Improve cost optimization and documentation (#141)

backend/indexing-job-template.yaml

@@ -18,6 +18,8 @@ spec:
     spec:
       serviceAccountName: PLACEHOLDER
       restartPolicy: OnFailure
+      nodeSelector:
+        workload: graphrag-indexing
       containers:
       - name: graphrag
         image: PLACEHOLDER
@@ -25,10 +27,10 @@ spec:
         resources:
           requests:
             cpu: "6"
-            memory: "24Gi"
+            memory: "56Gi"
           limits:
-            cpu: "10"
-            memory: "32Gi"
+            cpu: "8"
+            memory: "64Gi"
         envFrom:
         - configMapRef:
             name: graphrag

infra/core/aks/aks.bicep

@@ -26,7 +26,7 @@ param systemOsDiskSizeGB int = 128
 
 @description('The number of nodes for the system node pool.')
 @minValue(1)
-@maxValue(50)
+@maxValue(20)
 param systemNodeCount int = 1
 
 @description('The size of the system Virtual Machine.')
@@ -37,8 +37,11 @@ param systemVMSize string = 'standard_d4s_v5'
 @maxValue(50)
 param graphragNodeCount int = 1
 
-@description('The size of the GraphRAG Virtual Machine.')
-param graphragVMSize string = 'standard_e16as_v5' // 16 vcpus, 128 GiB memory
+@description('The VM size of nodes running the GraphRAG API.')
+param graphragVMSize string = 'standard_d8s_v5' // 8 vcpu, 32 GB memory
+
+@description('The VM size of nodes running GraphRAG indexing jobs.')
+param graphragIndexingVMSize string = 'standard_e8s_v5' // 8 vcpus, 64 GB memory
 
 @description('User name for the Linux Virtual Machines.')
 param linuxAdminUsername string = 'azureuser'
@@ -161,6 +164,32 @@ resource aks 'Microsoft.ContainerService/managedClusters@2024-02-01' = {
       type: 'VirtualMachineScaleSets'
     }
   }
+
+  resource graphragIndexingNodePool 'agentPools@2024-02-01' = {
+    name: 'indexing'
+    properties: {
+      enableAutoScaling: true
+      upgradeSettings: {
+        maxSurge: '50%'
+      }
+      minCount: 0
+      maxCount: 10
+      osDiskSizeGB: systemOsDiskSizeGB
+      count: 0
+      vmSize: graphragIndexingVMSize
+      osType: 'Linux'
+      mode: 'User'
+      enableEncryptionAtHost: enableEncryptionAtHost
+      vnetSubnetID: subnetId
+      nodeLabels: {
+        workload: 'graphrag-indexing'
+      }
+      tags: {
+        workload: 'graphrag'
+      }
+      type: 'VirtualMachineScaleSets'
+    }
+  }
 }
 
 resource aksManagedAutoUpgradeSchedule 'Microsoft.ContainerService/managedClusters/maintenanceConfigurations@2024-03-02-preview' = {

infra/main.bicep

@@ -186,6 +186,8 @@ module aks 'core/aks/aks.bicep' = {
   params:{
     clusterName: '${abbrs.containerServiceManagedClusters}${resourceBaseNameFinal}'
     location: location
+    graphragVMSize: 'standard_d8s_v5'           // 8 vcpu, 32 GB memory
+    graphragIndexingVMSize: 'standard_e8s_v5'   // 8 vcpus, 64 GB memory
     sshRSAPublicKey: aksSshRsaPublicKey
     logAnalyticsWorkspaceId: log.outputs.id
     subnetId: vnet.properties.subnets[1].id // aks subnet