Fixes #XXXXX - Add cache logging, health endpoint, and concurrency limiter to registration module

[pablomh]

Summary

Builds on smart-proxy#935 (in-memory registration script cache) with:

• Cache HIT/MISS logging — read_registration_cache logs registration_script cache=HIT age=Xs or cache=MISS, with source=shared|local when Redis is in use
• GET /register/health endpoint — returns 200 OK if the capsule can reach Foreman's /api/status, 503 if not; used by HAProxy health checks to route registration traffic away from degraded capsules
• with_concurrency_limit(&block) — block-based abstraction that wraps the POST / handler; owns semaphore acquire/release and returns 503 Retry-After: 30 when the limit is exhausted. The handler expresses what to do; the helper owns how to limit concurrency.
• Shared Redis/Valkey cache — optional :cache_url setting (redis://host:6379/0); when set, all capsule nodes in an LB pool share one cache so a single warm request benefits all nodes. Falls back to in-memory cache if Redis is unreachable.
• :max_concurrent_registrations setting — configurable via registration.yml; default unlimited (backward compatible)

Depends on

• smart-proxy#935 — in-memory registration script cache (base for this PR)

Design notes

with_concurrency_limit(&block) — top-down design

The POST / handler now reads as a statement of intent:

post '/' do
  with_concurrency_limit do
    resp = Proxy::Registration::ProxyRequest.new.host_register(request)
    handle_response(resp)
  end
end

The mechanical concern (semaphore acquire/release, 503 response, Retry-After header) lives entirely in with_concurrency_limit. The handler has no knowledge of how concurrency limiting works.

Shared cache architecture

One Redis/Valkey instance on the LB host, shared by all capsule nodes in the pool. Each capsule connects over the private network. If Redis is unreachable, the capsule falls back to its in-memory cache — registration continues working with per-node warm-up.

# /etc/foreman-proxy/settings.d/registration.yml
:cache_url: redis://lb-host-private-ip:6379/0
:max_concurrent_registrations: 50

Companion PRs

Repo	PR	What
smart-proxy	#935	In-memory cache (base)
smart-proxy	this PR	HIT/MISS logging, health endpoint, concurrency limiter, shared cache
satperf	registration-metrics	registration_cache Ansible role + HAProxy health check config

Test plan

• ruby test/registration/registration_api_test.rb
• Verify cache=HIT / cache=MISS lines appear in proxy log
• GET /register/health returns 200 when Foreman is reachable, 503 when not
• POST / returns 503 with Retry-After: 30 when semaphore is exhausted
• With :cache_url set, verify Redis keyspace_hits increments on cache hits
• Without :cache_url, verify fallback to in-memory cache works

🤖 Generated with Claude Code