staging: telemetry: add smartctl exporter

cluster/staging/app/telemetry/config/prometheus.yml

@@ -16,6 +16,17 @@ scrape_configs:
         cert_file: /etc/prometheus/consul-client.crt
         key_file: /etc/prometheus/consul-client.key
 
+  - job_name: 'smartctl-exporter'
+    scrape_interval: 60s
+    consul_sd_configs:
+    - server: 'https://localhost:8501'
+      services:
+        - 'smartctl-exporter'
+      tls_config:
+        ca_file: /etc/prometheus/consul.crt
+        cert_file: /etc/prometheus/consul-client.crt
+        key_file: /etc/prometheus/consul-client.key
+
   - job_name: 'garage'
     authorization:
       type: Bearer

cluster/staging/app/telemetry/deploy/telemetry-system.hcl

@@ -3,47 +3,92 @@ job "telemetry-system" {
   type = "system"
   priority = "100"
 
-  group "collector" {
+  group "node_exporter" {
-     network {
+    network {
-       port "node_exporter" { static = 9100 }
+      port "node_exporter" { static = 9100 }
-     }
+    }
 
-		task "node_exporter" {
+    task "node_exporter" {
-			driver = "docker"
+      driver = "docker"
 
-			config {
+      config {
-				image = "quay.io/prometheus/node-exporter:v1.8.1"
+        image = "quay.io/prometheus/node-exporter:v1.8.1"
-				network_mode = "host"
+        network_mode = "host"
-				volumes = [
+        volumes = [
-					"/:/host:ro,rslave"
+          "/:/host:ro,rslave"
-				]
+        ]
-				args = [ "--path.rootfs=/host" ]
+        args = [ "--path.rootfs=/host" ]
-			}
+      }
 
-			resources {
+      resources {
-				cpu = 50
+        cpu = 50
-				memory = 40
+        memory = 40
-			}
+      }
 
-       service {
+      service {
-         tags = [ "telemetry" ]
+        tags = [ "telemetry" ]
-         port = 9100
+        port = 9100
-         address_mode = "driver"
+        address_mode = "driver"
-         name = "node-exporter"
+        name = "node-exporter"
-         check {
+        check {
-           type = "http"
+          type = "http"
-           path = "/"
+          path = "/"
-           port = 9100
+          port = 9100
-           address_mode = "driver"
+          address_mode = "driver"
-           interval = "60s"
+          interval = "60s"
-           timeout = "5s"
+          timeout = "5s"
-           check_restart {
+          check_restart {
-             limit = 3
+            limit = 3
-             grace = "90s"
+            grace = "90s"
-             ignore_warnings = false
+            ignore_warnings = false
-           }
+          }
-         }
+        }
-       }
+      }
-		}
+    }
-	}
+  }
- }
+
+  group "smartctl_exporter" {
+    network {
+      port "smartctl_exporter" { static = 9101 }
+    }
+
+    task "smartctl_exporter" {
+      driver = "docker"
+      user = "root"
+
+      config {
+        image = "prometheuscommunity/smartctl-exporter:v0.13.0"
+        network_mode = "host"
+	privileged = true
+        args = [
+	  "--web.listen-address=0.0.0.0:9101"
+	]
+      }
+
+      resources {
+        cpu = 50
+        memory = 40
+      }
+
+      service {
+        tags = [ "telemetry" ]
+        port = 9101
+        address_mode = "driver"
+        name = "smartctl-exporter"
+        check {
+          type = "http"
+          path = "/"
+          port = 9101
+          address_mode = "driver"
+          interval = "60s"
+          timeout = "5s"
+          check_restart {
+            limit = 3
+            grace = "90s"
+            ignore_warnings = false
+          }
+        }
+      }
+    }
+  }
+}

doc/ports

@@ -47,5 +47,6 @@ ports so that we can avoid conflicts when adding services.
 8999    opendkim
 9090    prometheus
 9100    node_exporter
+9101    smartctl_exporter
 9334    tricot metrics
 9991    guichet