[tdprajwal]

What

Adds a registrycache plugin that keeps every active registry record in Redis for 24h, reloaded daily, and makes signature lookup, the agreement check and provider routing read through it. Supersedes #123.

Changes to existing records (a withdrawn key, an inactive participant, a changed baseUrl) can take up to the TTL to apply. New records apply immediately, and a refusal is rechecked with the registry.

Why

Most requests no longer need a registry call: two loads per day instead of one call per request.

Testing

Unit tests. On quick-start, an AgricultureFacility select was served end to end with the registry stopped.

Closes #109

[github-actions]

📊 Test Coverage: ✅ Passed — 86% of changed lines covered, min 80%

[github-actions]

🛡️ Trivy security scan (CRITICAL,HIGH,MEDIUM,LOW)

View full run

Go dependencies

No findings at CRITICAL,HIGH,MEDIUM,LOW.

Container image

No findings at CRITICAL,HIGH,MEDIUM,LOW.

[tdprajwal]

Self-review:

• A refusal from a cached copy is rechecked with the registry (signature lookup, routing, agreements); a refusal the registry gave is not asked again.
• A record not in the cache (inactive, new, unknown) or unreadable there is asked of the registry, so an inactive partner network is still refused.
• Only the background load reads the whole entity or writes the list. A record found missing or out of date asks for a refresh, within a minute.
• The list key includes the registry URL, so registries sharing a Redis keep apart. The reload runs a check before the list expires.
• ttl and preload are required. Removed the unused Get.
• Known, not changed: each lookup reads the entity list from Redis; the full-entity load is not paged (large registries fall back to per-record searches); the plugin repeats the registry url and has its own search code (follow-up); each module loads its own lists.

Retested on quick-start: a select served end to end, also with the registry stopped; lists deleted from Redis were reloaded within a minute.