Skip to content

[REQUEST] Support Redis Cluster mode (AWS ElastiCache cluster-mode-enabled) #9734

Open

Description

@subinksoman

Is your feature request related to a problem? Please describe.

Self-hosted Sim cannot use a Redis Cluster, which rules out AWS ElastiCache in
cluster-mode-enabled (and ElastiCache Serverless, which requires a cluster-capable
client). The only managed option left is a cluster-mode-disabled replication group,
which caps scaling at a single primary.

The env surface is a single endpoint:

REDIS_URL:            z.string().url().optional()
REDIS_TLS_SERVERNAME: z.string().min(1).optional()

There is no cluster flag and no seed-node list, and Sim never constructs an ioredis
Cluster — a standalone client is built from that one URL. Pointing it at a
configuration endpoint connects to a single shard, so any key hashing to another
slot fails with MOVED.

Notably ioredis 5.11.1 is already bundled and exports Cluster, so the capability
is present in the dependency tree and simply isn't reachable through configuration.

Describe the solution you'd like

Let REDIS_URL (or an additional variable) drive an ioredis Cluster client when
cluster mode is requested. Something like:

  • REDIS_CLUSTER_MODE=true, or a comma-separated REDIS_CLUSTER_NODES, selecting
    new Redis.Cluster(nodes, opts) instead of new Redis(url)
  • the existing REDIS_TLS_SERVERNAME / rediss:// handling carried through to
    redisOptions, since in-transit encryption is the norm on ElastiCache

The pub/sub paths matter as much as the key/value ones here: cluster pub/sub needs
either sharded pub/sub (SSUBSCRIBE) or a node-aware subscriber, so a transparent
swap of the client constructor alone may not be sufficient for live status.

Describe alternatives you've considered

  • Cluster-mode-disabled replication group — what I'm doing now. Works, but it
    caps scaling at a single primary.
  • A cluster-aware proxy (Envoy Redis filter, twemproxy) — not viable. Envoy's
    Redis proxy does not support pub/sub, which is exactly what Sim needs Redis for.
    Caching would work while live status silently never streams.
  • Patching Sim to build a Cluster — possible since ioredis is already present,
    but that means maintaining a fork and rebuilding the image on every upgrade.

Additional context

Observed on ghcr.io/simstudioai/simstudio:latest (pulled 2026-11-06), ioredis
5.11.1, Docker Compose self-host.

Per the compose comments, Redis is optional for storage (Postgres is the fallback)
but the pub/sub channels have no fallback — so for anyone wanting live status on a
managed Redis, the cluster-mode constraint decides the architecture. Happy to test
a patch against ElastiCache if that's useful.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions