From 54335e8f419a6a7ca599ffacb50ad36d892819eb Mon Sep 17 00:00:00 2001
From: caydyan <notevol@qq.com>
Date: Sun, 14 Jun 2026 11:08:33 +0800
Subject: [PATCH 1/6] Implement RunPod cloud adapter

---
 packages/cloud/runpod/README.md         |  17 +-
 packages/cloud/runpod/src/index.test.ts | 242 +++++++++++++-
 packages/cloud/runpod/src/index.ts      | 420 +++++++++++++++++++++---
 3 files changed, 630 insertions(+), 49 deletions(-)

diff --git a/packages/cloud/runpod/README.md b/packages/cloud/runpod/README.md
index cc8be9f2..1c4e9b73 100644
--- a/packages/cloud/runpod/README.md
+++ b/packages/cloud/runpod/README.md
@@ -4,10 +4,19 @@ Provides the RunPod (GPU) cloud provider adapter for sh1pt scale and deploy work
 
 ## What it does
 
-- Connects cloud provider credentials and project settings.
-- Supports infrastructure planning, deployment, or status workflows where implemented.
-- Includes a connection flow for account or credential setup.
-- Includes setup guidance for required credentials or provider configuration.
+- Connects to RunPod through the GraphQL API with `RUNPOD_API_KEY`.
+- Quotes GPU pods from either explicit `hourlyPrice` config or RunPod `gpuTypes` pricing.
+- Provisions on-demand GPU pods with `podFindAndDeployOnDemand`.
+- Lists account pods, checks a pod by ID, and terminates pods with `podTerminate`.
+- Requires `imageName` for real provisioning and supports `maxHourlyPrice` guardrails before any pod is created.
+
+Common config fields:
+
+- `gpuTypeId`: RunPod GPU type ID, such as `NVIDIA RTX A6000`.
+- `imageName`: container image for real pod creation, such as `runpod/pytorch`.
+- `cloudType`: `ALL`, `COMMUNITY`, or `SECURE`.
+- `hourlyPrice`: optional explicit offline quote value per GPU.
+- `ports`, `dockerArgs`, `volumeInGb`, `containerDiskInGb`, `minVcpuCount`, `minMemoryInGb`, `volumeMountPath`, `networkVolumeId`, and `env`.
 
 ## Package
 
diff --git a/packages/cloud/runpod/src/index.test.ts b/packages/cloud/runpod/src/index.test.ts
index 74d1cb34..346f3e2c 100644
--- a/packages/cloud/runpod/src/index.test.ts
+++ b/packages/cloud/runpod/src/index.test.ts
@@ -1,8 +1,242 @@
 import { contractTestCloud } from '@profullstack/sh1pt-core/testing';
-import cloud from './index.js';
+import { afterEach, describe, expect, it, vi } from 'vitest';
+import adapter from './index.js';
 
