chore: New default resource values for GKE (#10928)

This PR contains more optimal resource values for GKE networks
AztecProtocol · Dec 21, 2024 · 18e38d3 · 18e38d3
1 parent 6aa5369
commit 18e38d3
Show file tree

Hide file tree

Showing 3 changed files with 52 additions and 21 deletions.
diff --git a/spartan/aztec-network/values.yaml b/spartan/aztec-network/values.yaml
@@ -57,8 +57,8 @@ bootNode:
     enabled: "true"
   resources:
     requests:
-      memory: "2Gi"
-      cpu: "200m"
+      memory: "4Gi"
+      cpu: "1"
   deployContracts: true # Set to false to use manual contract addresses
   startupProbe:
     periodSeconds: 10
@@ -113,8 +113,8 @@ validator:
     failureThreshold: 120
   resources:
     requests:
-      memory: "2Gi"
-      cpu: "200m"
+      memory: "4Gi"
+      cpu: "1"
   archiverPollingInterval: 1000
   archiverViemPollingInterval: 1000
   attestationPollingInterval: 1000
@@ -144,8 +144,8 @@ proverNode:
     dataDirectory: ""
   resources:
     requests:
-      memory: "2Gi"
-      cpu: "200m"
+      memory: "4Gi"
+      cpu: "1"
   storage: "8Gi"
   archiverPollingInterval: 1000
   archiverViemPollingInterval: 1000
@@ -165,8 +165,8 @@ pxe:
     failureThreshold: 3
   resources:
     requests:
-      memory: "2Gi"
-      cpu: "200m"
+      memory: "4Gi"
+      cpu: "1"
 
 bot:
   enabled: true
@@ -194,8 +194,8 @@ bot:
     failureThreshold: 3
   resources:
     requests:
-      memory: "2Gi"
-      cpu: "200m"
+      memory: "4Gi"
+      cpu: "1"
 
 ethereum:
   externalHost: ""
@@ -219,8 +219,8 @@ ethereum:
     failureThreshold: 3
   resources:
     requests:
-      memory: "2Gi"
-      cpu: "200m"
+      memory: "4Gi"
+      cpu: "1"
   storage: "80Gi"
   deployL1ContractsPrivateKey:
 
@@ -237,7 +237,10 @@ proverAgent:
   bb:
     hardwareConcurrency: ""
   nodeSelector: {}
-  resources: {}
+  resources:
+    requests:
+      memory: "4Gi"
+      cpu: "1"
   pollInterval: 200
 
 proverBroker:
@@ -253,8 +256,8 @@ proverBroker:
   nodeSelector: {}
   resources:
     requests:
-      memory: "2Gi"
-      cpu: "200m"
+      memory: "4Gi"
+      cpu: "1"
 
 jobs:
   deployL1Verifier:
@@ -268,4 +271,7 @@ faucet:
   accountIndex: 0
   l1Assets: ""
   logLevel: ""
-  resources: {}
+  resources:
+    requests:
+      memory: "2Gi"
+      cpu: "200m"
diff --git a/spartan/aztec-network/values/rc-1.yaml b/spartan/aztec-network/values/rc-1.yaml
@@ -116,9 +116,6 @@ validator:
     - 0x7Ebb637fd68c523613bE51aad27C35C4DB199B9c
     - 0x3c3E2E178C69D4baD964568415a0f0c84fd6320A
 
-  resources:
-    requests:
-      memory: "512Mi"
   validator:
     disabled: false
 
@@ -129,7 +126,7 @@ bootNode:
   storageSize: "100Gi"
 
 proverAgent:
-  replicas: 24
+  replicas: 12
   bb:
     hardwareConcurrency: 31
   gke:

diff --git a/spartan/terraform/gke-cluster/main.tf b/spartan/terraform/gke-cluster/main.tf
@@ -115,6 +115,34 @@ resource "google_container_node_pool" "primary_nodes" {
   }
 }
 
+# Create node pool for simulated aztec nodes (validators, prover nodes, boot nodes)
+resource "google_container_node_pool" "aztec_nodes_simulated" {
+  name     = "aztec-node-pool-simulated"
+  location = var.zone
+  cluster  = google_container_cluster.primary.name
+
+  # Enable autoscaling
+  autoscaling {
+    min_node_count = 1
+    max_node_count = 256
+  }
+
+  # Node configuration
+  node_config {
+    machine_type = "t2d-standard-2"
+
+    service_account = google_service_account.gke_sa.email
+    oauth_scopes = [
+      "https://www.googleapis.com/auth/cloud-platform"
+    ]
+
+    labels = {
+      env = "production"
+    }
+    tags = ["aztec-gke-node", "aztec"]
+  }
+}
+
 # Create node pool for aztec nodes (validators, prover nodes, boot nodes)
 resource "google_container_node_pool" "aztec_nodes" {
   name     = "aztec-node-pool"
@@ -158,7 +186,7 @@ resource "google_container_node_pool" "spot_nodes" {
   # Enable autoscaling
   autoscaling {
     min_node_count = 0
-    max_node_count = 10
+    max_node_count = 1500
   }
 
   # Node configuration