# Probe: smoke
#
# Per-service L1+L2 health check across every showcase service on
# Railway. Discovery-driven: the `railway-services` source enumerates
# every `showcase-*` service in the orchestrator's project, and the
# driver fans out one invocation per service.
#
# Each driver invocation issues THREE HTTP calls:
#   1. GET  `${publicUrl}/smoke`            → primary `smoke:<slug>` tick
#   2. GET  `${publicUrl}/health`           → side-emit `health:<slug>`
#   3. POST `${publicUrl}/api/copilotkit/`  → side-emit `agent:<slug>`
#
# (1) and (2) are L1: liveness + health. (3) is L2: the CopilotKit
# runtime is mounted and answering — any non-404 response is green;
# 404 or transport failure is red. Matches the `checkAgentEndpoint`
# contract in the e2e helpers.
#
# `slug` is derived by stripping the `showcase-` prefix from the Railway
# service name. So `showcase-ag2` → keys `smoke:ag2` / `health:ag2` /
# `agent:ag2`. Each package service produces 3 independent probe
# rows — each can flip green/red without blinding its siblings.
#
# Auto-discovery: adding a new `showcase-*` service on Railway
# automatically picks it up on the next tick — no YAML edit required.
# Only infra services (aimock/harness/pocketbase/shell*) are excluded,
# because they don't expose a CopilotKit runtime or health endpoints.
# If a new non-monitored service is created with the `showcase-`
# prefix, extend `filter.nameExcludes` here.
#
# Schedule mirrors the pre-YAML cron cadence (every 15 minutes) — the
# accompanying alert rule in config/alerts/ covers green→red /
# red→green transitions and a cron-only heartbeat tick so silent
# green runs still produce observable liveness.
#
# timeout_ms (30s) bounds BOTH the discovery enumerate phase AND each
# per-target driver invocation at the invoker level. The driver also
# re-applies the timeout inside each HTTP call so one hung endpoint
# can't steal budget from its paired probes.
#
# This was originally 10s when the probe used a static target list.
# After migrating to discovery-driven (railway-services), the same
# timeout_ms governs the Railway enumerate call — a project-level
# GraphQL query PLUS per-service variable lookups (bounded at
# concurrency 8). With 35+ services, each variable round-trip ~1s,
# enumerate alone takes ceil(35/8) × 1s ≈ 5s on a good tick, and
# easily >10s on a slow one. The 10s cap caused persistent
# discoveryFailed:true — enumerate timed out before reaching the
# driver fan-out. 30s matches image-drift and qa (both Railway-
# discovery probes that work reliably at this fleet size).
#
# max_concurrency (6) caps simultaneous per-tick driver invocations —
# at 34 services the tick takes ≤ 6 pool-cycles and stays well under
# any Railway edge rate limit. Per-tick socket bound:
# max_concurrency × 3 endpoints = 18 inflight.
kind: smoke
id: smoke
schedule: "*/5 * * * *"
timeout_ms: 30000
max_concurrency: 6
discovery:
  source: railway-services
  filter:
    namePrefix: "showcase-"
    # Infra services that don't expose /smoke, /health, or the
    # CopilotKit runtime. Keep this list aligned with the excludes in
    # `src/probes/aimock-wiring.ts` — when a new infra service lands
    # on Railway under the `showcase-` prefix, add it in both places.
    nameExcludes:
      - "showcase-aimock"
      - "showcase-harness"
      - "showcase-pocketbase"
      - "showcase-shell"
      - "showcase-shell-dashboard"
      - "showcase-shell-docs"
      - "showcase-shell-dojo"
      # Decommissioned starters — Railway services stopped, deployments
      # halted (PR #4390). Excluded so probes don't scan dead endpoints.
      - "showcase-starter-ag2"
      - "showcase-starter-agno"
      - "showcase-starter-claude-sdk-python"
      - "showcase-starter-claude-sdk-typescript"
      - "showcase-starter-crewai-crews"
      - "showcase-starter-google-adk"
      - "showcase-starter-langgraph-fastapi"
      - "showcase-starter-langgraph-python"
      - "showcase-starter-langgraph-typescript"
      - "showcase-starter-langroid"
      - "showcase-starter-llamaindex"
      - "showcase-starter-mastra"
      - "showcase-starter-ms-agent-dotnet"
      - "showcase-starter-ms-agent-python"
      - "showcase-starter-pydantic-ai"
      - "showcase-starter-spring-ai"
      - "showcase-starter-strands"
  # key_template is interpolated against each RailwayServiceInfo record.
  # `${name}` is the full Railway service name (`showcase-ag2`). The
  # driver re-derives the slug by stripping the `showcase-` prefix, so
  # the writer sees `smoke:showcase-ag2` as the top-line key and
  # `health:ag2` / `agent:ag2` as the side-emit keys. We keep the
  # `smoke:` prefix carrying the full service name so existing alert
  # rules that dedupe on `smoke:*` continue to match.
  key_template: "smoke:${name}"