What is the architecture design of Zookeeper? What are the differences between Leader, Follower, and Observer roles?

Answer

Zookeeper adopts a master-slave architecture design, achieving data consistency and high availability through the ZAB protocol.

Cluster Roles

1. Leader

   • Handles all write requests
   • Responsible for coordinating cluster state
   • Maintains data replica synchronization
   • The only node that can process transaction requests

2. Follower

   • Handles client read requests
   • Participates in Leader election voting
   • Receives data synchronization from Leader
   • Forwards write requests to Leader

3. Observer

   • Only handles read requests
   • Does not participate in Leader election
   • Does not participate in data write voting
   • Receives data synchronization from Leader

Architecture Advantages

High Availability:

• Any node failure in the cluster does not affect service
• New Leader automatically elected when Leader fails
• Supports rapid fault recovery

Scalability:

• Improve read performance by adding Observer nodes
• Horizontally scale read capability
• Reduce load on Leader and Follower

Data Consistency:

• ZAB protocol guarantees strong consistency
• All Followers synchronize data with Leader
• Read requests may read stale data

Deployment Recommendations

Node Count:

• Production environment recommends 3, 5, or 7 nodes
• Odd number of nodes facilitates election voting
• Minimum 3 nodes to ensure high availability

Role Assignment:

• Leader + Follower: Ensure data consistency
• Add Observer: Improve read performance
• Typical configuration: 3 nodes (1 Leader + 2 Followers)

Network Requirements:

• Low-latency network between nodes
• Stable network connection
• Avoid cross-datacenter deployment

Client Connection

• Client connects to any available node
• Automatic reconnection mechanism
• Session timeout detection
• Watcher event notification