1 contributor
# autoSMART SMART Parameters Configuration # Defines which SMART parameters to monitor and their thresholds [monitoring] # Collection interval in seconds collection_interval = 300 # 5 minutes collection_timeout = 30 # 30 seconds timeout per disk # Madagascar integration madagascar_inventory_file = /etc/madagascar/disk_inventory.json madagascar_api_endpoint = http://madagascar.local/api/v1/disks [smart_parameters] # Format: parameter_name = threshold,weight,enabled,description # Critical parameters (high weight, immediate attention) Raw_Read_Error_Rate = 100000,0.9,true,"Raw read error rate from disk surface" Reallocated_Sector_Ct = 5,0.95,true,"Count of reallocated sectors" Current_Pending_Sector = 1,0.9,true,"Count of sectors waiting for reallocation" Offline_Uncorrectable = 1,0.95,true,"Count of uncorrectable sectors" UDMA_CRC_Error_Count = 100,0.7,true,"Count of UDMA CRC errors" # Important parameters (medium weight) Spin_Retry_Count = 3,0.8,true,"Count of spin-up retry attempts" End-to-End_Error = 1,0.8,true,"End-to-end error detection count" Reported_Uncorrect = 1,0.85,true,"Count of uncorrectable errors reported" High_Fly_Writes = 1,0.7,true,"Count of high fly write operations" Airflow_Temperature_Cel = 50,0.6,true,"Temperature of airflow in Celsius" # Monitoring parameters (lower weight, trending) Temperature_Celsius = 55,0.6,true,"Drive temperature in Celsius" Power_On_Hours = 43800,0.4,true,"Total power-on hours (5 years)" Load_Cycle_Count = 300000,0.5,true,"Count of load/unload cycles" Start_Stop_Count = 10000,0.4,true,"Count of start/stop cycles" Power_Cycle_Count = 10000,0.4,true,"Count of power-on cycles" # Performance parameters (informational) Seek_Error_Rate = 100000,0.3,true,"Rate of seek errors" Throughput_Performance = 80,0.3,true,"Overall throughput performance" Spin_Up_Time = 10000,0.4,true,"Time required to spin up" [thresholds] # Global threshold multipliers temperature_warning = 0.9 # Warning at 90% of threshold temperature_critical = 1.0 # Critical at 100% of threshold sector_warning = 0.5 # Warning at 50% of threshold sector_critical = 1.0 # Critical at 100% of threshold # Trend analysis trend_window_hours = 168 # 7 days for trend analysis trend_deviation_threshold = 2.0 # Standard deviations for anomaly [exclusions] # Disk models/serials to exclude from monitoring exclude_models = "Virtual,QEMU,VMware" exclude_serials = "" exclude_by_size_gb = 8 # Exclude disks smaller than 8GB