Understand how requests are coordinated
Cassandra has a masterless 「ring」 architecture that is elegant, easy to set up, and easy to maintain.
What is a cluster?
- Node: one Cassandra instance
- Rack: a logical set of nodes
- Data Center: a logical set of racks
- Cluster: the full set of nodes which map to a single complete token ring
What is a coordinator?
The node chosen by the client to receive a particular read or write request to its cluster
- Any node can coordinate any request
- Each client request may be coordinated by a a different node
- The coordinator manages the Replication Factor (RF)
- Replication factor (RF) – onto how many nodes should a write be copied?
The coordinator also applies the Consistency Level (CL)
Consistency level (CL) – how many nodes must acknowledge a read or write request
What is the partitioner?
A system on each node which hashes tokens from designated values in rows being added
How does a partitioner work?
A node's partitioner hashes a token from the partition key value of a write request
What partitioners does Cassandra offer?
Cassandra offers three partitioners
- Murmur3 Partitioner (default): uniform distribution based on Murmur3 hash
- Random Partitioner: uniform distribution based on MD5 hash
- ByteOrdered Partitioner (legacy only): lexical distribution based on key bytes
What are virtual nodes?
There's one token per node, and thusly a node owns exactly one contiguous range in the ringspace.Vnodes change this paradigm from one token or range per node, to many per node. Within a cluster these can be randomly selected and be non-contiguous, giving us many smaller ranges that belong to each node.
How are virtual nodes helpful?
Having many smaller token ranges (nodes) on each physical node allows for a more even distribution of data. This becomes evident when you add a physical node to the cluster, in that rebalancing (manually reassigning token ranges) is no longer necessary.
- token ranges are distributed, so machines bootstrap faster
- impact of virtual node failure is spread across entire cluster
- token range assignment automated