feat(monitoring): metricas dos tenants vendidos, e o alerta de fila do Asaas pausada

moldes/consolidado/70-runtime-core.yaml

@@ -58,6 +58,9 @@ kind: Service
 metadata:
   name: runtime-core
   namespace: {{TENANT}}-prod
+  labels:
+    app: runtime-core
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: runtime-core }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

moldes/consolidado/71-runtime-operacao.yaml

@@ -83,6 +83,9 @@ kind: Service
 metadata:
   name: runtime-operacao
   namespace: {{TENANT}}-prod
+  labels:
+    app: runtime-operacao
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: runtime-operacao }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

moldes/consolidado/72-runtime-saude.yaml

@@ -61,6 +61,9 @@ kind: Service
 metadata:
   name: runtime-saude
   namespace: {{TENANT}}-prod
+  labels:
+    app: runtime-saude
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: runtime-saude }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

moldes/consolidado/90-servico-bff.yaml

@@ -84,6 +84,9 @@ kind: Service
 metadata:
   name: servico-bff
   namespace: {{TENANT}}-prod
+  labels:
+    app: servico-bff
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: servico-bff }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

platform/monitoring/athleticmap-rules.yaml

@@ -191,3 +191,21 @@ spec:
           annotations:
             summary: "{{ $labels.deployment }} sem replica disponivel — demo, piloto e acme dependem dele"
             description: "Os tres tenants migrados param juntos. kubectl -n plataforma-prod get pods -l app={{ $labels.deployment }} e, se for o PgBouncer, lembre que a userlist e montada por subPath e so entra em vigor com restart."
+    - name: athleticmap.asaas
+      rules:
+        # Depois de 15 falhas seguidas o Asaas PAUSA a fila de avisos: os
+        # eventos ficam guardados 14 dias e depois somem. Nada quebra do nosso
+        # lado — so para de chegar pagamento. As vigias ja tentam reativar
+        # (no maximo 1x a cada 6 h); se ainda estiver pausada 30 min depois, a
+        # causa persiste e alguem tem de olhar o endpoint.
+        - alert: AthleticMapFilaDoAsaasPausada
+          expr: max by (namespace, origem) (atm_asaas_filas_pausadas) > 0
+          for: 30m
+          labels:
+            severity: critical
+          annotations:
+            summary: "Fila de avisos do Asaas pausada em {{ $labels.namespace }}"
+            description: >-
+              O Asaas parou de mandar avisos ({{ $labels.origem }}). Os guardados
+              somem em 14 dias. Veja o log (fila de avisos do Asaas PAUSADA) e o
+              endpoint de webhook; reativar: Integracoes > Webhooks > Remover penalidade.

platform/monitoring/tenants-servicemonitor.yaml

@@ -0,0 +1,30 @@
+# =============================================================================
+# As metricas dos tenants VENDIDOS — os que nascem do molde.
+#
+# Por ROTULO, e nao por nome de namespace: o `athleticmap-runtimes` lista
+# demo/piloto/acme pelo nome, e todo tenant novo ficava de fora (medido em
+# 2026-09-26: teste5 e zico-0 sem nenhuma metrica). O rotulo vem do molde.
+# Ver scripts/metricas_dos_tenants_e_filas.py.
+# =============================================================================
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: athleticmap-tenants
+  namespace: monitoring
+  labels:
+    release: monitoring
+spec:
+  namespaceSelector:
+    any: true
+  selector:
+    matchLabels:
+      athleticmap.io/metricas: tenant
+  endpoints:
+    - port: http
+      path: /actuator/prometheus
+      interval: 60s
+      scrapeTimeout: 20s
+      metricRelabelings:
+        - sourceLabels: [__name__]
+          regex: "atm_.*|up|jvm_memory_used_bytes|jvm_memory_max_bytes|jvm_threads_live_threads|process_cpu_usage|system_cpu_usage|http_server_requests_seconds_count|http_server_requests_seconds_sum|http_server_requests_seconds_max|hikaricp_connections_active|hikaricp_connections_pending|hikaricp_connections_max|flyway_migrations_.*|application_ready_time_seconds"
+          action: keep

tenants/escolinha-de-teste-2/70-runtime-core.yaml

@@ -58,6 +58,9 @@ kind: Service
 metadata:
   name: runtime-core
   namespace: escolinha-de-teste-2-prod
+  labels:
+    app: runtime-core
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: runtime-core }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

