Health, Readiness & Metrics

Three routes, three questions, and the reason they are three: an operator who cannot tell them apart either restarts a process that was serving fine or keeps routing to one that has been stale since Tuesday.

/healthz — liveness

Always 200. Configuration has no say in it.

A configuration that will not reload is a reason to stop routing to a process, never a reason to have it killed: the restart reads the same broken file, fails to start at all, and turns a degraded service into an outage.

/readyz — readiness

{
  "status": "degraded",
  "configs": {
    "db": {
      "installed": true,
      "generation": 4,
      "healthy": false,
      "consecutive_failures": 3,
      "stale_for": 812.4,
      "last_reason": "file changed",
      "last_failure": {"kind": "invalid", "path": "pool.max_size", "seconds_ago": 12.1},
      "ready": false
    }
  },
  "problems": ["db: 3 reloads refused"]
}

Three conditions, in the order they matter:

stale_after is off by default: a configuration nobody edits is not a problem, and a service whose settings change quarterly should not page anybody at midnight.

A group reports every member by key, and the worst member decides the code — so a page says which configuration, which is the difference between an alert somebody can act on and one they have to investigate.

/metrics — the engine's own series

from dynamic_config_web import CONTENT_TYPE, metrics_body

body = metrics_body(group)          # or metrics_body(db, cache)

Twelve series per configuration that has a remote store, six for one that does not, with the names the engine defines — they end up in dashboards and alert rules, so nothing here renames or filters them. Built per scrape and thrown away; the numbers are atomic loads, so a handler needs no cache in front of it.

metrics_body takes a ConfigGroup as well as configurations, which the engine's Exposition does not: it iterates the members and labels each with its own key.

What none of them carries

A value. Generations, counts, kinds, paths, seconds and reasons — nothing out of the document, not even in last_failure, whose message routinely would. /healthz and /readyz are unauthenticated in most deployments and /metrics is scraped by something that keeps everything for a year.

Values live in explain and check, behind a guard, on the next page.