check disk wearout levels

alerting/alerts.py

@@ -20,14 +20,15 @@ class AlertType(StrEnum):
     LOGIN = "LOGIN"  # TODO
     SMART = "SMART"  # TODO
     RAID = "RAID"
+    DISKS = "DISKS"
     UPS = "UPS"
     UPDATE = "UPDATE"
 
 
-class Severity(Enum):
+class Severity(StrEnum):
-    INFO = 1
+    INFO = "INFO"
-    WARNING = 2
+    WARNING = "WARNING"
-    CRITICAL = 3
+    CRITICAL = "CRITICAL"
 
 
 @dataclass

config.example.json

@@ -7,7 +7,21 @@
             ]
         },
         "raid": {
-            "lvs": ["Data/lvol0"]
+            "lvs": [
+                "Data/lvol0"
+            ]
+        },
+        "wearout": {
+            "disks": [
+                {
+                    "name": "/dev/sda",
+                    "severity": "WARNING"
+                },
+                {
+                    "name": "/dev/nvme0",
+                    "severity": "CRITICAL"
+                }
+            ]
         }
     }
 }

misc/checks.py

@@ -4,7 +4,7 @@ from datetime import timedelta
 
 from alerting import alerts
 from misc import cvars, docker_registry, sensors, vuln
-from misc.disks import LVAttr
+from misc.disks import LVAttr, WearoutIndicator, get_wearout_reading
 
 IS_TESTING = False
 
@@ -236,3 +236,35 @@ def raid_check() -> list[alerts.Alert]:
                 )
 
     return alert_list
+
+
+def disk_wearout_check() -> list[alerts.Alert]:
+    check_config = cvars.config.get()["checks"]["wearout"]
+    alert_list = []
+    for disk in check_config["disks"]:
+        try:
+            wearout_reading = get_wearout_reading(disk["name"])
+        except Exception as exc:
+            alert_list.append(
+                alerts.Alert(
+                    alert_type=alerts.AlertType.ERROR,
+                    message=f"Could not check wearout for disk {disk['name']}: {repr(exc)}, see logs",
+                    severity=alerts.Severity.CRITICAL,
+                )
+            )
+            logging.error(traceback.format_exc())
+            continue
+
+        if wearout_reading.current_reading < wearout_reading.threshold_reading:
+            match wearout_reading.indicator:
+                case WearoutIndicator.REALLOCATED_SECTORS:
+                    message = f"Disk {disk['name']} has reallocated sectors (curr {wearout_reading.current_reading}, thresh {wearout_reading.threshold_reading})"
+                case WearoutIndicator.SPARE_BLOCKS:
+                    message = f"Disk {disk['name']} has too few spare blocks (curr {wearout_reading.current_reading}, thresh {wearout_reading.threshold_reading})"
+            alert_list.append(
+                alerts.Alert(
+                    alert_type=alerts.AlertType.DISKS, message=message, severity=alerts.Severity[disk["severity"]]
+                )
+            )
+
+    return alert_list

misc/disks.py

@@ -1,7 +1,7 @@
 import json
 import subprocess
 from dataclasses import dataclass
-from enum import StrEnum
+from enum import Enum, StrEnum
 from typing import Optional, Self
 
