Caching

Query results can be cached in Redis. Install ioredis and give the cache a URL:

InverseModelModule.forRoot({
  url: process.env.DATABASE_URL!,
  cache: {
    url: process.env.REDIS_URL!,
    ttl: 60, // seconds
    ttlByTable: { audit_logs: 0, countries: 3600 },
  },
});

Option	Default	What it does
cache.url	(no cache)	Redis connection string.
cache.ttl	60	Seconds a result is kept.
cache.ttlByTable	{}	TTL per table or collection name; 0 leaves a table uncached.
cache.keyPrefix	inverse:	Prefix of every key the cache writes.

What is cached

Lists, counts, lookups by key, connections and relation batches. Each result is keyed by its table, its arguments and, when scope is set, the scope, so tenants never share results.

Invalidation

A mutation clears the cached results of the table it wrote to, without touching other tables. Inside a transaction, tables are cleared once it is over, never before it commits, so no reader caches what was there before.

Each table has a version number in Redis that is part of every key, and a write increments it. Old results are never read again and expire on their own, so invalidation is a single Redis command however many results a table has.

A query that filters by a relation is also cleared by writes to the related table.

When Redis is down

Reads go to the database and a warning is logged. The API keeps working, just without the cache.

Writes from outside the API

The cache only knows about writes made through the API. If other systems write to the same tables, use a short ttl for those tables, or 0 to leave them uncached.