Bug: RedisCacheStrategy fails to connect using Sentinel config

[apoca]

Describe the bug
The RedisCachePlugin fails to connect when using Redis Sentinel configuration. This is due to the internal usage of new IORedis.Redis(...) inside RedisCacheStrategy, which bypasses the automatic connection logic provided by the main ioredis constructor. As a result, the plugin throws a connect ETIMEDOUT or All sentinels are unreachable error, even though the same Sentinel setup works perfectly with BullMQJobQueuePlugin.

To Reproduce
Steps to reproduce the behavior:

1. Enable Redis Sentinel and configure the following in your Vendure setup:

RedisCachePlugin.init({
  redisOptions: {
    sentinels: [
      {
        host: process.env.REDIS_SENTINEL_HOST || 'localhost',
        port: parseInt(process.env.REDIS_SENTINEL_PORT || '26379', 10),
      },
    ],
    name: process.env.REDIS_SENTINEL_NAME || 'mymaster',
    maxRetriesPerRequest: null,
  },
  namespace: process.env.NAMESPACE,
})

2. Start Vendure
3. Observe the error in logs:

[RedisCacheStrategy] connect ETIMEDOUT
[RedisCacheStrategy] All sentinels are unreachable. Retrying from scratch...

4. Note: BullMQ plugin with the exact same Redis Sentinel configuration works as expected.

Expected behavior
RedisCachePlugin should correctly connect to Redis using Sentinel configuration, as supported by ioredis.

Proposed fix

In RedisCacheStrategy, the Redis client is currently instantiated like this:

this.client = new IORedis.Redis(this.options.redisOptions ?? {});

This causes the Redis constructor of a submodule to be used (IORedis.Redis), which does not correctly handle Sentinel configuration.

The correct way is:

this.client = new IORedis(this.options.redisOptions ?? {});

This uses the default export and ensures Sentinel discovery works properly, just like in the BullMQ plugin.

Environment (please complete the following information):

• @vendure/core version: 3.1.7
• Nodejs version: v18, v20
• Database (mysql/postgres etc): postgres

Additional context
I am happy to contribute a pull request to fix this issue, as it’s a one-line change with significant impact for setups using Redis Sentinel. Just let me know!