Question: Best practice for configuring with AWS ElastiCache (Cluster Mode Disabled)

[eortiz-tracktik]

Hello Relay Team,

First off, fantastic work on this project. The performance gains and features like connection pooling are incredibly compelling.

I'm working on integrating Relay into a Symfony application that uses AWS ElastiCache for Redis and I'd like to confirm the best practice for configuration, as the official documentation doesn't cover this specific AWS use case in detail.

Our Setup

• Service: AWS ElastiCache for Redis
• Configuration: Cluster Mode Disabled
• Topology: 1 Primary Node and 2+ Read Replicas.
• Key Constraint: ElastiCache in this mode does not expose the Redis Sentinel protocol on its data nodes. This means we cannot use Sentinel for auto-discovery.

Our Goals

1. Connection Pooling: Leverage Relay's core persistent connection management to reduce latency.
2. Read/Write Splitting: Distribute read commands across the read replicas (eventual consistency) while sending all write commands to the primary.
3. Consistent Reads: Have the ability, when necessary, to force a specific read command to go to the primary node to guarantee strong consistency.

Questions

1. What's the recommended configuration approach for an AWS ElastiCache setup without Sentinel?
2. Does Relay automatically identify the first host in the array as the primary (for writes) and the rest as replicas (for reads)?
3. How would we perform a strongly consistent read (i.e., force a GET command to go to the primary) when using this replication configuration? Is there a command-specific option or a client-level method to switch context?

Thank you for your time and guidance! We believe clarifying this will be very helpful for other AWS users in the community.

[tillkruss]

I'll let @michael-grunder respond to this, but my two cents:

• Yes, disable cluster mode unless you really want to use a Redis Cluster
• Usually the application itself (Laravel, WordPress) will handle the primary and replica discovery when supplying just a single AWS endpoint

[michael-grunder]

I actually haven't used AWS a great deal myself.

Unless Laravel has an abstraction to split writes and reads you'd currently need to use multiple clients in your app.

• A writer client pointing at the primary endpoint
• 1-N reader clients pointing at the replica(s)

[eortiz-tracktik]

Thanks for your quick response @tillkruss and @michael-grunder

So there aren't an option equivalent to this one on predis yet ?
https://github.com/predis/predis?tab=readme-ov-file#replication

[tillkruss]

No, Relay and PhpRedis don't have implementations like that. Usually the framework (Laravel, Symfony, etc.) handles that.

You can use Relay as the underlying client when using Predis.

[eortiz-tracktik]

Thanks a lot for you answer @tillkruss
I saw you are actively working on using relay as the underlying client when using Predis https://github.com/predis/predis/wiki/Using-Relay 🎉
We use some of the commands that are still incompatible https://github.com/predis/predis/search?q=relay-incompatible&type=code so I will keep an eye on this.

[tillkruss]

If you post some specific commands, we can probably make them compatible one by one.

Otherwise just use Relay directly, writing a primary-replicate abstraction is trivial for you, I thought Symfony has bundled in anyhow?