 
@@ -132,3 +132,58 @@ class LVAttr:
         json_obj = json.loads(subprocess.run(["lvs", "--reportformat=json", name], capture_output=True).stdout)
         attr_str = json_obj["report"][0]["lv"][0]["lv_attr"]
         return cls.from_str(attr_str, name)
+
+
+class WearoutIndicator(Enum):
+    REALLOCATED_SECTORS = 0
+    SPARE_BLOCKS = 1
+
+
+@dataclass
+class WearoutReading:
+    indicator: WearoutIndicator
+    current_reading: int
+    threshold_reading: int
+
+
+def _get_wearout_reading_from_smartctl_output(smartctl_output: dict) -> WearoutReading:
+    disk_protocol = smartctl_output["device"]["protocol"]
+    rotation_rate = smartctl_output.get("rotation_rate", 0)
+    match rotation_rate:
+        case 0:  # assuming non-rotating media is an SSD
+            indicator = WearoutIndicator.SPARE_BLOCKS
+            match disk_protocol:
+                case "ATA":
+                    attr_table = smartctl_output["ata_smart_attributes"]["table"]
+                    for a in attr_table:
+                        if a["name"] == "Available_Reservd_Space":
+                            value = a["value"]
+                            threshold = a["thresh"]
+                            break
+                    else:
+                        raise Exception(f"no Available_Reservd_Space on ATA SSD")
+                case "NVMe":
+                    value = smartctl_output["nvme_smart_health_information_log"]["available_spare"]
+                    threshold = smartctl_output["nvme_smart_health_information_log"]["available_spare_threshold"]
+        case _:
+            indicator = WearoutIndicator.REALLOCATED_SECTORS
+            match disk_protocol:
+                case "ATA":
+                    attr_table = smartctl_output["ata_smart_attributes"]["table"]
+                    for a in attr_table:
+                        if a["name"] == "Reallocated_Sector_Ct":
+                            value = a["value"]
+                            threshold = a["thresh"]
+                            break
+                    else:
+                        raise Exception(f"no Reallocated_Sector_Ct on ATA HDD")
+                case "NVMe":  # ? NVMe HDDs are very rare, if they even exist
+                    raise NotImplementedError
+
+    return WearoutReading(indicator, current_reading=value, threshold_reading=threshold)
+
+
+def get_wearout_reading(disk: str) -> WearoutReading:
+    smartctl_output = json.loads(subprocess.run(["smartctl", "-ja", disk], capture_output=True).stdout.decode("utf-8"))
+    wearout_reading = _get_wearout_reading_from_smartctl_output(smartctl_output)
+    return wearout_reading

service.py

@@ -38,6 +38,7 @@ async def main():
         interval_checker(checks.ram_check, datetime.timedelta(minutes=1)),
         interval_checker(checks.vuln_check, datetime.timedelta(days=1)),
         interval_checker(checks.raid_check, datetime.timedelta(days=1)),
+        interval_checker(checks.disk_wearout_check, datetime.timedelta(days=1)),
         scheduled_checker(
             checks.docker_registry_check, period=datetime.timedelta(days=1), when=datetime.time(hour=0, minute=0)
         ),