-contractTestCloud(cloud, {
-  sampleConfig: { cloudType: 'COMMUNITY' },
-  sampleSpec: { kind: 'gpu', gpu: { model: 'A100-40GB', count: 1 } },
+const API = 'https://api.runpod.io/graphql';
+
+afterEach(() => {
+  vi.unstubAllGlobals();
+});
+
+describe('RunPod cloud adapter', () => {
+  it('connects by querying the authenticated RunPod account', async () => {
+    const fetchMock = vi.fn(async (url: string, init: RequestInit) => {
+      expect(url).toBe(API);
+      expect(init.method).toBe('POST');
+      expect(init.headers).toEqual(expect.objectContaining({
+        Authorization: 'Bearer test-token',
+        'Content-Type': 'application/json',
+      }));
+      expect(JSON.parse(String(init.body)).query).toContain('myself');
+      return graphql({ myself: { id: 'user-1', email: 'ops@example.com' } });
+    });
+    vi.stubGlobal('fetch', fetchMock);
+
+    await expect(adapter.connect(connectCtx(), {})).resolves.toEqual({ accountId: 'user-1' });
+  });
+
+  it('quotes from configured hourly pricing without calling RunPod', async () => {
+    const fetchMock = vi.fn();
+    vi.stubGlobal('fetch', fetchMock);
+
+    const quote = await adapter.quote(
+      connectCtx(),
+      { kind: 'gpu', gpu: { model: 'NVIDIA RTX A6000', count: 2 } },
+      { hourlyPrice: 0.49, gpuTypeId: 'NVIDIA RTX A6000', cloudType: 'COMMUNITY' },
+    );
+
+    expect(quote).toMatchObject({
+      hourly: 0.98,
+      monthly: 715.4,
+      provider: 'runpod',
+      currency: 'USD',
+      sku: 'NVIDIA RTX A6000 x2',
+      availabilityZone: 'COMMUNITY',
+    });
+    expect(fetchMock).not.toHaveBeenCalled();
+  });
+
+  it('quotes from RunPod GPU type pricing when hourlyPrice is omitted', async () => {
+    vi.stubGlobal('fetch', vi.fn(async () => graphql({
+      gpuTypes: [{
+        id: 'NVIDIA RTX A6000',
+        displayName: 'RTX A6000',
+        communityPrice: 0.44,
+        securePrice: 0.79,
+        communitySpotPrice: 0.22,
+      }],
+    })));
+
+    const quote = await adapter.quote(
+      connectCtx(),
+      { kind: 'gpu', gpu: { model: 'RTX A6000', count: 2 }, spotOk: true },
+      { cloudType: 'COMMUNITY' },
+    );
+
+    expect(quote.hourly).toBe(0.44);
+    expect(quote.spot).toBe(true);
+  });
+
+  it('creates a RunPod pod through GraphQL', async () => {
+    const fetchMock = vi.fn(async (_url: string, init: RequestInit) => {
+      const body = JSON.parse(String(init.body));
+      expect(body.query).toContain('podFindAndDeployOnDemand');
+      expect(body.variables.input).toEqual({
+        cloudType: 'SECURE',
+        gpuCount: 1,
+        gpuTypeId: 'NVIDIA RTX A6000',
+        name: 'trainer',
+        imageName: 'runpod/pytorch',
+        ports: '8888/http',
+        volumeInGb: 40,
+        containerDiskInGb: 20,
+        minVcpuCount: 4,
+        minMemoryInGb: 16,
+        env: [{ key: 'JUPYTER_PASSWORD', value: 'secret' }],
+      });
+      return graphql({
+        podFindAndDeployOnDemand: pod({
+          id: 'pod-1',
+          name: 'trainer',
+          desiredStatus: 'RUNNING',
+          imageName: 'runpod/pytorch',
+          costPerHr: 0.79,
+        }),
+      });
+    });
+    vi.stubGlobal('fetch', fetchMock);
+
+    const instance = await adapter.provision(
+      provisionCtx(),
+      { kind: 'gpu', gpu: { model: 'NVIDIA RTX A6000', count: 1 }, cpu: 4, memory: 16, maxHourlyPrice: 1 },
+      {
+        cloudType: 'SECURE',
+        gpuTypeId: 'NVIDIA RTX A6000',
+        imageName: 'runpod/pytorch',
+        name: 'trainer',
+        hourlyPrice: 0.79,
+        ports: '8888/http',
+        volumeInGb: 40,
+        containerDiskInGb: 20,
+        env: { JUPYTER_PASSWORD: 'secret' },
+      },
+    );
+
+    expect(instance).toMatchObject({
+      id: 'pod-1',
+      kind: 'gpu',
+      status: 'running',
+      hourlyRate: 0.79,
+      sku: 'runpod/pytorch',
+    });
+  });
+
+  it('requires imageName before creating a real pod', async () => {
+    const fetchMock = vi.fn();
+    vi.stubGlobal('fetch', fetchMock);
+
+    await expect(adapter.provision(
+      provisionCtx(),
+      { kind: 'gpu', gpu: { model: 'NVIDIA RTX A6000', count: 1 } },
+      { hourlyPrice: 0.5 },
