Redis Cluster is the one database topology whose minimum is set by the protocol rather than by the workload. Redis's scaling guide, checked on 1 November 2026, describes a cluster of three primaries with a replica each, data split over 16,384 hash slots, and a second port per node for the cluster bus, and those facts decide the smallest honest cluster. This guide reads them and prices three sizes of the Redis Cluster package on MassiveGRID PaaS, built on Virtuozzo Application Platform (formerly Jelastic).
What Redis Cluster requires
The scaling guide's cluster tutorial builds its example with three masters and three replicas, the smallest layout in which every shard survives a node loss; a cluster with N primaries uses N-1 additional nodes for replicas at minimum. The key space is divided into 16,384 hash slots, each primary owning a range, and a client is redirected to the right node by slot. Every node listens on two ports: the data port, 6379 by default, and the cluster bus port, which is the data port plus 10,000 and must be reachable from every other node but never used by clients.
Availability has a rule of its own: the cluster keeps operating while a subset of nodes fails, but becomes unavailable when the majority of masters is unreachable. Replication is asynchronous, so a failover can lose the last writes a primary acknowledged, which the guide says plainly; a primary unreachable for longer than the configured node timeout is failed over by one of its replicas.
Redis Cluster on MassiveGRID PaaS: one-click install, about $29.52 a month for the Starter configuration (6 × Redis node, 256 MiB each) at $0.003372 per cloudlet-hour, before volume discounts, in New York, London, Frankfurt or Singapore. 14-day free trial, no credit card. See the sizing table.
Memory decides the node size
Redis is memory-bound, and the guide's facts leave one variable: how much data each primary holds. A session store for a web application is tens of megabytes; a cache of rendered fragments for a busy site is a few hundred; a queue or a feature store can be gigabytes. Each primary needs memory for its share of the data plus overhead, and each replica needs the same, because it holds a full copy of its primary. Size the node to the largest shard you expect, set maxmemory below the node's limit with an eviction policy appropriate to the use (allkeys-lru for a cache, noeviction for a queue), and let the platform's vertical scaling cover growth between resizes.
The three configurations on PaaS and their cost
The Redis Cluster package installs the six-node layout, three primaries with a replica each, with the cluster configuration and optional TLS. Cloudlets are 128 MiB RAM plus 400 MHz CPU, billed at $0.003372 per cloudlet-hour, $2.46 per cloudlet-month, no licence fee.
| Configuration | Per node | Nodes | Cloudlets | Per month (730 h) | Fits |
|---|---|---|---|---|---|
| Starter | 256 MiB | 6 | 12 | $29.52 | Sessions and small caches |
| Standard | 512 MiB | 6 | 24 | $59.04 | Application caches up to a few hundred MB |
| Production | 1 GiB | 6 | 48 | $118.08 | Larger datasets |
Redis is frugal with CPU, which is why the smallest honest cluster costs under $30 a month; a cluster with a dataset of several gigabytes per shard scales the per-node figure, not the node count. The figures are ceilings before volume discounts, and the PaaS pricing calculator prices larger nodes or more shards.
What the application has to do
Cluster mode is not transparent to a client written for a single Redis. The client must be cluster-aware, which every major library is, so that it follows slot redirections and keeps a map of which node owns which range; multi-key operations work only when the keys hash to the same slot, which hash tags ({user:42}:session) arrange. Sessions, caches and rate limiters fit cluster mode naturally; a workload that relies on transactions across arbitrary keys does not, and is better served by a single Redis with a replica. The session store post shows the client configuration for PHP and Node.js applications behind a load balancer.
Operations: failure, resharding and TLS
A primary failure is handled by the cluster, and the platform restarts or migrates the failed container so it rejoins as a replica; a lost replica is replaced the same way. Growing from three shards to four is a resharding operation that moves slots live, which the operations post walks through on the package. The Redis TLS add-on encrypts client connections and the cluster bus for environments where another application or a reviewer requires it, and Redis persistence (RDB snapshots, AOF) is configured per node for the data that must survive a full restart.
Frequently Asked Questions
What is the minimum number of nodes for a Redis Cluster?
Three primaries, each with a replica, six nodes in total, as in the example in Redis's scaling guide checked on 1 November 2026. Fewer primaries cannot keep a majority through a failure, and a primary without a replica loses its slots when it fails. The package on MassiveGRID PaaS installs that six-node layout from about $29.52 a month.
What ports does Redis Cluster use?
Each node uses its data port, 6379 by default, and a cluster bus port at the data port plus 10,000, which must be reachable from every other node and is never used by clients. On the platform the nodes talk over the internal network, and the TLS add-on can encrypt both.
Can my application use Redis Cluster without changes?
It needs a cluster-aware client, which the major libraries provide, and multi-key commands must use keys that hash to the same slot, which hash tags arrange. Session stores, caches and rate limiters fit; workloads built on cross-key transactions are better served by a single Redis with a replica.
The six-node cluster Redis describes, in one click
The Redis Cluster package on MassiveGRID PaaS installs three primaries with a replica each, the cluster configuration and optional TLS, from about $29.52 a month for session stores and small caches. 14-day free trial, no credit card.
Redis Cluster on PaaS