tests/smartctl_ata_hdd.json

@@ -0,0 +1,594 @@
+{
+  "json_format_version": [
+    1,
+    0
+  ],
+  "smartctl": {
+    "version": [
+      7,
+      4
+    ],
+    "pre_release": false,
+    "svn_revision": "5530",
+    "platform_info": "x86_64-linux-6.11.3-arch1-1",
+    "build_info": "(local build)",
+    "argv": [
+      "smartctl",
+      "-ja",
+      "/dev/sda"
+    ],
+    "drive_database_version": {
+      "string": "7.3/5528"
+    },
+    "exit_status": 0
+  },
+  "local_time": {
+    "time_t": 1731149584,
+    "asctime": "Sat Nov  9 13:53:04 2024 MSK"
+  },
+  "device": {
+    "name": "/dev/sda",
+    "info_name": "/dev/sda [SAT]",
+    "type": "sat",
+    "protocol": "ATA"
+  },
+  "model_name": "WDC WD20EARZ-00C5XB0",
+  "serial_number": "WD-WX32D83C15U7",
+  "wwn": {
+    "naa": 5,
+    "oui": 5358,
+    "id": 8959374949
+  },
+  "firmware_version": "01.01A01",
+  "user_capacity": {
+    "blocks": 3907029168,
+    "bytes": 2000398934016
+  },
+  "logical_block_size": 512,
+  "physical_block_size": 4096,
+  "rotation_rate": 5400,
+  "form_factor": {
+    "ata_value": 2,
+    "name": "3.5 inches"
+  },
+  "trim": {
+    "supported": false
+  },
+  "in_smartctl_database": false,
+  "ata_version": {
+    "string": "ACS-3 T13/2161-D revision 5",
+    "major_value": 2046,
+    "minor_value": 109
+  },
+  "sata_version": {
+    "string": "SATA 3.1",
+    "value": 126
+  },
+  "interface_speed": {
+    "max": {
+      "sata_value": 14,
+      "string": "6.0 Gb/s",
+      "units_per_second": 60,
+      "bits_per_unit": 100000000
+    },
+    "current": {
+      "sata_value": 3,
+      "string": "6.0 Gb/s",
+      "units_per_second": 60,
+      "bits_per_unit": 100000000
+    }
+  },
+  "smart_support": {
+    "available": true,
+    "enabled": true
+  },
+  "smart_status": {
+    "passed": true
+  },
+  "ata_smart_data": {
+    "offline_data_collection": {
+      "status": {
+        "value": 0,
+        "string": "was never started"
+      },
+      "completion_seconds": 19380
+    },
+    "self_test": {
+      "status": {
+        "value": 0,
+        "string": "completed without error",
+        "passed": true
+      },
+      "polling_minutes": {
+        "short": 2,
+        "extended": 208,
+        "conveyance": 5
+      }
+    },
+    "capabilities": {
+      "values": [
+        123,
+        3
+      ],
+      "exec_offline_immediate_supported": true,
+      "offline_is_aborted_upon_new_cmd": false,
+      "offline_surface_scan_supported": true,
+      "self_tests_supported": true,
+      "conveyance_self_test_supported": true,
+      "selective_self_test_supported": true,
+      "attribute_autosave_enabled": true,
+      "error_logging_supported": true,
+      "gp_logging_supported": true
+    }
+  },
+  "ata_sct_capabilities": {
+    "value": 12341,
+    "error_recovery_control_supported": false,
+    "feature_control_supported": true,
+    "data_table_supported": true
+  },
+  "ata_smart_attributes": {
+    "revision": 16,
+    "table": [
+      {
+        "id": 1,
+        "name": "Raw_Read_Error_Rate",
+        "value": 200,
+        "worst": 200,
+        "thresh": 51,
+        "when_failed": "",
+        "flags": {
+          "value": 47,
+          "string": "POSR-K ",
+          "prefailure": true,
+          "updated_online": true,
+          "performance": true,
+          "error_rate": true,
+          "event_count": false,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      },
+      {
+        "id": 3,
+        "name": "Spin_Up_Time",
+        "value": 175,
+        "worst": 175,
+        "thresh": 21,
+        "when_failed": "",
+        "flags": {
+          "value": 39,
+          "string": "POS--K ",
+          "prefailure": true,
+          "updated_online": true,
+          "performance": true,
+          "error_rate": false,
+          "event_count": false,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 2241,
+          "string": "2241"
+        }
+      },
+      {
+        "id": 4,
+        "name": "Start_Stop_Count",
+        "value": 100,
+        "worst": 100,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 11,
+          "string": "11"
+        }
+      },
+      {
+        "id": 5,
+        "name": "Reallocated_Sector_Ct",
+        "value": 200,
+        "worst": 200,
+        "thresh": 140,
+        "when_failed": "",
+        "flags": {
+          "value": 51,
+          "string": "PO--CK ",
+          "prefailure": true,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      },
+      {
+        "id": 7,
+        "name": "Seek_Error_Rate",
+        "value": 100,
+        "worst": 253,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 46,
+          "string": "-OSR-K ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": true,
