Virtual node (vnode) configuration
Virtual nodes simplify many tasks in DataStax Enterprise (DSE), such as eliminating the need to determine the partition range (calculate and assign tokens), rebalancing the cluster when adding or removing nodes, and replacing dead nodes. For a complete description of virtual nodes and how they work, see Virtual nodes.
DSE requires the same token architecture on all nodes in a datacenter. The nodes must all be vnode-enabled or single-token architecture. Across the entire cluster, datacenter architecture can vary. For example, a single cluster with:
-
A transaction-only datacenter running OLTP.
-
A single-token architecture search datacenter (no vnodes).
-
An analytics datacenter with vnodes.
Guidelines for using virtual nodes
-
DSE requires the same token architecture on all nodes in a datacenter.
The nodes must all be vnode-enabled or single-token architecture. Across the entire cluster, datacenter architecture can vary.
For example, a single cluster with:
-
A transaction-only datacenter running OLTP.
-
A single-token architecture search datacenter (no vnodes).
-
An analytics datacenter with vnodes.
-
-
DataStax recommends using 8 vnodes (tokens).
DataStax recommends not using vnodes with DSE Search. However, if you decide to use vnodes with DSE Search, do not use more than 8 vnodes and ensure that
allocate_tokens_for_local_replication_factor
option incassandra.yaml
is correctly configured for your environment.Using 8 vnodes distributes the workload between systems with a ~10% variance and has minimal impact on performance.
-
Ensure correct vnode configuration with cassandra.yaml settings:
-
When adding a vnode to an existing cluster or setting up nodes in a new datacenter, set the target replication factor (RF) of keyspaces in the datacenter with the
allocate_tokens_for_local_replication_factor
option. -
The allocation algorithm distributes the token ranges proportionately using the
num_tokens
settings.All systems in the datacenter should have the same
num_token
settings unless the systems performance varies between systems. To distribute more of the workload to the higher performance hardware, increase the number of tokens for those systems.The allocation algorithm efficiently balances the workload using fewer tokens; when systems are added to a datacenter, the algorithm maintains the balance. Using a higher number of tokens more evenly distributes the workload, but also significantly increases token management overhead.
Set the number of vnode tokens based on the workload distribution requirements of the datacenter:
Allocation algorithm workload distribution variance Replication factor 4 vnode (tokens) 8 vnode (tokens) 64 vnode (tokens) 128 vnode (tokens) 2
~17.5%
~12.5%
~3%
~1%
3
~14%
~10%
~2%
~1%
5
~11%
~7%
~1%
~1%
-
-
Add nodes to the cluster one at a time.
When adding multiple nodes to the cluster using the allocation algorithm, ensure that nodes are added one at a time. If nodes are added concurrently, the algorithm assigns the same tokens to different nodes.
Enabling vnodes
In the cassandra.yaml
file:
-
Uncomment
num_tokens
and set the required number of tokens. -
(Recommended) To use the allocation algorithm uncomment
allocate_tokens_for_local_replication_factor
and set it to the target replication factor for the keyspaces in the datacenter. If the replication varies, alternate between the replication factor (RF) settings. -
Comment out the
initial_token
or leave unset.
To upgrade existing clusters to vnodes, see Enabling virtual nodes on an existing production cluster.
Disabling vnodes
If you do not use vnodes, you must make sure that each node is responsible for roughly an equal amount of data. To ensure that each node is responsible for an equal amount of data, assign each node an initial-token value and calculate the tokens for each datacenter as described in Generating tokens. |
-
In the cassandra.yaml file:
-
Comment out the
num_tokens
andallocate_tokens_for_local_replication_factor
. -
Uncomment the
initial_token
and set it to 1 or to the value of a generated token for a multi-node cluster.
-