Redis - Replication

Redis Replication

Redis Replication

Key Facts

• Replication is Asynchronous
  • When a client writes data to the main Redis server (called the Master), the data is copied to other Redis servers (called Replicas).
  • However, this replication happens asynchronously—meaning the client doesn’t control when the data reaches the replica.
• Non-blocking Replication on Master
  • Replication doesn’t slow down the Master server for clients. Redis Master continues to respond to clients while sending data to replicas.
• Occasional Blocking on Replica
  • Sometimes, a replica might be syncing data from the Master and won’t respond to client requests until it’s up-to-date.
• Flexible Number of Replicas
  • A Redis Master can have as many replicas as needed for distributing data.
• Replicas are Usually Read-Only
  • By default, replicas only respond to read requests.
  • You can allow write requests on replicas, but changes made this way are only temporary; they’ll be overwritten when the Master syncs with the replica again.
• Replica as a Master
  • A replica can act as a Master to other replicas, creating a chain of replication if needed.

Why Use Replication?

1. High Availability: Replicas provide backups of data, helping keep data accessible even if one server fails.
2. Read Load Distribution: Multiple replicas can handle read requests, balancing the load.

3. Alternative to Persistence: Replication can provide data reliability even if persistence (writing to disk) is not configured.

   

Types of Replication

1. Full Sync

• When a new replica joins, it needs all the data from the Master. This is called Full Sync.
• How Full Sync Works:
  • Redis Master forks (creates a separate process) to copy all data to the replica, which lets the Master keep responding to clients.
  • Two Methods for Data Transfer:
    1. Disk-based: The Master saves data to disk and sends the file to the replica (older method).
    2. Diskless: The Master sends data directly to the replica without saving to disk.
• When Full Sync Happens:
  • Setting up a new replica
  • When a replica falls too far behind in syncing with the Master

2. Partial Sync

• In cases where only a small amount of data needs syncing, Partial Sync updates replicas with recent changes instead of copying everything.
• Key Components:
  • Replication Backlog: A limited-size memory buffer where recent writes are stored.
  • Replication ID: Each server gets a unique ID on restart, helping to manage syncs.
  • Replication Offset: Keeps track of the exact position of each update in the backlog.
  • Replica Client Output Buffer: Stores data that the Master sends to each replica.

Important Replication Configurations

• replicaof <master_ip> <master_port>
  • Connects a Redis instance to a Master as a replica.

  • Example:

      replicaof localhost 7001  # Connect to a Master on localhost at port 7001
      replicaof NO ONE          # Disconnects from Master
            
    

• repl-backlog-size <size>
  • Sets the size of the replication backlog buffer (e.g., 1mb).
• repl-timeout <seconds>
  • Sets a time limit for replicas to wait for data from the Master before timing out (e.g., 60 seconds).
• client-output-buffer-limit replica <hard limit> <soft limit> <time>
  • Configures limits for each replica’s output buffer. Exceeding this can cause disconnection.

Consistency in Redis

Redis doesn’t guarantee 100% consistency but offers settings to reduce inconsistency risk:

• Read-Only from Master
  • By reading only from the Master, clients get the most accurate data.
• Server Settings for Consistency
  • min-replicas-to-write and min-replicas-max-lag: This two config option goes hand-in-hand, i.e., if min-replicas-to-write that has a lag fewer than min-replicas-max-lag then accept the write.
• Client Option for Consistency
  • WAIT <num-replicas> <timeout-millis>: Ensures a certain number of replicas receive data before the command completes.

Replication Monitoring

• role

• info replication

Memory Usage Considerations

• Replication Backlog:
  • Shared buffer for all replicas, reducing available memory for user data.
• Replica Client Output Buffer:
  • Each replica has its buffer, with limits to control memory usage.

Common Replication Issues & Fixes

1. Out-of-Memory for Full Sync:

   • If a Full Sync fails due to lack of memory, allow Redis to allocate more memory with:

       echo 1 > /proc/sys/vm/overcommit_memory
             
     

2. Timeout on Replica During Full Sync:
   • Increase repl-timeout if replicas disconnect during Full Sync.
3. Output Buffer Limit Exceeded:
   • If the output buffer overflows, increase the client-output-buffer-limit replica setting to avoid disconnection.