Distributed Systems

1. What It Is - Replicas generally have two roles: Leader, responsible for handling client write requests; Follower, synchronizes data from the Leader and can handle client read requests. 2. Categories 2.1. Leader-Follower. 2.2. Leader-Leader. 2.3. Leaderless.

1. What Is Partitioning - Split one set of data into multiple parts and store them on different nodes. - There are two layers of mapping: take a field from the data as the key, then map key -> partition; then map partition -> machine/node. 2. Why Partitioning Is Needed - The data volume is too large to store on one node and must be distributed.

Distribute keys across partitions/nodes as evenly as possible. 1. Choosing the Split Key 1.1. Primary-Key ID - For example, an auto-increment primary key. The advantage is even data distribution; the disadvantage is that queries by business fields may be slow because all partitions need to be read. 1.2. Business ID - Such as user ID or product ID.

The routing component routes client read and write requests to the node that contains the corresponding partition. 1. Routing Component. 2. Request Processing 2.1. Add Data - 1. The client generates data containing a sharding key. 2. The client sends the add request to the routing component.

1. client - The client locally stores the relationship between partitions and server nodes and directly requests the correct node. 2. proxy - The client requests the routing layer, and the routing layer forwards the request to the correct node. 3. server - The client requests any server node, and that server node forwards the request to the correct node.

Distribute partitions/nodes across machines as evenly as possible. 1. Assignment Methods 1.1. Static Assignment - Create far more nodes than machines. Advantage: when migrating nodes to other machines, the cluster can still respond externally. Disadvantage: the maximum number of machines is fixed.