smart errors grafana

bundles/smartctl/files/telegraf_plugin_errors

@@ -0,0 +1,27 @@
+#!/usr/bin/env python3
+
+from subprocess import check_output
+from json import loads
+import time
+
+
+nanosecond = time.time_ns()
+
+for line in check_output(['/usr/sbin/smartctl', '--scan', '-d', 'scsi']).splitlines():
+    device = line.split()[0].decode()
+    smart = loads(check_output(['/usr/sbin/smartctl', '-a', device, '-j']))
+    attributes = {
+        attribute['name']: attribute['flags']['value']
+            for attribute in smart['ata_smart_attributes']['table']
+            if int(attribute['id']) in [
+                # https://www.backblaze.com/blog/what-smart-stats-indicate-hard-drive-failures/
+                5,
+                187,
+                188,
+                197,
+                198,
+            ]
+    }
+    attributes_string = ','.join(f'{k}={v}' for k, v in attributes.items())
+
+    print(f"smart_errors,host=${node.name},device={device} {attributes_string} {nanosecond}")

bundles/smartctl/items.py

@@ -1,6 +1,11 @@
 files = {
-    '/usr/local/share/telegraf/smartctl': {
-        'source': 'telegraf_plugin',
+    '/usr/local/share/telegraf/smartctl_power_mode': {
+        'source': 'telegraf_plugin_power_mode',
+        'content_type': 'mako',
+        'mode': '0755',
+    },
+    '/usr/local/share/telegraf/smartctl_errors': {
+        'source': 'telegraf_plugin_errors',
         'content_type': 'mako',
         'mode': '0755',
     },
@@ -22,5 +27,5 @@ for device, conf in node.metadata.get('smartctl').items():
             }
         else:
             raise ValueError(f'{node.name}: unkown smartctl option: {option}')
-        
+
         previous_action = [f'action:{action_name}']

bundles/smartctl/metadata.py

@@ -13,17 +13,29 @@ defaults = {
     'telegraf': {
         'config': {
             'inputs': {
-                'exec': {h({
-                    'commands': [
-                        f'sudo /usr/local/share/telegraf/smartctl',
-                    ],
-                    'data_format': 'influx',
-                    'interval': '20s',
-                })},
+                'exec': {
+                    h({
+                        'commands': [
+                            f'sudo /usr/local/share/telegraf/smartctl_power_mode',
+                        ],
+                        'data_format': 'influx',
+                        'interval': '20s',
+                    }),
+                    h({
+                        'commands': [
+                            f'sudo /usr/local/share/telegraf/smartctl_errors',
+                        ],
+                        'data_format': 'influx',
+                        'interval': '6h',
+                    })
+                },
             },
         },
     },
     'sudoers': {
-        'telegraf': {'/usr/local/share/telegraf/smartctl'},
+        'telegraf': {
+            '/usr/local/share/telegraf/smartctl_power_mode',
+            '/usr/local/share/telegraf/smartctl_errors',
+        },
     },
 }

data/grafana/rows/smartctl.py

@@ -33,4 +33,20 @@
         'min': 0,
         'tooltip': 'multi',
     },
+    'errors': {
+        'stacked': True,
+        'queries': {
+            'power_level': {
+                'filters': {
+                    '_measurement': 'smart_errors',
+                },
+            },
+        },
+        'display_name': '__field.labels.device} ${__field.name',
+        'min': 0,
+        'tooltip': 'multi',
+        'legend': {
+            'displayMode': 'hidden',
+        },
+    },
 }