computer-playbook/roles/web-app-prometheus/files/nginx/metricz.conf

  # Central nginx metrics endpoint — scraped ONCE by Prometheus from this domain.
  # Exposes HTTP request counts, latency histograms, bytes in/out, TLS handshakes,
  # and stack_up gauge for all vhosts via lua-resty-prometheus shared memory.
  # All metrics carry 'host' and 'app' labels so one scrape covers all 70+ apps.
  # NOT included in other app vhosts — only the prometheus domain exposes this.
  location = /metricz {
    access_log off;
    content_by_lua_block {
      -- Mirror the /healthz/live logic: a container is "up" only when BOTH
      -- Docker says it is healthy AND the HTTP backend is reachable.
      -- Stack is up only when both Docker health and HTTP probe pass.
      local up = 1

      -- Step 1: Docker HEALTHCHECK state (same source as /healthz/live).
      local cname = ngx.var.container_name
      if cname and cname ~= "" then
        local dict = ngx.shared.health_containers
        if dict then
          local status = dict:get(cname)
          if status == "down" then
            up = 0
          end
        end
      end

      -- Step 2: HTTP reachability.
      if up == 1 then
        local res = ngx.location.capture("/_live_probe")
        if not res or res.status >= 500 then
          up = 0
        end
      end

      metric_stack_up:set(up, {ngx.var.host, ngx.var.app_id})
      prometheus:collect()
    }
  }