+          "error_rate": true,
+          "event_count": false,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      },
+      {
+        "id": 9,
+        "name": "Power_On_Hours",
+        "value": 96,
+        "worst": 96,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 3275,
+          "string": "3275"
+        }
+      },
+      {
+        "id": 10,
+        "name": "Spin_Retry_Count",
+        "value": 100,
+        "worst": 253,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      },
+      {
+        "id": 11,
+        "name": "Calibration_Retry_Count",
+        "value": 100,
+        "worst": 253,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      },
+      {
+        "id": 12,
+        "name": "Power_Cycle_Count",
+        "value": 100,
+        "worst": 100,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 10,
+          "string": "10"
+        }
+      },
+      {
+        "id": 192,
+        "name": "Power-Off_Retract_Count",
+        "value": 200,
+        "worst": 200,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 6,
+          "string": "6"
+        }
+      },
+      {
+        "id": 193,
+        "name": "Load_Cycle_Count",
+        "value": 200,
+        "worst": 200,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 28,
+          "string": "28"
+        }
+      },
+      {
+        "id": 194,
+        "name": "Temperature_Celsius",
+        "value": 112,
+        "worst": 105,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 34,
+          "string": "-O---K ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": false,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 31,
+          "string": "31"
+        }
+      },
+      {
+        "id": 196,
+        "name": "Reallocated_Event_Count",
+        "value": 200,
+        "worst": 200,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      },
+      {
+        "id": 197,
+        "name": "Current_Pending_Sector",
+        "value": 200,
+        "worst": 200,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      },
+      {
+        "id": 198,
+        "name": "Offline_Uncorrectable",
+        "value": 100,
+        "worst": 253,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 48,
+          "string": "----CK ",
+          "prefailure": false,
+          "updated_online": false,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      },
+      {
+        "id": 199,
+        "name": "UDMA_CRC_Error_Count",
+        "value": 200,
+        "worst": 200,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      },
+      {
+        "id": 200,
+        "name": "Multi_Zone_Error_Rate",
+        "value": 100,
+        "worst": 253,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 8,
+          "string": "---R-- ",
+          "prefailure": false,
+          "updated_online": false,
+          "performance": false,
+          "error_rate": true,
+          "event_count": false,
+          "auto_keep": false
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      }
+    ]
+  },
+  "power_on_time": {
+    "hours": 3275
+  },
+  "power_cycle_count": 10,
+  "temperature": {
+    "current": 31
+  },
+  "ata_smart_error_log": {
+    "summary": {
+      "revision": 1,
+      "count": 0
+    }
+  },
+  "ata_smart_self_test_log": {
+    "standard": {
+      "revision": 1,
+      "table": [
+        {
+          "type": {
+            "value": 3,
+            "string": "Conveyance offline"
+          },
+          "status": {
+            "value": 0,
+            "string": "Completed without error",
+            "passed": true
+          },
+          "lifetime_hours": 0
+        }
+      ],
+      "count": 1,
+      "error_count_total": 0,
+      "error_count_outdated": 0
+    }
+  },
+  "ata_smart_selective_self_test_log": {
+    "revision": 1,
+    "table": [
+      {
+        "lba_min": 0,
+        "lba_max": 0,
+        "status": {
+          "value": 0,
+          "string": "Not_testing"
+        }
+      },
+      {
+        "lba_min": 0,
+        "lba_max": 0,
+        "status": {
+          "value": 0,
+          "string": "Not_testing"
+        }
+      },
+      {
+        "lba_min": 0,
+        "lba_max": 0,
+        "status": {
+          "value": 0,
+          "string": "Not_testing"
+        }
+      },
+      {
+        "lba_min": 0,
+        "lba_max": 0,
+        "status": {
+          "value": 0,
+          "string": "Not_testing"
+        }
+      },
+      {
+        "lba_min": 0,
+        "lba_max": 0,
+        "status": {
+          "value": 0,
+          "string": "Not_testing"
+        }
+      }
+    ],
+    "flags": {
+      "value": 0,
+      "remainder_scan_enabled": false
+    },
+    "power_up_scan_resume_minutes": 0
+  }
+}

