Add health checks and restarts to prometheus

2024-05-30 15:01:42 -07:00 · 2024-05-30 15:01:42 -07:00 · 7b41d29eb8
commit 7b41d29eb8
parent 90b7740343
1 changed files with 24 additions and 0 deletions
--- a/core/prometheus.nomad
+++ b/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" {