Add health checks and restarts to prometheus

core/prometheus.nomad

@@ -37,12 +37,36 @@ job "prometheus" {
         "traefik.enable=true",
         "traefik.http.routers.prometheus.entryPoints=websecure",
       ]
+
+      check {
+        type = "http"
+        path = "/-/healthy"
+        interval = "10s"
+        timeout = "3s"
+
+        check_restart {
+          limit = 3
+          grace = "5m"
+        }
+      }
     }
 
     service {
       name = "pushgateway"
       provider = "nomad"
       port = "pushgateway"
+
+      check {
+        type = "http"
+        path = "/-/healthy"
+        interval = "10s"
+        timeout = "3s"
+
+        check_restart {
+          limit = 3
+          grace = "5m"
+        }
+      }
     }
 
     task "prometheus" {