Goldpath.Caching 0.1.0-preview.8

Goldpath.Caching

Ring B caching: Microsoft HybridCache as the app surface (L1+L2, stampede-protected, tag invalidation) and Mediant [Cacheable]/[InvalidatesCache] as the query surface — composed, not wrapped. Goldpath adds the manifest-driven tier wiring and the tenant-scoped key convention.

Getting started

builder.AddGoldpathCaching();                 // L1 + Redis L2 (ConnectionStrings:redis)

// App surface — inject Microsoft's own type, no Goldpath wrapper:
public sealed class RateService(HybridCache cache, GoldpathCacheKeys keys)
{
    public async ValueTask<Rates> GetAsync(CancellationToken ct) =>
        await cache.GetOrCreateAsync(keys.For("rates", "current"),
            _ => FetchAsync(ct), tags: ["rates"], cancellationToken: ct);
}

// Query surface — Mediant's own attributes:
[Cacheable(300, CacheKeyPrefix = "rates")]
public sealed record GetRatesQuery : IQuery<Rates>;

Configuration

"Goldpath": {
  "DistributedCaching": {
    "Levels": ["l1", "l2"],          // ["l1"] = in-process only, no Redis dependency
    "RedisConnectionName": "redis",  // resolved from ConnectionStrings
    "DefaultTtlSeconds": 300
  }
}

Manifest: features.distributedCaching: { levels, redis.connectionName, defaultTtl }. In ["l1"] mode both surfaces stay fully in-process (memory distributed cache backs the query path) — unit tests and single-instance deployments need no Redis.

Advanced

• Key convention (GoldpathCacheKeys): goldpath:{tenant}:{area}:{key} — tenant scoping is baked in BEFORE the MultiTenancy module lands, because retrofitting a key format means invalidating production caches, and cross-tenant cache bleed is a security bug. Inject GoldpathCacheKeys for ambient-tenant keys; GoldpathCacheKeys.Compose(tenant, area, key) for background/message flows.
• Invalidation vocabulary: one area name across both surfaces — HybridCache tags on the app surface (RemoveByTagAsync("rates") works TODAY), CacheKeyPrefix on the query surface.
• The composition loop, closed: the gaps this module filed upstream shipped in Mediant 1.2.0 and are composed here — the query path rides the SAME HybridCache (L1+L2 on [Cacheable] hits, mediant#130) and [InvalidatesCache] really invalidates, tag-based (mediant#131). The pinned no-op test broke on the version bump exactly as designed and now asserts the real semantics.
• Stampede: both surfaces protected (HybridCache built-in; Mediant lock pool).

Providers

L2 = Redis (StackExchange; wire-compatible with Valkey/Garnet). Other L2 stores deferred until a project needs one (RFC D4) — YAGNI over a provider zoo.