From daea8a27490fb613378fc8a831925d7048127172 Mon Sep 17 00:00:00 2001
From: 7840vz <122374011+7840vz@users.noreply.github.com>
Date: Thu, 20 Apr 2023 16:00:26 +0300
Subject: [PATCH 1/2] feat: map ems severity to prom sev

Uses label 'severity' coming from EMS and maps to recommended prometheus scale of info/warning/critical: https://monitoring.mixins.dev/#guidelines-for-alert-names-labels-and-annotations
---
 container/prometheus/ems_alert_rules.yml | 921 +++++++++++++++++++++--
 1 file changed, 865 insertions(+), 56 deletions(-)

diff --git a/container/prometheus/ems_alert_rules.yml b/container/prometheus/ems_alert_rules.yml
index b8cb14ca5..ff6da70bc 100644
--- a/container/prometheus/ems_alert_rules.yml
+++ b/container/prometheus/ems_alert_rules.yml
@@ -1,5 +1,4 @@
 # Example Harvest ems alerts
-
 groups:
   - name: Harvest Ems Alert
     rules:
@@ -8,7 +7,22 @@ groups:
       - alert: Volume Anti-ransomware Monitoring State Changed
         expr: last_over_time(ems_events{message="arw.volume.state"} [4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Anti-ransomware state was changed to [{{ $labels.op }}] for Volume uuid [{{ $labels.volume_uuid }}]."
 
@@ -16,7 +30,22 @@ groups:
       - alert: Storage VM Anti-ransomware Monitoring State Changed
         expr: last_over_time(ems_events{message="arw.vserver.state"} [4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Anti-ransomware state was changed to [{{ $labels.op }}] for SVM name [{{ $labels.vserverName }}]."
 
@@ -24,7 +53,22 @@ groups:
       - alert: Ransomware Activity Detected
         expr: last_over_time(ems_events{message="callhome.arw.activity.seen"} [4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Ransomware activity detected for Volume uuid [{{ $labels.volume_uuid }}]."
 
@@ -32,7 +76,22 @@ groups:
       - alert: NVRAM Battery Low
         expr: last_over_time(ems_events{message="callhome.battery.low"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "NVRAM battery low for Node uuid [{{ $labels.node_uuid }}]"
 
@@ -40,7 +99,22 @@ groups:
       - alert: HA Interconnect Down
         expr: last_over_time(ems_events{message="callhome.hainterconnect.down"} [1d]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "HA interconnect down for Node uuid [{{ $labels.node_uuid }}]."
 
@@ -48,7 +122,22 @@ groups:
       - alert: Shadow Copy Failed
         expr: last_over_time(ems_events{message="cifs.shadowcopy.failure"} [4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Shadow copy failed for Object uuid [{{ $labels.object_uuid }}]."
 
@@ -56,7 +145,22 @@ groups:
       - alert: AWS Credentials Not Initialized
         expr: last_over_time(ems_events{message="cloud.aws.iamNotInitialized"} [5m]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "AWS credentials not initialized on Node uuid [{{ $labels.node_uuid }}]."
 
@@ -64,7 +168,22 @@ groups:
       - alert: Storage Switch Power Supplies Failed
         expr: last_over_time(ems_events{message="cluster.switch.pwr.fail"} [4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Storage switch power supplies failed on Cluster uuid [{{ $labels.cluster_uuid }}]."
 
@@ -72,7 +191,22 @@ groups:
       - alert: Disk Out of Service
         expr: last_over_time(ems_events{message="disk.outOfService"} [5m]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Disk out of service for Node uuid [{{ $labels.node_uuid }}]."
 
@@ -80,7 +214,22 @@ groups:
       - alert: FabricPool Space Usage Limit Reached
         expr: last_over_time(ems_events{message="fabricpool.full"} [4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "FabricPool space usage limit reached for Cluster uuid [{{ $labels.cluster_uuid }}]."
 
@@ -88,7 +237,22 @@ groups:
       - alert: FabricPool Space Usage Limit Nearly Reached
         expr: last_over_time(ems_events{message="fabricpool.nearly.full"} [4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "FabricPool space usage limit nearly reached for Cluster uuid [{{ $labels.cluster_uuid }}]."
 
