feat: add Redis Cluster support via REDIS_CLUSTER_MODE setting

[bebakouma]

Problem

When deploying ContextForge against a Redis Cluster (e.g. Bitnami redis-cluster Helm chart with multiple shards), the current standalone redis.asyncio.Redis client cannot handle MOVED/ASK redirects. This causes the application to crash-loop with redis.exceptions.MovedError on startup during leader election.

The error occurs because redis.asyncio.from_url() always creates a standalone client that connects to a single node. When a key lives on a different shard, Redis responds with a MOVED redirect that the standalone client cannot follow.

Solution

Add a REDIS_CLUSTER_MODE boolean environment variable (default: false). When set to true, the redis_client factory creates a redis.asyncio.RedisCluster instance that automatically discovers all shards and routes commands to the correct node.

Changes

• config.py — new redis_cluster_mode setting with Field descriptor
• utils/redis_client.py — refactored into _create_cluster_client() and _create_standalone_client() helpers; strips /N database path from URL in cluster mode (Redis Cluster only supports db 0); selects client type based on redis_cluster_mode setting

Usage

REDIS_CLUSTER_MODE=true
REDIS_URL=redis://:password@redis-cluster:6379   # no /0

Compatibility

• Backward compatible — default is false, existing standalone deployments are unaffected
• RedisCluster supports the same .set(), .get(), .ping(), .eval(), .publish(), .pubsub() APIs used throughout the codebase
• PubSub works on RedisCluster (broadcasts to all nodes)
• Leader election Lua script uses a single key, so it hashes to one slot (no cross-slot issues)

Environment

Tested against Bitnami redis-cluster Helm chart v12.0.2 (3 nodes) where the redis-cluster ClusterIP service load-balances across shards.

[crivetimihai]

Thanks @bebakouma — clean implementation. The cluster/standalone factory split, _strip_db_from_url() validation, and test coverage are well done.

One security issue to fix:

The cluster-mode log line logs the full URL:

f"url={_strip_db_from_url(settings.redis_url)}"

If REDIS_URL contains a password (e.g., redis://:secret@host:6379), the password is logged in plaintext. The standalone path doesn't log the URL. Please mask or remove the URL from the log output.

Minor suggestions (non-blocking):

• Consider adding REDIS_CLUSTER_MODE to the Helm chart values (charts/) for parity
• A brief mention in the deployment docs would help operators

DCO: Signed-off-by line is required on all commits — use git commit --amend -s.

Otherwise this looks ready for review. Nice work!