tenants/escolinha-de-teste-2/71-runtime-operacao.yaml

@@ -83,6 +83,9 @@ kind: Service
 metadata:
   name: runtime-operacao
   namespace: escolinha-de-teste-2-prod
+  labels:
+    app: runtime-operacao
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: runtime-operacao }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

tenants/escolinha-de-teste-2/72-runtime-saude.yaml

@@ -61,6 +61,9 @@ kind: Service
 metadata:
   name: runtime-saude
   namespace: escolinha-de-teste-2-prod
+  labels:
+    app: runtime-saude
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: runtime-saude }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

tenants/escolinha-de-teste-2/90-servico-bff.yaml

@@ -84,6 +84,9 @@ kind: Service
 metadata:
   name: servico-bff
   namespace: escolinha-de-teste-2-prod
+  labels:
+    app: servico-bff
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: servico-bff }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

tenants/escolinha-de-teste/70-runtime-core.yaml

@@ -58,6 +58,9 @@ kind: Service
 metadata:
   name: runtime-core
   namespace: escolinha-de-teste-prod
+  labels:
+    app: runtime-core
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: runtime-core }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

tenants/escolinha-de-teste/71-runtime-operacao.yaml

@@ -83,6 +83,9 @@ kind: Service
 metadata:
   name: runtime-operacao
   namespace: escolinha-de-teste-prod
+  labels:
+    app: runtime-operacao
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: runtime-operacao }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

tenants/escolinha-de-teste/72-runtime-saude.yaml

@@ -61,6 +61,9 @@ kind: Service
 metadata:
   name: runtime-saude
   namespace: escolinha-de-teste-prod
+  labels:
+    app: runtime-saude
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: runtime-saude }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

tenants/escolinha-de-teste/90-servico-bff.yaml

@@ -84,6 +84,9 @@ kind: Service
 metadata:
   name: servico-bff
   namespace: escolinha-de-teste-prod
+  labels:
+    app: servico-bff
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: servico-bff }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

tenants/teste5/70-runtime-core.yaml

@@ -58,6 +58,9 @@ kind: Service
 metadata:
   name: runtime-core
   namespace: teste5-prod
+  labels:
+    app: runtime-core
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: runtime-core }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

tenants/teste5/71-runtime-operacao.yaml

@@ -83,6 +83,9 @@ kind: Service
 metadata:
   name: runtime-operacao
   namespace: teste5-prod
+  labels:
+    app: runtime-operacao
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: runtime-operacao }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

tenants/teste5/72-runtime-saude.yaml

@@ -61,6 +61,9 @@ kind: Service
 metadata:
   name: runtime-saude
   namespace: teste5-prod
+  labels:
+    app: runtime-saude
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: runtime-saude }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

tenants/teste5/90-servico-bff.yaml

@@ -84,6 +84,9 @@ kind: Service
 metadata:
   name: servico-bff
   namespace: teste5-prod
+  labels:
+    app: servico-bff
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: servico-bff }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

tenants/zico-0/70-runtime-core.yaml

@@ -58,6 +58,9 @@ kind: Service
 metadata:
   name: runtime-core
   namespace: zico-0-prod
+  labels:
+    app: runtime-core
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: runtime-core }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

tenants/zico-0/71-runtime-operacao.yaml

@@ -83,6 +83,9 @@ kind: Service
 metadata:
   name: runtime-operacao
   namespace: zico-0-prod
+  labels:
+    app: runtime-operacao
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: runtime-operacao }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

tenants/zico-0/72-runtime-saude.yaml

@@ -61,6 +61,9 @@ kind: Service
 metadata:
   name: runtime-saude
   namespace: zico-0-prod
+  labels:
+    app: runtime-saude
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: runtime-saude }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]

tenants/zico-0/90-servico-bff.yaml

@@ -84,6 +84,9 @@ kind: Service
 metadata:
   name: servico-bff
   namespace: zico-0-prod
+  labels:
+    app: servico-bff
+    athleticmap.io/metricas: tenant
 spec:
   selector: { app: servico-bff }
-  ports: [{ port: 80, targetPort: 8083 }]
+  ports: [{ name: http, port: 80, targetPort: 8083 }]