Cache

Tunables for the L1 / L2 cache strands live under heimdall.cache. This page is the operator view (what to set); see Architecture / Caching for the design story.

Schema

heimdall:
  cache:
    l1:
      maxEntries: 50000
      defaultTtl: "00:05:00"
      provider: "memory"
    l2:
      maxEntries: 50000
      defaultTtl: "00:05:00"
      provider: "none"

L1 — in-process

l1 is the in-memory strand backed by Microsoft.Extensions.Caching.Hybrid. It always exists and is consulted on every request.

Key	Default	Notes
provider	memory	The MVP has no other backends here.
maxEntries	50000	Soft cap.
defaultTtl	00:05:00	Applied when an entry is stored without an explicit TTL.

The TTL is a hint; HybridCache may evict earlier under memory pressure.

L2 — distributed (stub)

The L2 strand is in place but unwired in the MVP — provider: "none" is the default and the only supported value. The DI registration uses an in-memory IDistributedCache stub so the HybridCache contract still works in tests and integration, but multiple Heimdall instances share nothing.

When Redis lands, configure it here:

heimdall:
  cache:
    l2:
      provider: "redis"
      defaultTtl: "00:30:00"

Until then, treat L2 settings as forward-compatible placeholders.

Per-feed TTL override

Each feed in heimdall.ecosystems.nuget.feeds may define its own cacheTtl to override the strand default for that feed's registration documents. See Feeds.