feat(supervisor): add configurable resource requests (#2474)

Author: nicktrn

apps/supervisor/src/env.ts

@@ -85,6 +85,10 @@ const Env = z.object({
   KUBERNETES_EPHEMERAL_STORAGE_SIZE_LIMIT: z.string().default("10Gi"),
   KUBERNETES_EPHEMERAL_STORAGE_SIZE_REQUEST: z.string().default("2Gi"),
   KUBERNETES_STRIP_IMAGE_DIGEST: BoolEnv.default(false),
+  KUBERNETES_CPU_REQUEST_MIN_CORES: z.coerce.number().min(0).default(0),
+  KUBERNETES_CPU_REQUEST_RATIO: z.coerce.number().min(0).max(1).default(0.75), // Ratio of CPU limit, so 0.75 = 75% of CPU limit
+  KUBERNETES_MEMORY_REQUEST_MIN_GB: z.coerce.number().min(0).default(0),
+  KUBERNETES_MEMORY_REQUEST_RATIO: z.coerce.number().min(0).max(1).default(1), // Ratio of memory limit, so 1 = 100% of memory limit
 
   // Placement tags settings
   PLACEMENT_TAGS_ENABLED: BoolEnv.default(false),

apps/supervisor/src/workloadManager/kubernetes.ts

@@ -20,6 +20,12 @@ export class KubernetesWorkloadManager implements WorkloadManager {
   private namespace = env.KUBERNETES_NAMESPACE;
   private placementTagProcessor: PlacementTagProcessor;
 
+  // Resource settings
+  private readonly cpuRequestMinCores = env.KUBERNETES_CPU_REQUEST_MIN_CORES;
+  private readonly cpuRequestRatio = env.KUBERNETES_CPU_REQUEST_RATIO;
+  private readonly memoryRequestMinGb = env.KUBERNETES_MEMORY_REQUEST_MIN_GB;
+  private readonly memoryRequestRatio = env.KUBERNETES_MEMORY_REQUEST_RATIO;
+
   constructor(private opts: WorkloadManagerOptions) {
     this.k8s = createK8sApi();
     this.placementTagProcessor = new PlacementTagProcessor({
@@ -63,6 +69,10 @@ export class KubernetesWorkloadManager implements WorkloadManager {
     return imageRef.substring(0, atIndex);
   }
 
+  private clamp(value: number, min: number, max: number): number {
+    return Math.min(Math.max(value, min), max);
+  }
+
   async create(opts: WorkloadManagerCreateOptions) {
     this.logger.log("[KubernetesWorkloadManager] Creating container", { opts });
 
@@ -295,9 +305,16 @@ export class KubernetesWorkloadManager implements WorkloadManager {
   }
 
   #getResourceRequestsForMachine(preset: MachinePreset): ResourceQuantities {
+    const cpuRequest = preset.cpu * this.cpuRequestRatio;
+    const memoryRequest = preset.memory * this.memoryRequestRatio;
+
+    // Clamp between min and max
+    const clampedCpu = this.clamp(cpuRequest, this.cpuRequestMinCores, preset.cpu);
+    const clampedMemory = this.clamp(memoryRequest, this.memoryRequestMinGb, preset.memory);
+
     return {
-      cpu: `${preset.cpu * 0.75}`,
-      memory: `${preset.memory}G`,
+      cpu: `${clampedCpu}`,
+      memory: `${clampedMemory}G`,
     };
   }