@@ -96,7 +260,22 @@ groups:
       - alert: Giveback of Aggregate Failed
         expr: last_over_time(ems_events{message="gb.netra.ca.check.failed"} [4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Giveback of aggregate failed of Aggregate uuid [{{ $labels.aggr_uuid }}]."
 
@@ -104,7 +283,22 @@ groups:
       - alert: LUN Destroyed
         expr: last_over_time(ems_events{message="LUN.destroy"} [5m]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "LUN destroyed of Object uuid [{{ $labels.object_uuid }}]."
 
@@ -112,7 +306,22 @@ groups:
       - alert: LUN Offline
         expr: last_over_time(ems_events{message="LUN.offline"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Lun offline for Object uuid [{{ $labels.object_uuid }}]"
 
@@ -120,7 +329,22 @@ groups:
       - alert: Node Root Volume Space Low
         expr: last_over_time(ems_events{message="mgmtgwd.rootvolrec.low.space"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Node root volume space low on Node uuid [{{ $labels.node_uuid }}]"
 
@@ -128,7 +352,22 @@ groups:
       - alert: System Cannot Operate Due to Main Unit Fan Failure
         expr: last_over_time(ems_events{message="monitor.fan.critical"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "System cannot operate due to main unit fan failure for Node uuid [{{ $labels.node_uuid }}]"
 
@@ -136,7 +375,22 @@ groups:
       - alert: Main Unit Fan Failed
         expr: last_over_time(ems_events{message="monitor.fan.failed"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Main unit fan failed for Node uuid [{{ $labels.node_uuid }}]"
 
@@ -144,7 +398,22 @@ groups:
       - alert: Main Unit Fan in Warning State
         expr: last_over_time(ems_events{message="monitor.fan.warning"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Main unit fan in warning state for Node uuid [{{ $labels.node_uuid }}]"
 
@@ -152,7 +421,22 @@ groups:
       - alert: Too Many CIFS Authentication
         expr: last_over_time(ems_events{message="Nblade.cifsManyAuths"}[1d]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Too many CIFS authentication on Object uuid [{{ $labels.object_uuid }}]"
 
@@ -160,7 +444,22 @@ groups:
       - alert: Max Times Open Per File Exceeded
         expr: last_over_time(ems_events{message="Nblade.cifsMaxOpenSameFile"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Max times open per file exceeded on Object uuid [{{ $labels.object_uuid }}]"
 
@@ -168,7 +467,22 @@ groups:
       - alert: Max Sessions Per User Exceeded
         expr: last_over_time(ems_events{message="Nblade.cifsMaxSessPerUsrConn"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Max sessions per user exceeded for Object uuid [{{ $labels.object_uuid }}]"
 
@@ -176,7 +490,22 @@ groups:
       - alert: NetBIOS Name Conflict
         expr: last_over_time(ems_events{message="Nblade.cifsNbNameConflict"}[1d]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "NetBIOS name conflict for Object uuid [{{ $labels.object_uuid }}]"
 
@@ -184,7 +513,22 @@ groups:
       - alert: Nonexistent Admin Share
         expr: last_over_time(ems_events{message="Nblade.cifsNoPrivShare"}[1d]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Nonexistent admin share for Object uuid [{{ $labels.object_uuid }}]"
 
@@ -192,7 +536,22 @@ groups:
       - alert: NFSv4 Store Pool Exhausted
         expr: last_over_time(ems_events{message="Nblade.nfsV4PoolExhaust"}[1d]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "NFSv4 store pool exhausted for Object uuid [{{ $labels.object_uuid }}]"
 
@@ -200,7 +559,22 @@ groups:
       - alert: Unauthorized User Access to Admin Share
         expr: last_over_time(ems_events{message="Nblade.vscanBadUserPrivAccess"}[1d]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Unauthorized user access to admin share for Object uuid [{{ $labels.object_uuid }}]"
 
@@ -208,7 +582,22 @@ groups:
       - alert: Antivirus Server Busy
         expr: last_over_time(ems_events{message="Nblade.vscanConnBackPressure"}[1d]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Antivirus server busy for Object uuid [{{ $labels.object_uuid }}]"
 
@@ -216,7 +605,22 @@ groups:
       - alert: No Registered Scan Engine
         expr: last_over_time(ems_events{message="Nblade.vscanNoRegdScanner"}[1d]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "No registered scan engine for Object uuid [{{ $labels.object_uuid }}]"
 
