Clustering Overview¶
Introduction to clustering¶
A cluster consists of multiple instances of a product that divide up the work and act as a single instance. This improves performance as the requests are distributed among several servers instead of just one. It is also more reliable as there are other instances to handle requests when one instance becomes unavailable. Following are several benefits of clustering.
-
High availability : Some systems require high availability percentages such as two-nines (99%). A server may go down due to many reasons such as system failures, planned outage, or hardware or network problems. Clustering for high availability results in fewer service interruptions. Since downtime is costly to any business, clustering has a direct and positive impact on costs.
-
Simplified administration : You can add and remove resources according to your size and time requirements. You can also launch compute jobs using simple APIs or management tools and automate workflows for maximum efficiency and scalability. Administration is simplified by using tools like the deployment synchronizer and log collector.
-
Increased scalability : Scalability is the ability of a system to accommodate a growing amount of work. Scalability is using resources more effectively. By distributing processing, we can make vertical or horizontal scalability possible.
-
Failover and switchover capabilities : Failover can occur automatically or manually. You can prepare a redundant backup system or use load-balanced servers to serve the failover function. You address failover through your system design and characteristics, and clustering helps you design your applications against interruptions and with improved recovery time. Even if a failover occurs, it is important to bring the system back up as quickly as possible.
-
Low cost : Clustering improves scalability and fault tolerance, so business continuity is guaranteed even in the case of node failure. Also, it facilitates automatically scaling up the system when there is a burst load, which means the business will not lose any unforeseen opportunities.
These characteristics are essential for enterprise applications deployed in a production environment. Therefore, you need a cluster when you go into production when performance and reliability are critical.
About membership schemes¶
A cluster should contain two or more instances of a product that are configured to run within the same domain. To make an instance a member of the cluster, configure it to either of the available membership schemes, which are as follows:
- Well Known Address (WKA) membership scheme
- Multicast membership scheme
- AWS membership scheme
- AWS ECS membership scheme
All of these membership schemes are ready to be used in production. You can select a scheme based on your production environment. Here's a comparison of the membership schemes:
Multicast | WKA | AWS/AWS ECS |
---|---|---|
All nodes should be in the same subnet | Nodes can be in different networks | Amazon EC2 nodes |
All nodes should be in the same multicast domain | No multicasting requirement | No multicasting requirement |
Multicasting should not be blocked | No multicasting requirement | No multicasting requirement |
No fixed IP addresses or hosts required | At least one well-known IP address or host required | No fixed IP addresses or hosts required |
Failure of any member does not affect membership discovery | New members can join with some WKA nodes down, but not if all WKA nodes are down | Failure of any member does not affect membership discovery |
Does not work on IaaSs such as Amazon EC2 | IaaS-friendly | Works on Amazon EC2 |
No WKA requirement | Requires keepalive, elastic IPs, or some other mechanism for re-mapping IP addresses of WK members in cases of failure | No WKA requirement |
Note
Some production environments do not support multicast. However, if your environment supports multicast, there are no issues in using multicast as your membership scheme.
About Well-Known Addresses (WKA)
The Well-Known Addresses (WKA) feature is a mechanism that allows cluster members to discover and join a cluster using unicast instead of multicast. WKA is enabled by specifying a small subset of cluster members (referred to as WKA members) that are able to start a cluster. The WKA member starts the cluster and the other members join the cluster through this WKA member. When the WKA member is down, the cluster breaks, and the members cannot communicate with each other.
The system should have at least two well-known address (WKA) members in order to work correctly and to recover if a single WKA member fails.
Clustering compatibility with WSO2 products¶
WSO2 products are compatible with each other if they are based on the same WSO2 Carbon version. See the release matrix for compatibility information.
About performance of WSO2 products in a cluster
If you are setting up multiple WSO2 products in a cluster, it is recommended to set up each product on a separate server.
Deciding how to set up your cluster¶
Note
If there are multiple WSO2 IS nodes in a deployment, it is mandatory to enable hazelcast clustering. Learn more.
When setting up your cluster, you must decide how you want to set up separate databases for clustering, whether to front your cluster with a load balancer, and whether to use sticky sessions. WSO2 Identity Server uses Hazelcast as the underlying clustering engine.
Top