influx upgrade

salt/influxdb/templates/alarm_deadman.json

@@ -0,0 +1,18 @@
+{
+  "apiVersion": "influxdata.com/v2alpha1",
+  "kind": "CheckDeadman",
+  "metadata": {
+    "name": "alarm-deadman"
+  },
+  "spec": {
+    "description": "Data has not been received from Telegraf for an extended period.",
+    "every": "1m0s",
+    "level": "CRIT",
+    "name": "Telegraf Data Outage",
+    "query": "from(bucket: \"telegraf/so_short_term\")\n  |> range(start: v.timeRangeStart, stop: v.timeRangeStop)\n  |> filter(fn: (r) => r[\"_measurement\"] == \"cpu\")\n  |> filter(fn: (r) => r[\"_field\"] == \"usage_idle\")\n  |> filter(fn: (r) => r[\"cpu\"] == \"cpu-total\")",
+    "staleTime": "15m0s",
+    "status": "active",
+    "statusMessageTemplate": "Check: ${ r._check_name } is: ${ r._level }",
+    "timeSince": "2m0s"
+  }
+}

salt/influxdb/templates/alarm_nsm_disk.json

@@ -0,0 +1,32 @@
+{
+  "apiVersion": "influxdata.com/v2alpha1",
+  "kind": "CheckThreshold",
+  "metadata": {
+    "name": "alarm-nsm-disk"
+  },
+  "spec": {
+    "description": "Percent used space on the root partition of at least one node has exceeded the alarm threshold.",
+    "every": "1m0s",
+    "name": "NSM Disk High Usage",
+    "query": "from(bucket: \"telegraf/so_short_term\")\n  |> range(start: v.timeRangeStart, stop: v.timeRangeStop)\n  |> filter(fn: (r) => r[\"_measurement\"] == \"disk\")\n  |> filter(fn: (r) => r[\"path\"] == \"/\")\n  |> filter(fn: (r) => r[\"_field\"] == \"used_percent\")\n  |> aggregateWindow(every: 1m, fn: max, createEmpty: false)\n  |> yield(name: \"max\")",
+    "status": "active",
+    "statusMessageTemplate": "Check: ${ r._check_name } is: ${ r._level }",
+    "thresholds": [
+      {
+        "level": "CRIT",
+        "type": "greater",
+        "value": 95
+      },
+      {
+        "level": "INFO",
+        "type": "greater",
+        "value": 85
+      },
+      {
+        "level": "WARN",
+        "type": "greater",
+        "value": 90
+      }
+    ]
+  }
+}

salt/influxdb/templates/alarm_root_disk.json

@@ -0,0 +1,32 @@
+{
+  "apiVersion": "influxdata.com/v2alpha1",
+  "kind": "CheckThreshold",
+  "metadata": {
+    "name": "alarm-root-disk"
+  },
+  "spec": {
+    "description": "Percent used space on the root partition of at least one node has exceeded the alarm threshold.",
+    "every": "1m0s",
+    "name": "Root Disk High Usage",
+    "query": "from(bucket: \"telegraf/so_short_term\")\n  |> range(start: v.timeRangeStart, stop: v.timeRangeStop)\n  |> filter(fn: (r) => r[\"_measurement\"] == \"disk\")\n  |> filter(fn: (r) => r[\"path\"] == \"/\")\n  |> filter(fn: (r) => r[\"_field\"] == \"used_percent\")\n  |> aggregateWindow(every: 1m, fn: max, createEmpty: false)\n  |> yield(name: \"max\")",
+    "status": "active",
+    "statusMessageTemplate": "Check: ${ r._check_name } is: ${ r._level }",
+    "thresholds": [
+      {
+        "level": "CRIT",
+        "type": "greater",
+        "value": 95
+      },
+      {
+        "level": "INFO",
+        "type": "greater",
+        "value": 85
+      },
+      {
+        "level": "WARN",
+        "type": "greater",
+        "value": 90
+      }
+    ]
+  }
+}