Skip to content

Latest commit

 

History

History
83 lines (67 loc) · 3.51 KB

redis_cluster_protocol.md

File metadata and controls

83 lines (67 loc) · 3.51 KB
Raw

Redis Cluster Protocol and Server Proxy

Redis Cluster is the official Redis distributed solution supporting sharding and failover. Compared to using a single instance redis, clients connecting to Redis Cluster need to implement the Redis Cluster Client Protocol. What it basically does is:

  • Redirecting the requests if we are not sending commands to the right node.
  • Caching the cluster routing table from one of the two commands, CLUSTER NODES and CLUSTER SLOTS.

This kind of client is called the Smart Client.

Redis Cluster Protocol

To be compatible with the existing Redis client, there're also some Redis Cluster Proxies like redis-cluster-proxy(official), aster, corvus, and samaritan, to adapt the cluster protocol to the widely supported single instance protocol.

Redis Cluster Proxy

How does Undermoon Implement the "Redis Cluster Protocol"?

Undermoon implements Redis Cluster Protocol based on a server-side proxy or Server Proxy. The Server Proxy will act just like the official Redis Cluster Redis and return redirection response if needed.

Server-side Proxy

Why another "Redis Cluster Protocol" implementation?

This implementation not only support horizontal scalability and high availability, but also enable you to build a self-managed distributed Redis supporting:

  • Redis resource pool management
  • Serving multiple clusters for different users
  • Spreading the flood evenly to all the physical machines
  • Scaling fast
  • Much easier operation and kubernetes integration.

Why server-side proxy?

Redis and most redis proxies such as redis-cluster-proxy, corvus, aster, codis are deployed in separated machines as the proxies normally need to spread the requests to different Redis instances.

Instead of routing requests, server-side proxy serves as a different role from these proxies and is analogous to the cluster module of Redis, which make it able to migrate the data and scale fast by using some customized migration protocols.

Server-side Proxy

A small bite of server-side proxy and Redis Cluster Protocol
# Run a redis-server first
$ redis-server
# Build and run the server_proxy
> cargo build
> make server  # runs on port 5299 and will forward commands to 127.0.0.1:6379
> redis-cli -p 5299
# Initialize the proxy by `UMCTL` commands.
127.0.0.1:5299> UMCTL SETCLUSTER v2 1 NOFLAGS mydb 127.0.0.1:6379 1 0-8000 PEER 127.0.0.1:7000 1 8001-16383

# Done! We can use it like a Redis Cluster!
# Unlike the official Redis Cluster, it only displays master nodes here
# instead of showing both masters and replicas.
127.0.0.1:5299> CLUSTER NODES
mydb________________9f8fca2805923328____ 127.0.0.1:5299 myself,master - 0 0 1 connected 0-8000
mydb________________d458dd9b55cc9ad9____ 127.0.0.1:7000 master - 0 0 1 connected 8001-16383

# As we initialize it using UMCTL SETCLUSTER,
# slots 8001-16383 belongs to another server proxy 127.0.0.1:7000
# so we get a redirection response.
#
# This is the key difference between normal Redis client and Redis Cluster client
# as we need to deal with the redirection.
127.0.0.1:5299> get a
(error) MOVED 15495 127.0.0.1:7000

# Key 'b' is what this proxy is responsible for so we process the request.
127.0.0.1:5299> set b 1
OK