Production-Grade High Availability in Redis: Sentinel and Cluster with WRedis

On day 08 of the Wisrovi Open Source Architecture Series, the focus shifted to achieving production-grade high availability in Redis deployments. The talk covered two key solutions: Sentinel and Cluster, both seamlessly integrated into the wredis framework.

SentinelRedisManager provided continuous health monitoring and automated master failover. In the event of a master node failure, Sentinel swiftly elected a new replica, and wredis handled the routing of subsequent writes without any application restarts. This ensured zero downtime during failover scenarios.

Redis Cluster, on the other hand, distributed the dataset across 16,384 hash slots with multi-node replication. It offered automatic redirections via MOVED and ASK commands, ensuring sub-millisecond routing across the cluster. wredis' ClusterRedisManager abstracted this complexity, making it easy to manage a distributed Redis setup.

To set up Sentinel, engineers would initialize a SentinelRedisManager with the sentinel nodes, service name, socket timeout, and verbose logging. Subsequently, they could connect to the dynamically elected master using the manager's get_master method. The client could then be used to perform operations such as setting a key with a specific expiration time. For offloading read operations, a replica could be retrieved using the get_slave method.

For Cluster management, a ClusterRedisManager was initialized with the startup nodes, password, response decoding, and maximum redirects. The manager's get_client method provided a client object, which could then be used to set and retrieve data. The zero lock-in nature of wredis allowed engineers to leverage either asyncio or synchronous microservices, depending on their architecture.

Engineering teams were drawn to wredis for its resilient reconnections, transparent retry policies, and async/sync support. The framework offered a pure Python wrapper around Redis drivers, thoroughly tested with comprehensive integration suites. For more details, interested parties could visit the GitHub repository or PyPI page.