WSO2 Clustering Briefings

Intro

• High availability AND Scaling
• How to?
• Vertical scaling
• Horizontal scaling - have a load balancer
• Clustering for the scalability
• nodes should be stateless
• Clustering for availability
• Session/state replication
• TCP communication is involved. TCP communication is done via sockets. So it requires IP/port for each member node.
• Static membership - can't add nodes without restarting the cluster
• Dynamic membership - nodes can join/leave
• Hybrid membership  - dynamic nodes revolves around the static members

Membership management- membership discovery
1. Multicast based membership mgt

• After the discovery, communication is happening p2p
• Only for testing purposes

2. Well-known Address based(WKA)

• no need of public channel
• At-least 2 well-known members
• Well-known members notify the newly joined dynamic members

• In cloud and internet you can't multicast, so WKA is preferred
• Hazlecast, apache tribes handles membership mgt
• Hazlecast
• Clustering and highly scalable data distribution
• Partitions the data-> in-memory data grid which is distributed too
• Auto-scaling
• allocation/deallocation of nodes based on the current load
• Deployment types
• Single node
• Primary-secondary
• Have a secondary node in case of primary node failures
• Primary-secondary with multiple LB
• multi datacenters
• Management and Worker nodes separation
• Management nodes controls the cluster, upload artifacts, do the management work etc.
• Workers perform the main work, shouldn't do any mgt work
• This Separation is done using sub domains, mainly routing the requests based on it
• A usecase- deployment synchronization
• Caching- when expensive data transfers to be done
• JCache

ELB - Elastic load balancing

• Only HTTP traffic, not TCP
• Proxy port
• When multiple HTTP/HTTPs ports are there portmapping should be done - axis2.xml
• Tenant-aware LB
• tenant partitioning - define a tenant range
• LB routes the tenant requests to the relevant cluster based on the tenant range
• Private Jet Mode - dont share the resources with others - done using tenant range
• loadbalancer.conf

LB Cluster

In case of failure of LB node, secondary node has to learn from the beginning. To avoid this state replication is done.