Snitches are used in the following ways:
To determine to which datacenters and racks the Scylla nodes belong to
To inform Scylla about the network topology so that requests are routed efficiently
To allow Scylla to distribute replicas by grouping machines into data centers and racks.
Note, that if you do not choose a Snitch when creating a Scylla cluster, the SimpleSnitch is selected by default.
Scylla supports the following snitches:
Note
For production clusters, it is strongly recommended to use GossipingPropertyFileSnitch or Ec2MultiRegionSnitch.
Other snitches are limited and will make it harder for you to add a Data Center (DC) later.
Warning
Do not disable access to instance metadata if you’re using Ec2Snitch or Ec2MultiRegionSnitch. With access to medatata disabled, the information about datacenter names and racks may be missing or incorrect, or the instance may fail to boot.
Use the SimpleSnitch when working with single cluster deployments and all the nodes are under the same datacenter The SimpleSnitch binds all the nodes to the same Rack and datacenter and is recommended to be used only in single datacenter deployments.
RackInferringSnitch binds nodes to DCs and racks according to their broadcast IP addresses.
For Example:
If a node has a Broadcast IP 192.168.100.200; then it would belong to a DC ‘168’ and Rack ‘100’.
Use the GossipingPropertyFileSnitch when working with multi-cluster deployments where the nodes are in various datacenters.
It is recommended to use the GossipingPropertyFileSnitch in production installations.
This snitch allows Scylla to explicitly define which DC and rack a specific node belongs to.
In addition, it reads its configuration from the cassandra-rackdc.properties file, which is located in the /etc/scylla/ directory.
For Example:
prefer_local=true
dc=my_data_center
rack=my_rack
Setting prefer_local to true instructs Scylla to use an internal IP address for interactions with nodes in the same DC.
An example use case is when your host uses different addresses for LAN and WAN sessions. You want your cluster to be accessible by clients outside the Scylla nodes’ LAN while still allowing Scylla nodes to communicate over internal LAN keeping latency low. In AWS, this is similar to a VM’s “Public” and “Private” addresses. To set an internal and external address, set a LAN address as a listen_address and use a WAN address as a broadcast_address.
If you set prefer_local: true nodes in the same DC would use their LAN addresses to communicate with each other and WAN addresses to access nodes in different DCs and communicate with Clients.
Use the Ec2Snitch when working on EC2 with a single cluster deployments where all nodes are located in the same region. This basic snitch reads its configuration from Amazon’s EC2 registry services. When using EC2, the region name is treated as the datacenter name and availability zones are treated as racks within a datacenter. If the setup includes a single datacenter, there is no need to specify any parameters. As private IPs are used, this snitch does not work well across multiple regions. It should also be noted that according to this snitch, a DC is a region and if a region is down, the entire cluster will be down.
If you are working with multiple datacenters, specify the DC and set the parameter dc_suffix=<DCNAME> in the cassandra-rackdc.properties file, which is located in the /etc/scylla/ directory.
For Example, suppose you had created a 5 node cluster and added the following configuration settings to each node’s /etc/scylla/cassandra-rackdc.properties file as shown:
Node number |
Parameter to add to the specific node’s /etc/scylla/cassandra-rackdc.properties |
|---|---|
1 |
dc_suffix=_dc1-europe |
2 |
dc_suffix=_dc1-europe |
3 |
dc_suffix=_dc2-asia |
4 |
dc_suffix=_dc2-asia |
5 |
dc_suffix=_dc3-australia |
This action adds a suffix to the name of each of the datacenters for the region.
Running the nodetool status command shows all three datacenters:
Datacenter: us-east_dc1-europe
==============================
Status=Up/Down
|/ State=Normal/Leaving/Joining/Moving
-- Address Load Tokens Owns Host ID Rack
UN 172.20.0.4 111.23 KB 256 ? eaabc5db-61ff-419b-b1a7-f70af23edb1b Rack1
UN 172.20.0.5 127.09 KB 256 ? bace1b4e-67c6-4bdb-8eba-398162b7b56e Rack1
Datacenter: us-east_dc2-asia
============================
Status=Up/Down
|/ State=Normal/Leaving/Joining/Moving
-- Address Load Tokens Owns Host ID Rack
UN 172.20.0.6 110.59 KB 256 ? bda5fb11-9369-48fb-91be-82c8d821f758 Rack1
UN 172.20.0.3 111.26 KB 256 ? b9ea3516-5e1e-4ffb-abff-c6a6701cb41b Rack1
Datacenter: us-east_dc3-australia
=================================
Status=Up/Down
|/ State=Normal/Leaving/Joining/Moving
-- Address Load Tokens Owns Host ID Rack
UN 172.20.0.7 111.23 KB 256 ? eaabc5db-61ff-419b-b1a7-f70af23edb1b Rack1
Note
The datacenter naming convention in this example is based on location.
