Memcached Cluster Architecture (Nodes, Sharding, and High Availability Design)

Memcached is a distributed memory caching system, and its cluster architecture can significantly improve its availability and performance. However, effectively designing a highly available cluster requires considering many factors.

Node Selection

When designing a Memcached cluster, node selection is critical. Typically, one of the following patterns is adopted:

  • Single Sharding Mode: Each node stores only a portion of the data, which minimizes network traffic.
  • Multi-Sharding Mode: Each node stores different data fragments, which improves cluster availability.
  • Shared Mode: All nodes store the same data, which requires more disk space and bandwidth.

Sharding Design

When using single or multi-sharding modes, sharding design is crucial. Sharding involves dividing a large cache into smaller, independent cache areas to better handle data and improve availability. Typically, the following sharding strategies are adopted:

  • Hash Sharding: Uses a hash function to split data keys into different parts, balancing data distribution.
  • Range Sharding: Divides cache areas based on data ranges, such as splitting data by time order or key value order.
  • Displacement Sharding: Uses a displacement algorithm to generate a series of numbers representing a cache block.

Cluster Swapping

Cluster swapping refers to how to quickly transfer data to another node when a node becomes inactive or a network failure occurs. Typically, the following swapping strategies are adopted:

  • Redundant Data: Each node maintains redundant data, allowing for rapid transfer of data from failed nodes to other healthy nodes.
  • Slow Write: Stores newly written data on a slow node, avoiding potential issues during normal operation.

Read Load Balancing

Data in the cluster needs to be read from the correct node. Typically, the following read load balancing strategies are adopted:

  • Round-Robin: Distributes all requests to each node.
  • IP Hash: Uses a hash function corresponding to the IP address to determine which node reads the data.
  • Geographic Hash: Selects a nearby node based on the hash value corresponding to the geographic location.

Event Handling

Event handling is very important when designing a highly available cluster. Typically, the following event handling strategies are adopted:

  • Scheduled Tasks: Uses scheduled tasks on each node to periodically check data status.
  • Redundant Data: Stores redundant data on slow nodes, avoiding potential issues during normal operation.

Optimization and Monitoring

Finally, highly available clusters require optimization and monitoring. Typically, the following optimization strategies are adopted:

  • Load Balancing: Uses a load balancer to balance read and write traffic.
  • Configuration Management: Defines cluster settings through configuration files or system variables.

Problem Solving

When designing a highly available cluster, the following issues often arise:

  • Data Loss: How to avoid data loss?
  • Node Failure: How to handle a failed node?
  • Network Failure: How to quickly transfer to another node during a network failure?

Examples

Here are several design examples for highly available Memcached clusters:

Single Sharding Mode

# Single Sharding Mode

*   Node 1: 127.0.0.1:11211
*   Node 2: 127.0.0.1:11212
*   Node 3: 127.0.0.1:11213

Multi-Sharding Mode

# Multi-Sharding Mode

*   Node 1 (KEY 00-99): 127.0.0.1:11211
*   Node 2 (KEY 100-199): 127.0.0.1:11212
*   Node 3 (KEY 200-299): 127.0.0.1:11213

Shared Mode

# Shared Mode

*   Node 1: 127.0.0.1:11211
*   Node 2: 127.0.0.1:11212
*   Node 3: 127.0.0.1:11213

Read Load Balancing Strategy

# Read Load Balancing Strategy

*   Round-Robin: 127.0.0.1:11211, 127.0.0.1:11212, 127.0.0.1:11213
*   IP Hash: 127.0.0.1:11211, 127.0.0.1:11212, 127.0.0.1:11213