency level. Replication factor says how many copies are stored. Consistency level says how many responses are required now. They are related, as umbrellas and rain are related, but neither is the other.

Inside each replica, a write reaches a commit log and a memtable before it becomes an immutable SSTable on disk. That makes writes friendly to the foreground path. It also creates later work: reads may reconcile several structures, and compaction must merge files and process old values. A deletion is a tombstone, not an instant disappearance, because replicas need time to learn that an old value is gone.

That is why repair matters. Hints can help after a missed write, but repair compares replica data and streams differences so the copies converge. Adding nodes can spread token ranges. It cannot rescue an unbounded partition key, cancel compaction, or turn a relational join into a Cassandra query. The database is polite about these limits only until production traffic arrives.

Read the Intro when you need the whole request and storage path. Use Slides for the chain from query to replicas to SSTables. Keep the Cheatsheet nearby when distinguishing replication from consistency, or tombstones from repair. The Reference tab is the route from this mental map into CQL, modeling, and production operations.

Don't Panic — Apache Cassandra Fundamentals

Relevant careers