You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
scrutiny/webapp/backend/pkg/models/measurements/smart_ata_attribute.go

164 lines
5.2 KiB

package measurements
import (
"fmt"
"github.com/analogj/scrutiny/webapp/backend/pkg"
"github.com/analogj/scrutiny/webapp/backend/pkg/thresholds"
"strconv"
"strings"
)
type SmartAtaAttribute struct {
AttributeId int `json:"attribute_id"`
Value int64 `json:"value"`
Threshold int64 `json:"thresh"`
Worst int64 `json:"worst"`
RawValue int64 `json:"raw_value"`
RawString string `json:"raw_string"`
WhenFailed string `json:"when_failed"`
//Generated data
TransformedValue int64 `json:"transformed_value"`
Status int64 `json:"status,omitempty"`
StatusReason string `json:"status_reason,omitempty"`
FailureRate float64 `json:"failure_rate,omitempty"`
}
func (sa *SmartAtaAttribute) GetStatus() int64 {
return sa.Status
}
func (sa *SmartAtaAttribute) Flatten() map[string]interface{} {
idString := strconv.Itoa(sa.AttributeId)
return map[string]interface{}{
fmt.Sprintf("attr.%s.attribute_id", idString): idString,
fmt.Sprintf("attr.%s.value", idString): sa.Value,
fmt.Sprintf("attr.%s.worst", idString): sa.Worst,
fmt.Sprintf("attr.%s.thresh", idString): sa.Threshold,
fmt.Sprintf("attr.%s.raw_value", idString): sa.RawValue,
fmt.Sprintf("attr.%s.raw_string", idString): sa.RawString,
fmt.Sprintf("attr.%s.when_failed", idString): sa.WhenFailed,
//Generated Data
fmt.Sprintf("attr.%s.transformed_value", idString): sa.TransformedValue,
fmt.Sprintf("attr.%s.status", idString): sa.Status,
fmt.Sprintf("attr.%s.status_reason", idString): sa.StatusReason,
fmt.Sprintf("attr.%s.failure_rate", idString): sa.FailureRate,
}
}
func (sa *SmartAtaAttribute) Inflate(key string, val interface{}) {
if val == nil {
return
}
keyParts := strings.Split(key, ".")
switch keyParts[2] {
case "attribute_id":
attrId, err := strconv.Atoi(val.(string))
if err == nil {
sa.AttributeId = attrId
}
case "value":
sa.Value = val.(int64)
case "worst":
sa.Worst = val.(int64)
case "thresh":
sa.Threshold = val.(int64)
case "raw_value":
sa.RawValue = val.(int64)
case "raw_string":
sa.RawString = val.(string)
case "when_failed":
sa.WhenFailed = val.(string)
//generated
case "transformed_value":
sa.TransformedValue = val.(int64)
case "status":
sa.Status = val.(int64)
case "status_reason":
sa.StatusReason = val.(string)
case "failure_rate":
sa.FailureRate = val.(float64)
}
}
//populate attribute status, using SMART Thresholds & Observed Metadata
// Chainable
func (sa *SmartAtaAttribute) PopulateAttributeStatus() *SmartAtaAttribute {
if strings.ToUpper(sa.WhenFailed) == pkg.SmartWhenFailedFailingNow {
//this attribute has previously failed
sa.Status = pkg.SmartAttributeStatusFailed
sa.StatusReason = "Attribute is failing manufacturer SMART threshold"
} else if strings.ToUpper(sa.WhenFailed) == pkg.SmartWhenFailedInThePast {
sa.Status = pkg.SmartAttributeStatusWarning
sa.StatusReason = "Attribute has previously failed manufacturer SMART threshold"
}
if smartMetadata, ok := thresholds.AtaMetadata[sa.AttributeId]; ok {
sa.ValidateThreshold(smartMetadata)
}
return sa
}
// compare the attribute (raw, normalized, transformed) value to observed thresholds, and update status if necessary
func (sa *SmartAtaAttribute) ValidateThreshold(smartMetadata thresholds.AtaAttributeMetadata) {
//TODO: multiple rules
// try to predict the failure rates for observed thresholds that have 0 failure rate and error bars.
// - if the attribute is critical
// - the failure rate is over 10 - set to failed
// - the attribute does not match any threshold, set to warn
// - if the attribute is not critical
// - if failure rate is above 20 - set to failed
// - if failure rate is above 10 but below 20 - set to warn
//update the smart attribute status based on Observed thresholds.
var value int64
if smartMetadata.DisplayType == thresholds.AtaSmartAttributeDisplayTypeNormalized {
value = int64(sa.Value)
} else if smartMetadata.DisplayType == thresholds.AtaSmartAttributeDisplayTypeTransformed {
value = sa.TransformedValue
} else {
value = sa.RawValue
}
for _, obsThresh := range smartMetadata.ObservedThresholds {
//check if "value" is in this bucket
if ((obsThresh.Low == obsThresh.High) && value == obsThresh.Low) ||
(obsThresh.Low < value && value <= obsThresh.High) {
sa.FailureRate = obsThresh.AnnualFailureRate
if smartMetadata.Critical {
if obsThresh.AnnualFailureRate >= 0.10 {
sa.Status = pkg.SmartAttributeStatusFailed
sa.StatusReason = "Observed Failure Rate for Critical Attribute is greater than 10%"
}
} else {
if obsThresh.AnnualFailureRate >= 0.20 {
sa.Status = pkg.SmartAttributeStatusFailed
sa.StatusReason = "Observed Failure Rate for Attribute is greater than 20%"
} else if obsThresh.AnnualFailureRate >= 0.10 {
sa.Status = pkg.SmartAttributeStatusWarning
sa.StatusReason = "Observed Failure Rate for Attribute is greater than 10%"
}
}
//we've found the correct bucket, we can drop out of this loop
return
}
}
// no bucket found
if smartMetadata.Critical {
sa.Status = pkg.SmartAttributeStatusWarning
sa.StatusReason = "Could not determine Observed Failure Rate for Critical Attribute"
}
return
}