+    )).rejects.toThrow('config.imageName is required');
+    expect(fetchMock).not.toHaveBeenCalled();
+  });
+
+  it('honors maxHourlyPrice before provisioning', async () => {
+    const fetchMock = vi.fn();
+    vi.stubGlobal('fetch', fetchMock);
+
+    await expect(adapter.provision(
+      provisionCtx(),
+      { kind: 'gpu', gpu: { model: 'NVIDIA RTX A6000', count: 1 }, maxHourlyPrice: 0.01 },
+      { hourlyPrice: 0.5, imageName: 'runpod/pytorch' },
+    )).rejects.toThrow('exceeds maxHourlyPrice');
+    expect(fetchMock).not.toHaveBeenCalled();
+  });
+
+  it('lists pods from the authenticated account', async () => {
+    vi.stubGlobal('fetch', vi.fn(async () => graphql({
+      myself: {
+        pods: [
+          pod({ id: 'pod-1', name: 'trainer', desiredStatus: 'RUNNING', costPerHr: 0.44 }),
+          pod({ id: 'pod-2', name: 'stopped', desiredStatus: 'EXITED', costPerHr: 0 }),
+        ],
+      },
+    })));
+
+    const instances = await adapter.list(connectCtx(), {});
+
+    expect(instances.map((instance) => [instance.id, instance.status])).toEqual([
+      ['pod-1', 'running'],
+      ['pod-2', 'stopped'],
+    ]);
+  });
+
+  it('checks status for a single pod', async () => {
+    const fetchMock = vi.fn(async (_url: string, init: RequestInit) => {
+      const body = JSON.parse(String(init.body));
+      expect(body.query).toContain('query Pod');
+      expect(body.variables).toEqual({ input: { podId: 'pod-1' } });
+      return graphql({ pod: pod({ id: 'pod-1', desiredStatus: 'RUNNING', costPerHr: 0.44 }) });
+    });
+    vi.stubGlobal('fetch', fetchMock);
+
+    await expect(adapter.status(connectCtx(), 'pod-1', {})).resolves.toMatchObject({
+      id: 'pod-1',
+      status: 'running',
+      publicIp: '203.0.113.10',
+    });
+  });
+
+  it('terminates a pod through GraphQL', async () => {
+    const fetchMock = vi.fn(async (_url: string, init: RequestInit) => {
+      const body = JSON.parse(String(init.body));
+      expect(body.query).toContain('podTerminate');
+      expect(body.variables).toEqual({ input: { podId: 'pod-1' } });
+      return graphql({ podTerminate: null });
+    });
+    vi.stubGlobal('fetch', fetchMock);
+
+    await adapter.destroy(provisionCtx(), 'pod-1', {});
+
+    expect(fetchMock).toHaveBeenCalledTimes(1);
+  });
+
+  it('reports GraphQL errors', async () => {
+    vi.stubGlobal('fetch', vi.fn(async () => new Response(JSON.stringify({
+      errors: [{ message: 'not authorized' }],
+    }), { status: 200 })));
+
+    await expect(adapter.connect(connectCtx(), {})).rejects.toThrow('RunPod GraphQL failed: not authorized');
+  });
+});
+
+contractTestCloud(adapter, {
+  sampleConfig: { cloudType: 'COMMUNITY', hourlyPrice: 0.001 },
+  sampleSpec: { kind: 'gpu', gpu: { model: 'NVIDIA RTX A6000', count: 1 } },
   requiredSecrets: ['RUNPOD_API_KEY'],
 });
+
+function connectCtx() {
+  return {
+    secret: (key: string) => key === 'RUNPOD_API_KEY' ? 'test-token' : undefined,
+    log: vi.fn(),
+  };
+}
+
+function provisionCtx(dryRun = false) {
+  return {
+    ...connectCtx(),
+    dryRun,
+  };
+}
+
+function graphql(data: unknown) {
+  return new Response(JSON.stringify({ data }));
+}
+
+function pod(overrides: Record<string, unknown> = {}) {
+  return {
+    id: 'pod-1',
+    name: 'trainer',
+    desiredStatus: 'RUNNING',
+    createdAt: '2026-06-14T00:00:00Z',
+    costPerHr: 0.44,
+    imageName: 'runpod/pytorch',
+    runtime: {
+      ports: [{ ip: '203.0.113.10', isIpPublic: true, publicPort: 8888, privatePort: 8888, type: 'http' }],
+    },
+    ...overrides,
+  };
+}
diff --git a/packages/cloud/runpod/src/index.ts b/packages/cloud/runpod/src/index.ts
index 33ac9714..54e366d1 100644
--- a/packages/cloud/runpod/src/index.ts
+++ b/packages/cloud/runpod/src/index.ts
@@ -1,58 +1,230 @@
-import { defineCloud, tokenSetup, type Instance, type Quote, type InstanceSpec } from '@profullstack/sh1pt-core';
+import {
+  defineCloud,
+  tokenSetup,
+  type CloudConnectContext,
+  type Instance,
+  type InstanceSpec,
+  type ProvisionContext,
+  type Quote,
+} from '@profullstack/sh1pt-core';
+
+type Numberish = number | string;
+type CloudType = 'ALL' | 'COMMUNITY' | 'SECURE';
 
