gilles/scrutiny
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add support for disabling repeat notifications

Browse Source 
* Add a new database function for getting the tail

* Update ShouldNotify() to handle ignoring repeat notifications if set
This commit is contained in:
Aram Akhavan
2023-11-26 22:43:50 -08:00
parent 6417d71311
commit 4e5c76b259
4 changed files with 139 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files
webapp/backend/pkg/notify/notify.go
+62 -33
View File
@@ -15,11 +15,13 @@ import (
"github.com/analogj/go-util/utils"
"github.com/analogj/scrutiny/webapp/backend/pkg"
"github.com/analogj/scrutiny/webapp/backend/pkg/config"
"github.com/analogj/scrutiny/webapp/backend/pkg/database"
"github.com/analogj/scrutiny/webapp/backend/pkg/models"
"github.com/analogj/scrutiny/webapp/backend/pkg/models/measurements"
"github.com/analogj/scrutiny/webapp/backend/pkg/thresholds"
"github.com/containrrr/shoutrrr"
shoutrrrTypes "github.com/containrrr/shoutrrr/pkg/types"
"github.com/gin-gonic/gin"
"github.com/sirupsen/logrus"
"golang.org/x/sync/errgroup"
)
@@ -30,7 +32,7 @@ const NotifyFailureTypeSmartFailure = "SmartFailure"
const NotifyFailureTypeScrutinyFailure = "ScrutinyFailure"
// ShouldNotify check if the error Message should be filtered (level mismatch or filtered_attributes)
func ShouldNotify(device models.Device, smartAttrs measurements.Smart, statusThreshold pkg.MetricsStatusThreshold, statusFilterAttributes pkg.MetricsStatusFilterAttributes) bool {
func ShouldNotify(device models.Device, smartAttrs measurements.Smart, statusThreshold pkg.MetricsStatusThreshold, statusFilterAttributes pkg.MetricsStatusFilterAttributes, repeatNotifications bool, c *gin.Context, deviceRepo database.DeviceRepo) bool {
// 1. check if the device is healthy
if device.DeviceStatus == pkg.DeviceStatusPassed {
return false
@@ -54,52 +56,79 @@ func ShouldNotify(device models.Device, smartAttrs measurements.Smart, statusThr
requiredAttrStatus = pkg.AttributeStatusFailedScrutiny
}
// 2. check if the attributes that are failing should be filtered (non-critical)
// 3. for any unfiltered attribute, store the failure reason (Smart or Scrutiny)
if statusFilterAttributes == pkg.MetricsStatusFilterAttributesCritical {
hasFailingCriticalAttr := false
var statusFailingCriticalAttr pkg.AttributeStatus
// This is the only case where individual attributes need not be considered
if statusFilterAttributes == pkg.MetricsStatusFilterAttributesAll && repeatNotifications {
return pkg.DeviceStatusHas(device.DeviceStatus, requiredDeviceStatus)
}
for attrId, attrData := range smartAttrs.Attributes {
//find failing attribute
if attrData.GetStatus() == pkg.AttributeStatusPassed {
continue //skip all passing attributes
}
var failingAttributes []string
for attrId, attrData := range smartAttrs.Attributes {
var status pkg.AttributeStatus = attrData.GetStatus()
if status == pkg.AttributeStatusPassed {
continue
}
// merge the status's of all critical attributes
statusFailingCriticalAttr = pkg.AttributeStatusSet(statusFailingCriticalAttr, attrData.GetStatus())
//found a failing attribute, see if its critical
if device.IsScsi() && thresholds.ScsiMetadata[attrId].Critical {
hasFailingCriticalAttr = true
} else if device.IsNvme() && thresholds.NmveMetadata[attrId].Critical {
hasFailingCriticalAttr = true
if statusFilterAttributes == pkg.MetricsStatusFilterAttributesCritical {
critical := false
if device.IsScsi() {
critical = thresholds.ScsiMetadata[attrId].Critical
} else if device.IsNvme() {
critical = thresholds.NmveMetadata[attrId].Critical
} else {
//this is ATA
attrIdInt, err := strconv.Atoi(attrId)
if err != nil {
continue
}
if thresholds.AtaMetadata[attrIdInt].Critical {
hasFailingCriticalAttr = true
critical = thresholds.AtaMetadata[attrIdInt].Critical
}
if !critical {
continue
}
}
failingAttributes = append(failingAttributes, attrId)
}
if repeatNotifications {
lastPoints, err := deviceRepo.GetSmartAttributeHistoryTail(c, c.Param("wwn"), 1, 1, failingAttributes)
if err == nil && len(lastPoints) > 1 {
for _, attrId := range failingAttributes {
if old, ok := lastPoints[0].Attributes[attrId].(*measurements.SmartAtaAttribute); ok {
if current, ok := smartAttrs.Attributes[attrId].(*measurements.SmartAtaAttribute); ok {
if old.TransformedValue != current.TransformedValue {
return true
}
}
}
if old, ok := lastPoints[0].Attributes[attrId].(*measurements.SmartNvmeAttribute); ok {
if current, ok := smartAttrs.Attributes[attrId].(*measurements.SmartNvmeAttribute); ok {
if old.TransformedValue != current.TransformedValue {
return true
}
}
}
if old, ok := lastPoints[0].Attributes[attrId].(*measurements.SmartScsiAttribute); ok {
if current, ok := smartAttrs.Attributes[attrId].(*measurements.SmartScsiAttribute); ok {
if old.TransformedValue != current.TransformedValue {
return true
}
}
}
}
}
if !hasFailingCriticalAttr {
//no critical attributes are failing, and notifyFilterAttributes == "critical"
return false
} else {
// check if any of the critical attributes have a status that we're looking for
return pkg.AttributeStatusHas(statusFailingCriticalAttr, requiredAttrStatus)
}
return true
} else {
// 2. SKIP - we are processing every attribute.
// 3. check if the device failure level matches the wanted failure level.
return pkg.DeviceStatusHas(device.DeviceStatus, requiredDeviceStatus)
for _, attr := range failingAttributes {
attrStatus := smartAttrs.Attributes[attr].GetStatus()
if pkg.AttributeStatusHas(attrStatus, requiredAttrStatus) {
return true
}
}
}
return false
}
// TODO: include user label for device.