# Memorystore Redis

> zop.dev provisions a google_redis_instance (tier BASIC by default, memory_size 1 GB, replica_count 1, connect_mode DIRECT_PEERING, redis_version REDIS_7_0) plus a firewall rule and a Kubernetes ExternalName service, so applications connect through a stable in-cluster name that survives endpoint changes, resizes, and instance replacement.

Source: https://zop.dev/integrations/provision/gcp/memorystore-redis
Updated: 2026-08-19

---

Memorystore is Google's managed Redis service. zop.dev provisions the instance with network access and an in-cluster DNS alias so applications connect by a stable service name.

## A Redis instance, a firewall rule, and an alias

google_redis_instance (cluster or single), firewall rule, and a Kubernetes ExternalName service.

## Tier, memory, replicas, and connect mode

```text
tier BASIC|STANDARD (default BASIC); memory_size 1 GB; replica_count 1; connect_mode DIRECT_PEERING; redis_version REDIS_7_0.
```

## Only BASIC and STANDARD are cataloged

Only BASIC and STANDARD tiers are cataloged for Memorystore.

## BASIC restarts empty

The default tier is BASIC: one node, no replica. Maintenance windows, scaling operations, and
failures all share one consequence: the instance comes back with nothing in it. For a true
cache that repopulates from a database, a cold start is a latency blip and BASIC's price is
the right trade. For sessions, queues, rate-limit counters, or anything else that cannot be
recomputed, STANDARD is the floor. It adds the replica and automatic failover that make a
flush an event instead of a data loss. The tier choice is really a question about what the
application put in Redis, and it deserves an explicit answer per environment.

## Per-GB-hour, always on

Memorystore bills by provisioned capacity per hour: `memory_size` (1 GB default) times tier
rate, continuously, idle or busy. STANDARD roughly doubles BASIC at the same size because the
replica is a second node. Capacity can be grown later, so the cheap strategy is starting at
the default and scaling on evidence rather than provisioning for a guess.

## Two settings frozen at creation

`connect_mode` is permanent for the instance's life, and DIRECT_PEERING is the value here.
Moving to private-services-access networking means a new instance and a client cutover.
`redis_version` (REDIS_7_0) upgrades forward but never back. Both are worth a deliberate
glance before the first byte is cached, because both outlive every other setting.

## The alias is the contract

The Kubernetes ExternalName service is what applications should connect to: a stable
in-cluster name pointing at the instance's endpoint, with the firewall rule admitting the
traffic. Re-provisioning, resizing, or replacing the instance then becomes an alias update:
zero application config changes, which is precisely what makes the fixed-at-creation
settings above survivable.