Redis High-availability evolution

Recently collated Redis, a deeper understanding of Sentinel, specifically summarized as follows

1. master-Slave Redis

Master-Slave Redis is actually a primary standby mode that allows you to switch from Redis to service after the master Redis outage.

Disadvantages:

1. People pay attention to whether Master is down

2. Unable to complete auto switch master/slave

3. The function of the slave node is not fully utilized

Master-Slave Mode:

2.sentinel

To address this determination, Redis officially provides Sentinel to ensure high availability of Redis

Figure 1 shows the relationship between Sentinel and Redis, where the Sentinel system creates two links for each Redis instance (master, slave): Command connection, subscription connection

Command connection: Send info command to keep communication with Redis

Subscribe to connections: Automatically discover other Sentinel instances by subscribing to a connection

Figure 2 shows a minimal sentinel, which is comprised of at least three sentinel instances, and when the monitored Redis is judged to be the supervisor downline, it is necessary to elect a minor sentinel from Sentinel for master-slave switchover.

Advantages:

1.SENTIENL can monitor the health status of master-slave node, reduce the cost of human monitoring

2.sentinel can complete master-slave switching

Disadvantages:

1. From the node is still not fully utilized

2. Unable to scale out, only one master of the server is available

Sentinel Mode:

Figure 1

　　　　　　

Figure 2

3. sharding

The idea of sharding is to achieve the scale-out of Redis by using the consistent hashing algorithm.

1. Storage of Sentinel configuration information via zookeeper

2. Implement a consistent hashing algorithm on the client, which determines that the Redis command is executed by the Redis instance through the routing algorithm

3. To share the pressure of a single shard by increasing shard

Disadvantages:

1. Data migration is involved in the expansion, if the Redis is only cached data is easy to handle, but if there is a strong reliance on Redis business data, the migration can only stop processing

2. Unable to solve hot and cold data problems

Shard Mode:

The following sections focus on:

Jedisclient the implementation principle of the customer list

Redis High-availability evolution