Module runtime 

Process-wide holder of the cache’s single shared, multiplexed backend and the RedisCacheStore over it — Rust port of the Java CacheRuntime. Every v1.cache.redis worker instance calls store and shares this one connection: redis.cache.instances is worker concurrency, not a connection count (design spec §4.4: no pool; the client pipelines over one in-order connection).

The backend is built lazily on first use, re-resolving configuration on every attempt until it connects. This is deliberate (spec §6): the function is registered before a credential-bootstrap application publishes a vault password, and — unlike sync-over-async, which must keep an eager Pub/Sub subscriber live — a cache has nothing to maintain at start-up and must not fail application start-up when Redis is briefly unreachable. While the connection cannot be built the store stays empty and each call retries (a late credential is picked up); once built, the client owns reconnection under it and the store is reused.

The one connection is released on shutdown through the platform’s lifecycle (Platform::on_shutdown), registered from the build so the cleanup is wired only when a connection has actually been opened.

Functions

current

The store if one has been built (diagnostics; None before the first successful call and after shutdown).

set

Install a store as the process-wide instance — the reuse/test seam (a store built against an embedded or in-process server). Replaces any current one.

shutdown

Release the shared backend (idempotent) — registered with Platform::on_shutdown when the connection opens. Dropping the last handle closes the connection; a later call rebuilds from live configuration.

store

The shared store, built on first use and reused thereafter. If the connection cannot be built yet, the error propagates to the caller (fail-fast — a cache failure is the caller’s concern via the flow’s exception handler) and the next call retries with freshly resolved configuration.