chore: update rpc deployment

spartan/terraform/deploy-rpc/environments/mainnet/main.tf

@@ -58,6 +58,53 @@ locals {
       aztec_docker_image = var.CANONICAL_AZTEC_DOCKER_IMAGE
       hosts              = ["v5.mainnet.rpc.aztec-labs.com", "canonical.mainnet.rpc.aztec-labs.com"]
       storage_size       = "8Gi"
+      min_replicas       = 1
+      force_update       = false
+      recreate_pods      = false
+      log_level          = "info"
+      extra_helm_values = [
+        yamlencode({
+          nodeSelector = {
+            local-ssd = "false"
+            node-type = "network"
+            cores     = "2"
+          }
+          affinity = {
+            nodeAffinity = {
+              preferredDuringSchedulingIgnoredDuringExecution = [
+                {
+                  weight = 100
+                  preference = {
+                    matchExpressions = [
+                      {
+                        key      = "cores"
+                        operator = "In"
+                        values   = ["2"]
+                      }
+                    ]
+                  }
+                }
+              ]
+            }
+          }
+          node = {
+            resources = {
+              requests = {
+                cpu = "1"
+              }
+              limits = {
+                cpu = "2"
+              }
+            }
+            updateStrategy = {
+              type = "RollingUpdate"
+              rollingUpdate = {
+                partition = 0
+              }
+            }
+          }
+        })
+      ]
       env = merge(local.env, {
         ROLLUP_VERSION = ""
       })
@@ -107,10 +154,10 @@ module "environment" {
     google     = google
   }
 
-  NAMESPACE       = "mainnet-rpc"
-  RELEASE_PREFIX  = "mainnet"
-  RPCS            = local.rpcs
-  ALLOW_ANONYMOUS = false
-  CONSUMERS = local.consumers
+  NAMESPACE           = "mainnet-rpc"
+  RELEASE_PREFIX      = "mainnet"
+  RPCS                = local.rpcs
+  ALLOW_ANONYMOUS     = false
+  CONSUMERS           = local.consumers
   IRM_METRICS_ENABLED = true
 }

spartan/terraform/deploy-rpc/modules/environment/main.tf

@@ -61,12 +61,17 @@ module "rpc" {
   RELEASE_PREFIX = var.RELEASE_PREFIX
 
   AZTEC_DOCKER_IMAGE                            = each.value.aztec_docker_image
+  LOG_LEVEL                                     = each.value.log_level
   ENV                                           = each.value.env
   L1_RPC_SECRET_NAME                            = each.value.l1_rpc_secret_name
   L1_CONSENSUS_HOST_URLS_SECRET_NAME            = each.value.l1_consensus_host_urls_secret_name
   L1_CONSENSUS_HOST_API_KEYS_SECRET_NAME        = each.value.l1_consensus_host_api_keys_secret_name
   L1_CONSENSUS_HOST_API_KEY_HEADERS_SECRET_NAME = each.value.l1_consensus_host_api_key_headers_secret_name
   STORAGE_SIZE                                  = each.value.storage_size
+  EXTRA_HELM_VALUES                             = each.value.extra_helm_values
+  MIN_REPLICAS                                  = each.value.min_replicas
+  FORCE_UPDATE                                  = each.value.force_update
+  RECREATE_PODS                                 = each.value.recreate_pods
   OTEL_COLLECTOR_ENDPOINT_GCP_SECRET_NAME       = var.OTEL_COLLECTOR_ENDPOINT_GCP_SECRET_NAME
 
   depends_on = [kubernetes_namespace_v1.rpc]

spartan/terraform/deploy-rpc/modules/environment/variables.tf

@@ -12,12 +12,17 @@ variable "RPCS" {
   description = "RPC instances keyed by public route alias."
   type = map(object({
     aztec_docker_image                            = string
+    log_level                                     = optional(string, "info")
     l1_rpc_secret_name                            = string
     l1_consensus_host_urls_secret_name            = string
     l1_consensus_host_api_keys_secret_name        = string
     l1_consensus_host_api_key_headers_secret_name = string
     hosts                                         = list(string)
     storage_size                                  = string
+    extra_helm_values                             = optional(list(string), [])
+    min_replicas                                  = optional(number, 1)
+    force_update                                  = optional(bool, true)
+    recreate_pods                                 = optional(bool, true)
     env                                           = map(string)
   }))
 }

spartan/terraform/deploy-rpc/modules/rpc/main.tf

@@ -26,20 +26,20 @@ resource "helm_release" "rpc" {
   namespace        = var.NAMESPACE
   create_namespace = false
   upgrade_install  = true
-  force_update     = true
-  recreate_pods    = true
+  force_update     = var.FORCE_UPDATE
+  recreate_pods    = var.RECREATE_PODS
   reuse_values     = false
   timeout          = 3600 # 1h to sync
   wait             = true
   wait_for_jobs    = true
   take_ownership   = true
 
-  values = [
+  values = concat([
     file("${path.module}/values/prod.yaml"),
     file("${path.module}/values/prod-res.yaml"),
     yamlencode({
       fullnameOverride = local.workload_name
-      replicaCount     = 1
+      replicaCount     = var.MIN_REPLICAS
       extraObjects = concat(
         [
           {
@@ -149,7 +149,7 @@ resource "helm_release" "rpc" {
       }
 
       node = {
-        logLevel = "info"
+        logLevel = var.LOG_LEVEL
         startupProbe = {
           failureThreshold = 120
         }
@@ -212,7 +212,7 @@ resource "helm_release" "rpc" {
         }
       }
     })
-  ]
+  ], var.EXTRA_HELM_VALUES)
 }
 
 resource "kubernetes_manifest" "hpa" {
@@ -229,7 +229,7 @@ resource "kubernetes_manifest" "hpa" {
         kind       = "StatefulSet"
         name       = local.workload_name
       }
-      minReplicas = 1
+      minReplicas = var.MIN_REPLICAS
       maxReplicas = 4
       behavior = {
         scaleUp = {

spartan/terraform/deploy-rpc/modules/rpc/variables.tf

@@ -13,6 +13,12 @@ variable "RELEASE_PREFIX" {
   type        = string
 }
 
+variable "LOG_LEVEL" {
+  description = "Default and per-module log levels for the RPC node."
+  type        = string
+  default     = "info"
+}
+
 variable "AZTEC_DOCKER_IMAGE" {
   description = "Aztec Docker image in repository:tag form."
   type        = string
@@ -53,6 +59,30 @@ variable "STORAGE_SIZE" {
   type        = string
 }
 
+variable "EXTRA_HELM_VALUES" {
+  description = "Additional YAML values applied after the standard RPC values."
+  type        = list(string)
+  default     = []
+}
+
+variable "MIN_REPLICAS" {
+  description = "Minimum number of RPC pods kept ready by the HPA."
+  type        = number
+  default     = 1
+}
+
+variable "FORCE_UPDATE" {
+  description = "Whether Helm may replace resources during an upgrade."
+  type        = bool
+  default     = true
+}
+
+variable "RECREATE_PODS" {
+  description = "Whether Helm restarts pods during an upgrade."
+  type        = bool
+  default     = true
+}
+
 variable "OTEL_COLLECTOR_ENDPOINT_GCP_SECRET_NAME" {
   description = "GCP Secret Manager secret containing the OpenTelemetry collector base URL."
   type        = string