Distributed agreement of Lease mechanism

1. What is the Lease mechanism

The Lease mechanism, translated as the lease mechanism, is a commonly used protocol in distributed systems and a common tool for maintaining data consistency in distributed systems.

The Lease mechanism has the following features:

• Lease is the issuer's commitment to data consistency over time;
• After the issuer issues the Lease, regardless of whether it is received or not, as long as the lease does not expire, the issuer will abide by the agreement;
• Lease holders can only use the promise within the validity period of the Lease. Once the Lease expires, the holder needs to give up the execution and re-apply for the Lease.

Take car rental as an example:

2. What problem does the Lease mechanism solve?

In a distributed system, how to confirm whether a node is working properly?If there are 5 copies 1-5.Among them, No. 1 is the master copy.

The most intuitive processing method in distribution is to maintain a heartbeat between each replica and the primary replica, hoping to judge whether the other party is still alive by the existence of the heartbeat.

The heartbeat method can't solve the problem of whether the nodes are normal under the distributed environment at all.Consider the following scenario:

1. At a certain time, the master node of Node1 suddenly has network jitter or network interruption (note: not downtime), resulting in the slave node unable to receive the heartbeat.

2. One of the remaining secondary nodes will be selected as the primary node.

There are four main solutions:

• Design a distributed protocol that can tolerate dual masters
• Raft Protocol - High sync low via Term version.
• Use the lease mechanism
• Involving Decentralization-Gossip Protocol

2. The principle of Lease

      1. Introduce the central node to be responsible for issuing Lease

     2. There is a network problem

If there is network jitter during 01:05, other nodes will fail to apply for Lease, because the central node will admit that there is a master node before 01:10, and other nodes are not allowed to apply for Lease

     3. If the network recovers

     4. If it reaches 01:10, the master node will renew the contract, and then issue a new lease

      5. If the master node goes down, other nodes will fail to apply for Lease, and the master node will be acknowledged

      6. The secondary node applied for Lease, and the application was successful. Because the Lease expired

3, fault tolerance of lease

The master node is down: The lease mechanism is inherently tolerant of network and lease receiver errors, and the time is the remaining expiration of the lease.Duration

Central node exception: The downtime of the issuer may cause all nodes to have no lease, and the system is in an unavailable state. The solutionJust use a small cluster instead of a single node as the issuer.

Time difference problem: There may also be an error in the clock between the central node and the master node, only the central node needs to consider the clock errorThat's it.

The length of lease time generally takes the experience value of 1-10 seconds.If it is too short, the network will be under great pressure. If it is too long, it will take too long to recover the commitment, which will affect the availability.

4. Application

1. In GFS (Google File System), the Master decides which is the master copy through the lease mechanism, and the lease is carried in the heartbeat response message to each node.When the heartbeat is not received, wait for the lease to expire before issuing it to other nodes.
2. In chubby, after paxos elects the master, the slave node will issue a lease to the master, and no other nodes will be selected as the master within the time limit.On the other hand, the master node sends a lease to each client node to judge whether the client is alive or dead.