@@ -224,7 +628,22 @@ groups:
       - alert: No Vscan Connection
         expr: last_over_time(ems_events{message="Nblade.vscanNoScannerConn"}[1d]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "No vscan connection for Object uuid [{{ $labels.object_uuid }}]"
 
@@ -232,7 +651,22 @@ groups:
       - alert: Virus Detected
         expr: last_over_time(ems_events{message="Nblade.vscanVirusDetected"}[1w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Virus detected for Object uuid [{{ $labels.object_uuid }}]"
 
@@ -240,7 +674,22 @@ groups:
       - alert: Non-responsive AntiVirus Server
         expr: last_over_time(ems_events{message="Nblade.vscanConnInactive"}[5m]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Non-responsive antiVirus server for Object uuid [{{ $labels.object_uuid }}]"
 
@@ -248,7 +697,22 @@ groups:
       - alert: NVMe Namespace Destroyed
         expr: last_over_time(ems_events{message="NVMeNS.destroy"}[5m]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "NVMe namespace destroyed for Object uuid [{{ $labels.object_uuid }}]"
 
@@ -256,7 +720,22 @@ groups:
       - alert: NVMe Namespace Offline
         expr: last_over_time(ems_events{message="NVMeNS.offline"}[5m]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "NVMe namespace offline for Object uuid [{{ $labels.object_uuid }}]"
 
@@ -264,7 +743,22 @@ groups:
       - alert: NVMe Namespace Online
         expr: last_over_time(ems_events{message="NVMeNS.online"}[5m]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "NVMe namespace online for Object uuid [{{ $labels.object_uuid }}]"
 
@@ -272,7 +766,22 @@ groups:
       - alert: NVMe-oF Grace Period Active
         expr: last_over_time(ems_events{message="nvmf.graceperiod.active"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "NVMe-oF grace period active for Cluster uuid [{{ $labels.cluster_uuid }}]"
 
@@ -280,7 +789,22 @@ groups:
       - alert: NVMe-oF Grace Period Expired
         expr: last_over_time(ems_events{message="nvmf.graceperiod.expired"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "NVMe-oF grace period expired for Cluster uuid [{{ $labels.cluster_uuid }}]"
 
@@ -288,7 +812,22 @@ groups:
       - alert: NVMe-oF Grace Period Start
         expr: last_over_time(ems_events{message="nvmf.graceperiod.start"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "NVMe-oF grace period start for Cluster uuid [{{ $labels.cluster_uuid }}]"
 
@@ -296,7 +835,22 @@ groups:
       - alert: Cloud Tier Unreachable
         expr: last_over_time(ems_events{message="object.store.unavailable"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Cloud tier unreachable for Node uuid [{{ $labels.node_uuid }}]"
 
@@ -304,7 +858,22 @@ groups:
       - alert: Object Store Host Unresolvable
         expr: last_over_time(ems_events{message="objstore.host.unresolvable"}[1d]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Object store host unresolvable for Node uuid [{{ $labels.node_uuid }}]"
 
@@ -312,7 +881,22 @@ groups:
       - alert: Object Store Intercluster LIF Down
         expr: last_over_time(ems_events{message="objstore.interclusterlifDown"}[1d]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Object store intercluster LIF down for Node uuid [{{ $labels.node_uuid }}]"
 
@@ -320,7 +904,22 @@ groups:
       - alert: Object Store Signature Mismatch
         expr: last_over_time(ems_events{message="osc.signatureMismatch"}[1d]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Object store signature mismatch for Node uuid [{{ $labels.node_uuid }}]"
 
@@ -328,7 +927,22 @@ groups:
       - alert: QoS Monitor Memory Maxed Out
         expr: last_over_time(ems_events{message="qos.monitor.memory.maxed"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "QoS monitor memory maxed out for Object uuid [{{ $labels.object_uuid }}]"
 
@@ -336,7 +950,22 @@ groups:
       - alert: SAN [active-active] State Changed
         expr: last_over_time(ems_events{message="scsiblade.san.config.active"}[5m]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "SAN [active-active] state changed for Object uuid [{{ $labels.object_uuid }}]"
 
@@ -344,7 +973,22 @@ groups:
       - alert: FC Target Port Commands Exceeded
         expr: last_over_time(ems_events{message="scsitarget.fct.port.full"}[5m]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "FC target port commands exceeded for Port Name [{{ $labels.portname }}]"
 
