feat(supervisor): add per-machine-preset resource request ratios

Adds support to configure CPU/memory request ratios per machine preset.
Falls back to the global request ratio configs if no specific override
is specified.

Runs across different machine presets have different usage patters, so
this enables use to manage the available capacity better.

Author: myftija

apps/supervisor/src/env.ts

@@ -89,6 +89,25 @@ const Env = z.object({
   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
+
+  // Per-preset overrides of the global KUBERNETES_CPU_REQUEST_RATIO
+  KUBERNETES_CPU_REQUEST_RATIO_MICRO: z.coerce.number().min(0).max(1).optional(),
+  KUBERNETES_CPU_REQUEST_RATIO_SMALL_1X: z.coerce.number().min(0).max(1).optional(),
+  KUBERNETES_CPU_REQUEST_RATIO_SMALL_2X: z.coerce.number().min(0).max(1).optional(),
+  KUBERNETES_CPU_REQUEST_RATIO_MEDIUM_1X: z.coerce.number().min(0).max(1).optional(),
+  KUBERNETES_CPU_REQUEST_RATIO_MEDIUM_2X: z.coerce.number().min(0).max(1).optional(),
+  KUBERNETES_CPU_REQUEST_RATIO_LARGE_1X: z.coerce.number().min(0).max(1).optional(),
+  KUBERNETES_CPU_REQUEST_RATIO_LARGE_2X: z.coerce.number().min(0).max(1).optional(),
+
+  // Per-preset overrides of the global KUBERNETES_MEMORY_REQUEST_RATIO
+  KUBERNETES_MEMORY_REQUEST_RATIO_MICRO: z.coerce.number().min(0).max(1).optional(),
+  KUBERNETES_MEMORY_REQUEST_RATIO_SMALL_1X: z.coerce.number().min(0).max(1).optional(),
+  KUBERNETES_MEMORY_REQUEST_RATIO_SMALL_2X: z.coerce.number().min(0).max(1).optional(),
+  KUBERNETES_MEMORY_REQUEST_RATIO_MEDIUM_1X: z.coerce.number().min(0).max(1).optional(),
+  KUBERNETES_MEMORY_REQUEST_RATIO_MEDIUM_2X: z.coerce.number().min(0).max(1).optional(),
+  KUBERNETES_MEMORY_REQUEST_RATIO_LARGE_1X: z.coerce.number().min(0).max(1).optional(),
+  KUBERNETES_MEMORY_REQUEST_RATIO_LARGE_2X: z.coerce.number().min(0).max(1).optional(),
+
   KUBERNETES_MEMORY_OVERHEAD_GB: z.coerce.number().min(0).optional(), // Optional memory overhead to add to the limit in GB
   KUBERNETES_SCHEDULER_NAME: z.string().optional(), // Custom scheduler name for pods
   KUBERNETES_LARGE_MACHINE_POOL_LABEL: z.string().optional(), // if set, large-* presets affinity for machinepool=<value>

apps/supervisor/src/workloadManager/kubernetes.ts

@@ -4,7 +4,12 @@ import {
   type WorkloadManagerCreateOptions,
   type WorkloadManagerOptions,
 } from "./types.js";
-import type { EnvironmentType, MachinePreset, PlacementTag } from "@trigger.dev/core/v3";
+import type {
+  EnvironmentType,
+  MachinePreset,
+  MachinePresetName,
+  PlacementTag,
+} from "@trigger.dev/core/v3";
 import { PlacementTagProcessor } from "@trigger.dev/core/v3/serverOnly";
 import { env } from "../env.js";
 import { type K8sApi, createK8sApi, type k8s } from "../clients/kubernetes.js";
@@ -14,6 +19,26 @@ type ResourceQuantities = {
   [K in "cpu" | "memory" | "ephemeral-storage"]?: string;
 };
 
+const cpuRequestRatioByMachinePreset: Record<MachinePresetName, number | undefined> = {
+  micro: env.KUBERNETES_CPU_REQUEST_RATIO_MICRO,
+  "small-1x": env.KUBERNETES_CPU_REQUEST_RATIO_SMALL_1X,
+  "small-2x": env.KUBERNETES_CPU_REQUEST_RATIO_SMALL_2X,
+  "medium-1x": env.KUBERNETES_CPU_REQUEST_RATIO_MEDIUM_1X,
+  "medium-2x": env.KUBERNETES_CPU_REQUEST_RATIO_MEDIUM_2X,
+  "large-1x": env.KUBERNETES_CPU_REQUEST_RATIO_LARGE_1X,
+  "large-2x": env.KUBERNETES_CPU_REQUEST_RATIO_LARGE_2X,
+};
+
+const memoryRequestRatioByMachinePreset: Record<MachinePresetName, number | undefined> = {
+  micro: env.KUBERNETES_MEMORY_REQUEST_RATIO_MICRO,
+  "small-1x": env.KUBERNETES_MEMORY_REQUEST_RATIO_SMALL_1X,
+  "small-2x": env.KUBERNETES_MEMORY_REQUEST_RATIO_SMALL_2X,
+  "medium-1x": env.KUBERNETES_MEMORY_REQUEST_RATIO_MEDIUM_1X,
+  "medium-2x": env.KUBERNETES_MEMORY_REQUEST_RATIO_MEDIUM_2X,
+  "large-1x": env.KUBERNETES_MEMORY_REQUEST_RATIO_LARGE_1X,
+  "large-2x": env.KUBERNETES_MEMORY_REQUEST_RATIO_LARGE_2X,
+};
+
 export class KubernetesWorkloadManager implements WorkloadManager {
   private readonly logger = new SimpleStructuredLogger("kubernetes-workload-provider");
   private k8s: K8sApi;
@@ -321,8 +346,11 @@ export class KubernetesWorkloadManager implements WorkloadManager {
   }
 
   #getResourceRequestsForMachine(preset: MachinePreset): ResourceQuantities {
-    const cpuRequest = preset.cpu * this.cpuRequestRatio;
-    const memoryRequest = preset.memory * this.memoryRequestRatio;
+    const cpuRatio = cpuRequestRatioByMachinePreset[preset.name] ?? this.cpuRequestRatio;
+    const memoryRatio = memoryRequestRatioByMachinePreset[preset.name] ?? this.memoryRequestRatio;
+
+    const cpuRequest = preset.cpu * cpuRatio;
+    const memoryRequest = preset.memory * memoryRatio;
 
     // Clamp between min and max
     const clampedCpu = this.clamp(cpuRequest, this.cpuRequestMinCores, preset.cpu);