Aller au contenu

Sondes et métriques

Généré depuis src/http/healthcheck.cr, src/event/loss.cr, src/event/consumer_health.cr et src/metrics.cr par make gen-docs-health-metrics. Ne modifiez pas ce fichier à la main.

Cette page sert à lire une réponse de sonde ou à nommer une métrique dans vos tableaux de bord et vos alertes.

Le serveur expose deux sondes et un point de collecte :

  • GET /Healthcheck vérifie la préparation.
  • GET /Healthcheck/live vérifie la vivacité.
  • GET /metrics sert le format d’exposition texte Prometheus 0.0.4.

Les deux sondes répondent sans authentification. La sonde de préparation répond HTTP 503 quand une dépendance est injoignable ou qu’un consommateur de flux est dégradé.

Tout ce que renvoie GET /Healthcheck. Un chemin [] est une entrée du champ tableau : la sonde renvoie une entrée de pool par pool de base de données et une entrée de consommateur par groupe de consommateurs de flux. La sonde de vivacité ne renvoie que les deux champs live.*.

Champ Signification
database.error The first pool error, when a pool is broken.
database.pools[].database The database entry name from the route table.
database.pools[].error The error this pool raised, when it is broken.
database.pools[].role The pool role: rw or ro.
database.pools[].status ok when this pool answers the SELECT 1 probe.
database.status ok when every pool answers the SELECT 1 probe.
event_lag.max_lag Largest consumer-group lag the last lag-check pass observed.
event_loss.messages_dropped Dispatched messages dropped by handler errors since process start.
event_loss.publish_events_dropped Events dropped by failed publish flushes since process start.
event_loss.publish_flush_failures Failed publish flushes since process start.
events.consumers[].consecutive_failures Failed loop turns in a row for this consumer.
events.consumers[].last_error The last error message this consumer reported.
events.consumers[].last_success_at RFC-3339 time of the consumer’s last successful loop turn.
events.consumers[].name The stream consumer group name.
events.consumers[].restarts Supervised restarts of this consumer since process start.
events.status ok, in_process under the in-process manager, or degraded when a consumer reports consecutive failures.
redis.error The Redis error, when the probe fails.
redis.status ok when Redis answers, in_process under the in-process manager, error when unreachable.
service.env The APP_ENV the process boots in.
service.instance_id One instance id per boot; matches the boot log and the shard-lock holders in Redis.
status ok when every dependency passes, degraded otherwise; HTTP 200 answers ok, HTTP 503 answers degraded.
live.status Always ok while the process answers.
live.uptime_seconds Seconds since the process started.

Chaque nom de métrique émis par le serveur, avec son type d’exposition. Le même registre rend aussi tout nom non catalogué comme un compteur avec une ligne d’aide générique ; le catalogue ci-dessous est l’ensemble documenté complet.

Nom Type Signification
facet_consumer_consecutive_failures gauge Consecutive failed loop turns per stream consumer group.
facet_consumer_restarts_total counter Supervised fiber restarts per stream consumer group since process start.
facet_cron_task_failures_total counter Cron task firings that raised, per task id.
facet_cron_task_lease_skips_total counter Cron task dispatches skipped because another instance holds the tick lease, per task id.
facet_db_checkout_total counter Pooled connection checkouts, by database entry and pool role (rw/ro).
facet_event_lag_max gauge Maximum consumer-group lag observed by the last Cron::EventLagCheck pass.
facet_event_loss_messages_dropped_total counter Dispatched messages swallowed by handler errors since process start.
facet_event_loss_publish_events_dropped_total counter Events dropped by failed publish flushes since process start.
facet_event_loss_publish_flush_failures_total counter Failed Event::Publisher.flush calls since process start.
facet_events_published_total counter Resource events written to the Redis streams by this process.
facet_fiber_loop_errors_total counter Long-lived loop turns that rescued an exception, by fiber name.
facet_fiber_restarts_total counter Supervised fiber restarts, by fiber name, since process start.
facet_http_requests_total counter HTTP requests handled, by method, route group, and status class.
facet_jobs_abandoned_total counter Background job rows requeued by the expired-lease sweep, per kind.
facet_jobs_claimed_total counter Background job rows claimed by the worker, per kind.
facet_jobs_enqueued_total counter Background job rows inserted, per kind.
facet_jobs_failed_total counter Background job attempts written as failed, per kind.
facet_jobs_succeeded_total counter Background job attempts written as succeeded, per kind.
facet_messages_processed_total counter Stream messages dispatched to a subscription handler, per consumer group.