tests/smartctl_ata_ssd.json

@@ -0,0 +1,680 @@
+{
+  "json_format_version": [
+    1,
+    0
+  ],
+  "smartctl": {
+    "version": [
+      7,
+      4
+    ],
+    "pre_release": false,
+    "svn_revision": "5530",
+    "platform_info": "x86_64-linux-6.11.6-arch1-1",
+    "build_info": "(local build)",
+    "argv": [
+      "smartctl",
+      "-ja",
+      "/dev/sda"
+    ],
+    "drive_database_version": {
+      "string": "7.3/5528"
+    },
+    "exit_status": 0
+  },
+  "local_time": {
+    "time_t": 1731149676,
+    "asctime": "Sat Nov  9 13:54:36 2024 MSK"
+  },
+  "device": {
+    "name": "/dev/sda",
+    "info_name": "/dev/sda [SAT]",
+    "type": "sat",
+    "protocol": "ATA"
+  },
+  "model_family": "WD Blue / Red / Green SSDs",
+  "model_name": "WDC WDS100T2G0A-00JH30",
+  "serial_number": "20299A802244",
+  "wwn": {
+    "naa": 5,
+    "oui": 6980,
+    "id": 37501727029
+  },
+  "firmware_version": "UH510000",
+  "user_capacity": {
+    "blocks": 1953529856,
+    "bytes": 1000207286272
+  },
+  "logical_block_size": 512,
+  "physical_block_size": 512,
+  "rotation_rate": 0,
+  "form_factor": {
+    "ata_value": 3,
+    "name": "2.5 inches"
+  },
+  "trim": {
+    "supported": true,
+    "deterministic": true,
+    "zeroed": false
+  },
+  "in_smartctl_database": true,
+  "ata_version": {
+    "string": "ACS-2 T13/2015-D revision 3",
+    "major_value": 1008,
+    "minor_value": 272
+  },
+  "sata_version": {
+    "string": "SATA 3.2",
+    "value": 255
+  },
+  "interface_speed": {
+    "max": {
+      "sata_value": 14,
+      "string": "6.0 Gb/s",
+      "units_per_second": 60,
+      "bits_per_unit": 100000000
+    },
+    "current": {
+      "sata_value": 3,
+      "string": "6.0 Gb/s",
+      "units_per_second": 60,
+      "bits_per_unit": 100000000
+    }
+  },
+  "smart_support": {
+    "available": true,
+    "enabled": true
+  },
+  "smart_status": {
+    "passed": true
+  },
+  "ata_smart_data": {
+    "offline_data_collection": {
+      "status": {
+        "value": 0,
+        "string": "was never started"
+      },
+      "completion_seconds": 120
+    },
+    "self_test": {
+      "status": {
+        "value": 0,
+        "string": "completed without error",
+        "passed": true
+      },
+      "polling_minutes": {
+        "short": 2,
+        "extended": 182
+      }
+    },
+    "capabilities": {
+      "values": [
+        21,
+        3
+      ],
+      "exec_offline_immediate_supported": true,
+      "offline_is_aborted_upon_new_cmd": true,
+      "offline_surface_scan_supported": false,
+      "self_tests_supported": true,
+      "conveyance_self_test_supported": false,
+      "selective_self_test_supported": false,
+      "attribute_autosave_enabled": true,
+      "error_logging_supported": true,
+      "gp_logging_supported": true
+    }
+  },
+  "ata_smart_attributes": {
+    "revision": 1,
+    "table": [
+      {
+        "id": 5,
+        "name": "Reallocated_Sector_Ct",
+        "value": 100,
+        "worst": 100,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      },
+      {
+        "id": 9,
+        "name": "Power_On_Hours",
+        "value": 100,
+        "worst": 100,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 7976,
+          "string": "7976"
+        }
+      },
+      {
+        "id": 12,
+        "name": "Power_Cycle_Count",
+        "value": 100,
+        "worst": 100,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 1218,
+          "string": "1218"
+        }
+      },
+      {
+        "id": 165,
+        "name": "Block_Erase_Count",
+        "value": 100,
+        "worst": 100,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 2063,
+          "string": "2063"
+        }
+      },
+      {
+        "id": 166,
+        "name": "Minimum_PE_Cycles_TLC",
+        "value": 100,
+        "worst": 100,
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 15,
+          "string": "15"
+        }
+      },
+      {
+        "id": 167,
+        "name": "Max_Bad_Blocks_per_Die",
+        "value": 100,
+        "worst": 100,
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      },
+      {
+        "id": 168,
+        "name": "Maximum_PE_Cycles_TLC",
+        "value": 100,
+        "worst": 100,
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 36,
+          "string": "36"
+        }
+      },
+      {
+        "id": 169,
+        "name": "Total_Bad_Blocks",
+        "value": 100,
+        "worst": 100,
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 1075,
+          "string": "1075"
+        }
+      },
+      {
+        "id": 170,
+        "name": "Grown_Bad_Blocks",
+        "value": 100,
