Remove garage files at bad location, add basic telemetry

1 files changed, 49 insertions, 0 deletions

diff --git a/cluster/prod/app/telemetry/deploy/telemetry-system.hcl b/cluster/prod/app/telemetry/deploy/telemetry-system.hcl
new file mode 100644
index 0000000..e4bde1a
--- /dev/null
+++ b/cluster/prod/app/telemetry/deploy/telemetry-system.hcl
@@ -0,0 +1,49 @@
+job "telemetry-system" {
+	datacenters = ["neptune", "orion"]
+	type = "system"
+	priority = "100"
+
+	group "collector" {
+    network {
+      port "node_exporter" { static = 9100 }
+    }
+
+		task "node_exporter" {
+			driver = "docker"
+
+			config {
+				image = "quay.io/prometheus/node-exporter:v1.1.2"
+				network_mode = "host"
+				volumes = [
+					"/:/host:ro,rslave"
+				]
+				args = [ "--path.rootfs=/host" ]
+			}
+
+			resources {
+				cpu = 50
+				memory = 40
+			}
+
+      service {
+        tags = [ "telemetry" ]
+        port = 9100
+        address_mode = "driver"
+        name = "node-exporter"
+        check {
+          type = "http"
+          path = "/"
+          port = 9100
+          address_mode = "driver"
+          interval = "60s"
+          timeout = "5s"
+          check_restart {
+            limit = 3
+            grace = "90s"
+            ignore_warnings = false
+          }
+        }
+      }
+		}
+	}
+}