A problem with clustered Redis and a solution

Two problems are highlighted when using Redis in cluster mode.

1. Unreliable Partial Data Sync

2. The two connection issue

The Problem - Unreliable Partial Data Sync

1. The client requests to write to the master node.

2. Slave receives a replication command from master.

3. The master gives the client an OK response.

In this case, should the system crash after step 2, the client may believe that the data is secure because it obtained approval from the master prior to the replication instruction being issued.

The Solution: Partial Data Synchronisation

1. The client sends the master node a write request.

2. The master instructs the slave to replicate.

3. The master gives the client a "OK."

In this case, the client is certain that the slave receives a replication command. Since the slave returned to the client with an OK after receiving a replication request from the master, it is assumed that data replication has been successful in this instance.

Although it's not ideal, this works best when taking into consideration the performance, cost of utilising full data synchronisation.

The Problem : Two Connection Issue

1. The client sends a read request to a node using a key.

2. The Node doesn’t have the slot that owns the key and replies with MOVED command.

3. The client sends a read request to the appropriate node using a key.

4. The value is provided to the client against that key.

The Solution : Client side key-to-node cache

1. The client checks for the key in the cache. A request goes to a "default" node if the key cannot be found.

2. The correct node is updated in the cache in the event that a MOVED error is received.

3. The client calls the correct node.

4. The value is handed over to the client using that key.

---

Have Questions or Suggestions ?

Let's connect.