+        "worst": 100,
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      },
+      {
+        "id": 171,
+        "name": "Program_Fail_Count",
+        "value": 100,
+        "worst": 100,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      },
+      {
+        "id": 172,
+        "name": "Erase_Fail_Count",
+        "value": 100,
+        "worst": 100,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      },
+      {
+        "id": 173,
+        "name": "Average_PE_Cycles_TLC",
+        "value": 100,
+        "worst": 100,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 15,
+          "string": "15"
+        }
+      },
+      {
+        "id": 174,
+        "name": "Unexpected_Power_Loss",
+        "value": 100,
+        "worst": 100,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 144,
+          "string": "144"
+        }
+      },
+      {
+        "id": 184,
+        "name": "End-to-End_Error",
+        "value": 100,
+        "worst": 100,
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      },
+      {
+        "id": 187,
+        "name": "Reported_Uncorrect",
+        "value": 100,
+        "worst": 100,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      },
+      {
+        "id": 188,
+        "name": "Command_Timeout",
+        "value": 100,
+        "worst": 100,
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      },
+      {
+        "id": 194,
+        "name": "Temperature_Celsius",
+        "value": 70,
+        "worst": 58,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 34,
+          "string": "-O---K ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": false,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 249108103198,
+          "string": "30 (Min/Max 0/58)"
+        }
+      },
+      {
+        "id": 199,
+        "name": "UDMA_CRC_Error_Count",
+        "value": 100,
+        "worst": 100,
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      },
+      {
+        "id": 230,
+        "name": "Media_Wearout_Indicator",
+        "value": 100,
+        "worst": 100,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 5879860561241,
+          "string": "0x055903000559"
+        }
+      },
+      {
+        "id": 232,
+        "name": "Available_Reservd_Space",
+        "value": 100,
+        "worst": 100,
+        "thresh": 5,
+        "when_failed": "",
+        "flags": {
+          "value": 51,
+          "string": "PO--CK ",
+          "prefailure": true,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 100,
+          "string": "100"
+        }
+      },
+      {
+        "id": 233,
+        "name": "NAND_GB_Written_TLC",
+        "value": 100,
+        "worst": 100,
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 15654,
+          "string": "15654"
+        }
+      },
+      {
+        "id": 234,
+        "name": "NAND_GB_Written_SLC",
+        "value": 100,
+        "worst": 100,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 32244,
+          "string": "32244"
+        }
+      },
+      {
+        "id": 241,
+        "name": "Host_Writes_GiB",
+        "value": 100,
+        "worst": 100,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 48,
+          "string": "----CK ",
+          "prefailure": false,
+          "updated_online": false,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 13847,
+          "string": "13847"
+        }
+      },
+      {
+        "id": 242,
+        "name": "Host_Reads_GiB",
+        "value": 100,
+        "worst": 100,
+        "thresh": 0,
+        "when_failed": "",
+        "flags": {
+          "value": 48,
+          "string": "----CK ",
+          "prefailure": false,
+          "updated_online": false,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 31195,
+          "string": "31195"
+        }
+      },
+      {
+        "id": 244,
+        "name": "Temp_Throttle_Status",
+        "value": 0,
+        "worst": 100,
+        "flags": {
+          "value": 50,
+          "string": "-O--CK ",
+          "prefailure": false,
+          "updated_online": true,
+          "performance": false,
+          "error_rate": false,
+          "event_count": true,
+          "auto_keep": true
+        },
+        "raw": {
+          "value": 0,
+          "string": "0"
+        }
+      }
+    ]
+  },
+  "power_on_time": {
+    "hours": 7976
+  },
+  "power_cycle_count": 1218,
+  "temperature": {
+    "current": 30
+  },
+  "ata_smart_error_log": {
+    "summary": {
+      "revision": 1,
+      "count": 0
+    }
+  },
+  "ata_smart_self_test_log": {
+    "standard": {
+      "revision": 1,
+      "count": 0
+    }
+  }
+}

tests/smartctl_nvme_ssd.json

@@ -0,0 +1,145 @@
+{
+  "json_format_version": [
+    1,
+    0
+  ],
+  "smartctl": {
+    "version": [
+      7,
+      4
+    ],
+    "pre_release": false,
+    "svn_revision": "5530",
+    "platform_info": "x86_64-linux-6.11.3-arch1-1",
+    "build_info": "(local build)",
+    "argv": [
+      "smartctl",
+      "-ja",
+      "/dev/nvme0"
+    ],
+    "exit_status": 0
+  },
+  "local_time": {
+    "time_t": 1731149045,
+    "asctime": "Sat Nov  9 13:44:05 2024 MSK"
+  },
+  "device": {
+    "name": "/dev/nvme0",
+    "info_name": "/dev/nvme0",
+    "type": "nvme",
+    "protocol": "NVMe"
+  },
+  "model_name": "Samsung SSD 970 EVO Plus 1TB",
+  "serial_number": "S4EWNM0W921977B",
+  "firmware_version": "2B2QEXM7",
+  "nvme_pci_vendor": {
+    "id": 5197,
+    "subsystem_id": 5197
+  },
+  "nvme_ieee_oui_identifier": 9528,
+  "nvme_total_capacity": 1000204886016,
+  "nvme_unallocated_capacity": 0,
+  "nvme_controller_id": 4,
+  "nvme_version": {
+    "string": "1.3",
+    "value": 66304
+  },
+  "nvme_number_of_namespaces": 1,
+  "nvme_namespaces": [
+    {
+      "id": 1,
+      "size": {
+        "blocks": 1953525168,
+        "bytes": 1000204886016
+      },
+      "capacity": {
+        "blocks": 1953525168,
+        "bytes": 1000204886016
+      },
+      "utilization": {
+        "blocks": 686279048,
+        "bytes": 351374872576
+      },
+      "formatted_lba_size": 512,
+      "eui64": {
+        "oui": 9528,
+        "ext_id": 383083641036
+      }
+    }
+  ],
+  "user_capacity": {
+    "blocks": 1953525168,
+    "bytes": 1000204886016
+  },
+  "logical_block_size": 512,
+  "smart_support": {
+    "available": true,
+    "enabled": true
+  },
+  "smart_status": {
+    "passed": true,
+    "nvme": {
+      "value": 0
+    }
+  },
+  "nvme_smart_health_information_log": {
+    "critical_warning": 0,
+    "temperature": 47,
+    "available_spare": 100,
+    "available_spare_threshold": 10,
+    "percentage_used": 0,
+    "data_units_read": 111588,
+    "data_units_written": 1802957,
+    "host_reads": 2570341,
+    "host_writes": 36266417,
+    "controller_busy_time": 133,
+    "power_cycles": 31,
+    "power_on_hours": 432,
+    "unsafe_shutdowns": 18,
+    "media_errors": 0,
+    "num_err_log_entries": 63,
+    "warning_temp_time": 0,
+    "critical_comp_time": 0,
+    "temperature_sensors": [
+      47,
+      51
+    ]
+  },
+  "temperature": {
+    "current": 47
+  },
+  "power_cycle_count": 31,
+  "power_on_time": {
+    "hours": 432
+  },
+  "nvme_error_information_log": {
+    "size": 64,
+    "read": 16,
+    "unread": 0,
+    "table": [
+      {
+        "error_count": 63,
+        "submission_queue_id": 0,
+        "command_id": 8,
+        "status_field": {
+          "value": 8194,
+          "do_not_retry": false,
+          "status_code_type": 0,
+          "status_code": 2,
+          "string": "Invalid Field in Command"
+        },
+        "phase_tag": false,
+        "lba": {
+          "value": 0
+        },
+        "nsid": 0
+      }
+    ]
+  },
+  "nvme_self_test_log": {
+    "current_self_test_operation": {
+      "value": 0,
+      "string": "No self-test in progress"
+    }
+  }
+}

tests/test_disks.py

@@ -1,6 +1,12 @@
+import json
 import unittest
 
-from misc.disks import LVAttr
+from misc.disks import (
+    LVAttr,
+    WearoutIndicator,
+    WearoutReading,
+    _get_wearout_reading_from_smartctl_output,
+)
 
 
 class TestDisks(unittest.TestCase):
@@ -22,6 +28,30 @@ class TestDisks(unittest.TestCase):
             ),
         )
 
+    def test_wearout_reading_nvme_ssd(self):
+        with open("tests/smartctl_nvme_ssd.json") as f:
+            smartctl_output = json.load(f)
+        self.assertEqual(
+            _get_wearout_reading_from_smartctl_output(smartctl_output),
+            WearoutReading(indicator=WearoutIndicator.SPARE_BLOCKS, current_reading=100, threshold_reading=10),
+        )
+
+    def test_wearout_reading_ata_hdd(self):
+        with open("tests/smartctl_ata_hdd.json") as f:
+            smartctl_output = json.load(f)
+        self.assertEqual(
+            _get_wearout_reading_from_smartctl_output(smartctl_output),
+            WearoutReading(indicator=WearoutIndicator.REALLOCATED_SECTORS, current_reading=200, threshold_reading=140),
+        )
+
+    def test_wearout_reading_ata_ssd(self):
+        with open("tests/smartctl_ata_ssd.json") as f:
+            smartctl_output = json.load(f)
+        self.assertEqual(
+            _get_wearout_reading_from_smartctl_output(smartctl_output),
+            WearoutReading(indicator=WearoutIndicator.SPARE_BLOCKS, current_reading=100, threshold_reading=5),
+        )
+
 
 if __name__ == "__main__":
     unittest.main()