You can use other conventions, such as DC1, DC2 or 100, 200, or analytics, search, Scylla, and more.
Providing a separator such as a dash keeps the name of the DC readable as the dc_suffix property adds the suffix to the DC name.
Note
Ec2Snitch and Ec2MultiRegionSnitch will define DC/RACK differently for AWS Availability Zones (AZs) that end with 1x compared to other AZs:
For the former class of AZs, e.g. us-east-1d, the Snitch will set DC='us-east', RACK='1d'
For the latter class of AZs, e.g. us-east-4c, the Snitch will set DC='us-east-4', RACK='4c'
Use the Ec2MultiRegionSnitch when working on EC2 and using multi-cluster deployments where the nodes are in various regions.
This snitch works like the Ec2Snitch, but in addition, it sets the node’s broadcast_address and broadcast_rpc_address to the node’s public IP address.
This setting allows nodes from other zones to communicate with the node regardless of what is configured in the node’s scylla.yaml configuration file for broadcast_address and broadcast_rpc_address parameters.
Ec2MultiRegionSnitch also unconditionally imposes the “prefer local” policy on a node (similar to GossipingPropertyFileSnitch when prefer_local is set to true).
In EC2, the region name is treated as the datacenter name and availability zones are treated as racks within a datacenter.
To change the DC and rack names, do the following:
Edit the cassandra-rackdc.properties file with the preferred datacenter name. The file can be found under /etc/scylla/
The dc_suffix defines a suffix added to the datacenter name as described below.
For Example:
Node - region DC='us-west' and Rack Rack='1' will be us-west-1 dc_suffix= scylla_node_west
Node - region DC='us-east' and Rack Rack='2' will be us-east-2 dc_suffix= scylla_node_east
us-west-1_scylla_node_west
us-east-2_scylla_node_east
Use the GoogleCloudSnitch for deploying Scylla on the Google Cloud Engine (GCE) platform across one or more regions. The region is treated as a datacenter, and the availability zones are treated as racks within the datacenter. All communication occurs over private IP addresses within the same logical network.
To use the GoogleCloudSnitch, add the snitch name to the scylla.yaml file, which is located in the /etc/scylla/ directory for all nodes in the cluster.
You can add a suffix to the data center name as an additional identifier. This suffix is appended to the Zone name without adding any spaces. To add this suffix edit the cassandra-rackdc.properties file, which can be found under /etc/scylla/ and set the dc_suffix with an appropriate text string. It may help to add an underscore or dash in front. Keep in mind that this property file is used for all Scylla snitches. When using GoogleCloudSnitch, all other properties are ignored.
Example
You have two datacenters running on GCE. One is for the office in Miami and is in region us-east1, zone us-east-1-b. The other office is in Portland and is in region us-west1,, zone us-west-1-b.
It’s important to note that:
DC1 is us-east1 with rack name b
DC2 is us-west1 with rack b
Racks are important for distributing replicas, but not for datacenter naming as this Snitch can work across multiple regions without additional configuration.
After creating the instances on GCE, edit the scylla.yaml file to select the GoogleCloudSnitch.
endpoint_snitch: GoogleCloudSnitch
As you want to set the data center suffix for the nodes in each datacenter, you open each node’s properties file in the cassandra-rackdc.properties The file can be found under /etc/scylla/.
You set the following parameters for Miami:
# node 1 - 192.0.2.2 (you use the same properties for node #2 (192.0.2.3) and #3 (192.0.2.4))
dc_suffix=_scylla_node_Miami
and for Portland:
# node 4 192.0.2.5
dc_suffix=_scylla_node_Portland
Start the cluster, one node at a time, and then run nodetool status to check connectivity.
nodetool status
Datacenter: us-east1_scylla_node_Miami
======================================
Status=Up/Down
|/ State=Normal/Leaving/Joining/Moving
-- Address Load Tokens Owns Host ID Rack
UN 192.0.2.2 1.27 MB 256 ? 5b1d864f-a026-4076-bb19-3e7dd693abf1 b
UN 192.0.2.3 954.89 KB 256 ? 783a815e-6e9d-4ab5-a092-bbf15fd76a9f b
UN 192.0.2.4 1.02 MB 256 ? 1edf5b52-6ae3-41c1-9ec1-c431d34a1aa1 b
Datacenter: us-west1_scylla_node_Portland
======================================
Status=Up/Down
|/ State=Normal/Leaving/Joining/Moving
-- Address Load Tokens Owns Host ID Rack
UN 192.0.2.5 670.16 KB 256 ? f0a44a49-0035-4146-8fdc-30e66c037f95 b
Related Topics