Choosing services

Cache

In-process memory, PostgreSQL or Redis / Valkey (ElastiCache, Azure Cache for Redis, Memorystore) for rate limits and invalidations shared by every replica.

The cache store holds small, short-lived state that every app instance has to agree on: rate limits (API keys, the OAuth token endpoint, sign-in and reset, agent runs, signups), invalidations (a revoked key or token, a changed setting) and idempotency claims for task ingestion. None of it is the source of truth: losing the store resets rate-limit windows and makes caches reload, nothing more.

Option	`CACHE_PROVIDER`	On-prem	AWS	Azure	Google Cloud	Switch
In-process memory	`memory` (default)	yes	yes	yes	yes	config
PostgreSQL (UNLOGGED table in the app database)	`postgres`	yes	yes	yes	yes	config
Redis, Valkey, KeyDB, Dragonfly	`redis`	yes	native: ElastiCache / MemoryDB with IAM auth (`CACHE_REDIS_AUTH=aws_iam`), or an AUTH token	native: Azure Cache for Redis / Azure Managed Redis with Entra ID	native: Memorystore (IAM for Valkey)	config

When to pick which

Memory for a single app instance.
Postgres as soon as there is more than one instance (or a separate worker). There's nothing extra to run, and it's the right default for most installs.
Redis / Valkey when you already run one, or at high request rates.

Rules

With memory, each replica counts on its own: N replicas allow up to N times every limit, and other replicas see a revoked credential only when their credential cache expires (60 seconds).
Redis Cluster mode is not supported. Use a non-clustered (primary and replica) instance.
If a shared store is unreachable, rate limits fall back to per-instance counting and credentials are re-checked against the database.
Several installs can share one Redis with different CACHE_KEY_PREFIX values.

bash

usertasks-admin switch cache --to postgres --yes
usertasks-admin switch cache --to redis --set redis_url=rediss://cache.internal:6379 --yes

Reference: Cache.