@@ -352,7 +996,22 @@ groups:
       - alert: Shelf Fan Failed
         expr: last_over_time(ems_events{message="ses.status.fanError"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Shelf fan failed for Node uuid [{{ $labels.node_uuid }}]"
 
@@ -360,7 +1019,22 @@ groups:
       - alert: Node Panic
         expr: last_over_time(ems_events{message="sk.panic"}[1d]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Node panic for Node uuid [{{ $labels.node_uuid }}]"
 
@@ -368,7 +1042,22 @@ groups:
       - alert: SnapMirror Relationship Out of Sync
         expr: last_over_time(ems_events{message="sms.status.out.of.sync"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "SnapMirror relationship out of sync for Relationship id [{{ $labels.relationship_id }}]"
 
@@ -376,7 +1065,22 @@ groups:
       - alert: Service Processor Offline
         expr: last_over_time(ems_events{message="sp.ipmi.lost.shutdown"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Service processor offline for Node uuid [{{ $labels.node_uuid }}]"
 
@@ -384,7 +1088,22 @@ groups:
       - alert: Service Processor Not Configured
         expr: last_over_time(ems_events{message="sp.notConfigured"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Service processor not configured for Node uuid [{{ $labels.node_uuid }}]"
 
@@ -392,7 +1111,22 @@ groups:
       - alert: Unassigned Disks
         expr: last_over_time(ems_events{message="unowned.disk.reminder"}[5m]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Unassigned disks for Cluster uuid [{{ $labels.cluster_uuid }}]"
 
@@ -400,15 +1134,45 @@ groups:
       - alert: Storage VM Stop Succeeded
         expr: last_over_time(ems_events{message="vserver.stop.succeeded"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
-          summary: "Storage VM stop succeeded for instance uuid [{{ $labels.inst_uuid }}]"
+          summary: "Storage VM stop succeeded for instance uuid [{{ $labels.instuuid }}]"
 
       # Alert for READDIR timeout ems
       - alert: READDIR Timeout
         expr: last_over_time(ems_events{message="wafl.readdir.expired"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "READDIR timeout for Object uuid [{{ $labels.object_uuid }}]"
 
@@ -416,7 +1180,22 @@ groups:
       - alert: Volume Automatic Resizing Succeeded
         expr: last_over_time(ems_events{message="wafl.vol.autoSize.done"}[5m]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Volume automatic resizing succeeded for Object uuid [{{ $labels.object_uuid }}]"
 
@@ -424,7 +1203,22 @@ groups:
       - alert: Volume Offline
         expr: last_over_time(ems_events{message="wafl.vvol.offline"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Volume offline for instance uuid [{{ $labels.inst_uuid }}]"
 
@@ -432,6 +1226,21 @@ groups:
       - alert: Volume Restricted
         expr: last_over_time(ems_events{message="wafl.vvol.restrict"}[4w]) == 1
         labels:
-          severity: "error"
+          severity: >
+            {{- if $labels.severity -}}
+            {{- if eq $labels.severity "alert" -}}
+            critical
+            {{- else if eq $labels.severity "error" -}}
+            warning
+            {{- else if eq $labels.severity "emergency" -}}
+            critical
+            {{- else if eq $labels.severity "notice" -}}
+            info
+            {{- else if eq $labels.severity "informational" -}}
+            info
+            {{- else -}}
+            {{ $labels.severity }}
+            {{- end -}}
+            {{- end -}}
         annotations:
           summary: "Volume restricted for instance uuid [{{ $labels.inst_uuid }}]"

From 80d9cf722363293f4a9be4e61cd12c63f60500aa Mon Sep 17 00:00:00 2001
From: 7840vz <122374011+7840vz@users.noreply.github.com>
Date: Thu, 20 Apr 2023 16:06:28 +0300
Subject: [PATCH 2/2] fix: revert whitespace

---
 container/prometheus/ems_alert_rules.yml | 1 +
 1 file changed, 1 insertion(+)

diff --git a/container/prometheus/ems_alert_rules.yml b/container/prometheus/ems_alert_rules.yml
index ff6da70bc..880da26b3 100644
--- a/container/prometheus/ems_alert_rules.yml
+++ b/container/prometheus/ems_alert_rules.yml
@@ -1,4 +1,5 @@
 # Example Harvest ems alerts
+
 groups:
   - name: Harvest Ems Alert
     rules: