feat: refactor Grafana alerting rules into a consolidated system module and update individual alert files

hosts/web-arm/modules/grafana/alerting/cpu_usage.nix

@@ -1,12 +1,6 @@
 { lib, pkgs, config, ... }:
 {
-  services.grafana.provision.alerting.rules.settings.groups = [
+  grafanaAlertRuleDefinitions = [
-    {
-      name = "CPUUsageAlerts";
-      folder = "System Alerts";
-      interval = "1m";
-
-      rules = [
     {
       uid = "high-cpu-usage-alert-uid";
       title = "HighCPUUsage";
@@ -61,6 +55,4 @@
       };
     }
   ];
-    }
-  ];
 }

hosts/web-arm/modules/grafana/alerting/disk_usage.nix

@@ -1,18 +1,11 @@
 { lib, pkgs, config, ... }:
 {
-  services.grafana.provision.alerting.rules.settings.groups = [
+  grafanaAlertRuleDefinitions = [
-    {
-      # orgId = 1; # Defaults to 1 for provisioned rules
-      name = "DiskUsageAlerts";      # Name of the rule group
-      folder = "System Alerts";       # The folder these rules belong to in Grafana UI
-      interval = "1m";              # How often to evaluate rules in this group
-
-      rules = [
     {
       uid = "high-disk-usage-alert-uid"; # Optional: provide a stable UID for the rule itself
       title = "HighDiskUsage";          # Name of the alert rule (was 'alert' in vmalert)
 
-          # Condition for the alert to fire. 'C' refers to the refId of the threshold expression.
+      # Condition for the alert to fire. 'D' refers to the refId of the threshold expression.
       condition = "D"; # Condition is now D
       # Removed rule-level relativeTimeRange
 
@@ -80,6 +73,4 @@
       # isPaused = false; # Default is not paused
     }
   ];
-    }
-  ];
 }

hosts/web-arm/modules/grafana/alerting/host_down.nix

@@ -1,12 +1,6 @@
 { lib, pkgs, config, ... }:
 {
-  services.grafana.provision.alerting.rules.settings.groups = [
+  grafanaAlertRuleDefinitions = [
-    {
-      name = "HostStatusAlerts";
-      folder = "System Alerts";
-      interval = "1m";
-
-      rules = [
     {
       uid = "host-down-alert-uid";
       title = "HostDown";
@@ -57,6 +51,4 @@
       };
     }
   ];
-    }
-  ];
 }

hosts/web-arm/modules/grafana/alerting/inode_usage.nix

@@ -1,12 +1,6 @@
 { lib, pkgs, config, ... }:
 {
-  services.grafana.provision.alerting.rules.settings.groups = [
+  grafanaAlertRuleDefinitions = [
-    {
-      name = "InodeUsageAlerts";
-      folder = "System Alerts";
-      interval = "1m";
-
-      rules = [
     {
       uid = "high-inode-usage-alert-uid";
       title = "HighInodeUsage";
@@ -66,6 +60,4 @@
       };
     }
   ];
-    }
-  ];
 }

hosts/web-arm/modules/grafana/alerting/ram_usage.nix

@@ -1,12 +1,6 @@
 { lib, pkgs, config, ... }:
 {
-  services.grafana.provision.alerting.rules.settings.groups = [
+  grafanaAlertRuleDefinitions = [
-    {
-      name = "RAMUsageAlerts";
-      folder = "System Alerts";
-      interval = "1m";
-
-      rules = [
     {
       uid = "high-ram-usage-alert-uid";
       title = "HighRAMUsage";
@@ -64,6 +58,4 @@
       };
     }
   ];
-    }
-  ];
 }

hosts/web-arm/modules/grafana/alerting/system/default.nix

@@ -0,0 +1,21 @@
+{ lib, pkgs, config, ... }:
+let
+  # Import rule definitions from refactored alert files in the parent 'alerting' directory
+  cpuAlertRules = (import ../cpu_usage.nix { inherit lib pkgs config; }).grafanaAlertRuleDefinitions;
+  diskAlertRules = (import ../disk_usage.nix { inherit lib pkgs config; }).grafanaAlertRuleDefinitions;
+  hostDownAlertRules = (import ../host_down.nix { inherit lib pkgs config; }).grafanaAlertRuleDefinitions;
+  inodeAlertRules = (import ../inode_usage.nix { inherit lib pkgs config; }).grafanaAlertRuleDefinitions;
+  ramAlertRules = (import ../ram_usage.nix { inherit lib pkgs config; }).grafanaAlertRuleDefinitions;
+
+  allSystemRules = cpuAlertRules ++ diskAlertRules ++ hostDownAlertRules ++ inodeAlertRules ++ ramAlertRules;
+in
+{
+  services.grafana.provision.alerting.rules.settings.groups = [
+    {
+      name = "System Alerts"; # This is the Grafana alert group name
+      folder = "System Alerts"; # This is the Grafana folder name
+      interval = "1m"; 
+      rules = allSystemRules;
+    }
+  ];
+}

hosts/web-arm/modules/grafana/default.nix

@@ -28,11 +28,13 @@ let
 in
 {
   imports = [
-    ./alerting/disk_usage.nix
+    # Individual alert files removed, now handled by alerting/system/default.nix
-    ./alerting/cpu_usage.nix
+    # ./alerting/disk_usage.nix
-    ./alerting/host_down.nix
+    # ./alerting/cpu_usage.nix
-    ./alerting/inode_usage.nix
+    # ./alerting/host_down.nix
-    ./alerting/ram_usage.nix
+    # ./alerting/inode_usage.nix
+    # ./alerting/ram_usage.nix
+    ./alerting/system/default.nix # Added: Imports the consolidated system alerts module
     # ... other rule files can be added here ...
     ./datasources/victoriametrics.nix
   ];
@@ -99,7 +101,7 @@ in
     };
     provision = {
       alerting = {
-        rules.settings.groups = lib.mkMerge []; # Allows rule groups to be merged
+        rules.settings.groups = lib.mkMerge []; # Allows rule groups to be merged (including the one from system/default.nix)
         contactPoints = {
           settings = {
             apiVersion = 1; # As per Grafana provisioning API