Let's do a quick demo to see this in action. We will set up a Redis cluster on your local machine with one leader and one follower.
Replication with Redis
Start the primary Redis instance on port 6379 (default port).
redis-server
And then start the replica Redis instance on port 6380, and specify the primary instance as its master:
redis-server --port 6380 --slaveof 127.0.0.1 6379
Replication from Leader to Follower
And run the following Python code to write to the primary and read from both primary and replica.
import redis # Connect to the primary and replica nodes primary = redis.StrictRedis(host='localhost', port=6379, db=0) replica = redis.StrictRedis(host='localhost', port=6379, db=0) # Write to the primary node primary.set('key', 'value') # Read from the primary node print("Primary node value:", primary.get('key')) # Read from the replica node print("Replica node value:", replica.get('key'))
You should see
Primary node value:value Replica node value:value
Even though we only wrote to the primary node, the replica node still got the changes because it received the data changes from the primary.
We can also see the replica syncing with the master when it started.
:S 26 Apr 2023 17:10:42.924 * MASTER <-> REPLICA sync started 11333:S 26 Apr 2023 17:10:42.925 * Non blocking connect for SYNC fired the event. 11333:S 26 Apr 2023 17:10:42.925 * Master replied to PING, replication can continue... 11333:S 26 Apr 2023 17:10:42.925 * Trying a partial resynchronization (request 28d99e54a99142d4d9de07c45f2d0e82eea125ef:2457). 11333:S 26 Apr 2023 17:10:42.925 * Successful partial resynchronization with master. 11333:S 26 Apr 2023 17:10:42.925 * MASTER <-> REPLICA sync: Master accepted a Partial Resynchronization
Read From Any Node
Typically in a single-leader setup,
- Read requests can go to any node
- Write requests only go to the primary
In the example above, if you try replica.set('key', 'value'), you'll get an error:
File "/Library/Frameworks/Python.framework/Versions/3.10/lib/python3.10/site-packages/redis/connection.py", line 882, in read_response raise response redis.exceptions.ReadOnlyError: You can't write against a read only replica.
Failover
Let's briefly shut down the leader and turn it back up and see what happens. In the real world, this can be a network partition between the leader and the follower.
11550:S 26 Apr 2023 20:22:20.879 # MASTER timeout: no data nor PING received... 11550:S 26 Apr 2023 20:22:20.879 # Connection with master lost. 11550:S 26 Apr 2023 20:22:20.879 * Caching the disconnected master state. 11550:S 26 Apr 2023 20:22:20.879 * Reconnecting to MASTER 127.0.0.1:6379 11550:S 26 Apr 2023 20:22:20.880 * MASTER <-> REPLICA sync started 11550:S 26 Apr 2023 20:22:20.880 * Non blocking connect for SYNC fired the event. 11550:S 26 Apr 2023 20:22:20.881 * Master replied to PING, replication can continue... 11550:S 26 Apr 2023 20:22:20.881 * Trying a partial resynchronization (request 28d99e54a99142d4d9de07c45f2d0e82eea125ef:7966). 11550:S 26 Apr 2023 20:22:20.881 * Successful partial resynchronization with master.
We can see the follower attempting to reconnect and trying to synchronize (SYNC event) with the leader as soon as it's able to reach it.