-// RunPod — GPU-first, pay-by-the-second. GraphQL API. Two pod types:
-//   - Community Cloud: cheapest, non-SLA, host may reclaim
-//   - Secure Cloud: SLA, redundant, 1.5–2× the cost
 interface Config {
-  apiKey?: string;                     // stored as RUNPOD_API_KEY secret
-  cloudType?: 'COMMUNITY' | 'SECURE';
+  apiBaseUrl?: string;
+  cloudType?: CloudType;
+  gpuTypeId?: string;
+  imageName?: string;
+  name?: string;
+  hourlyPrice?: Numberish;
+  volumeInGb?: Numberish;
+  containerDiskInGb?: Numberish;
+  minVcpuCount?: Numberish;
+  minMemoryInGb?: Numberish;
+  dockerArgs?: string;
+  ports?: string;
+  volumeMountPath?: string;
   networkVolumeId?: string;
+  env?: Record<string, string>;
+}
+
+interface RunpodGraphqlResponse<T> {
+  data?: T;
+  errors?: Array<{ message?: string }>;
+}
+
+interface RunpodGpuType {
+  id?: string;
+  displayName?: string;
+  memoryInGb?: number;
+  communityPrice?: number;
+  securePrice?: number;
+  communitySpotPrice?: number;
+  secureSpotPrice?: number;
+}
+
+interface RunpodPod {
+  id?: string;
+  name?: string;
+  desiredStatus?: string;
+  createdAt?: string;
+  lastStartedAt?: string;
+  costPerHr?: number;
+  adjustedCostPerHr?: number;
+  imageName?: string;
+  machineId?: string;
+  gpuCount?: number;
+  runtime?: {
+    ports?: Array<{
+      ip?: string;
+      isIpPublic?: boolean;
+      publicPort?: number;
+      privatePort?: number;
+      type?: string;
+    }>;
+  };
 }
 
+const API = 'https://api.runpod.io/graphql';
+
+const POD_FIELDS = `
+  id
+  name
+  desiredStatus
+  createdAt
+  lastStartedAt
+  costPerHr
+  adjustedCostPerHr
+  imageName
+  machineId
+  gpuCount
+  runtime {
+    ports {
+      ip
+      isIpPublic
+      publicPort
+      privatePort
+      type
+    }
+  }
+`;
+
 export default defineCloud<Config>({
   id: 'cloud-runpod',
   label: 'RunPod (GPU)',
   supports: ['gpu'],
 
-  async connect(ctx) {
-    const key = ctx.secret('RUNPOD_API_KEY');
-    if (!key) throw new Error('RUNPOD_API_KEY not set — `sh1pt secret set RUNPOD_API_KEY ...`');
-    ctx.log('runpod connected');
-    return { accountId: 'runpod-account' };
+  async connect(ctx, config) {
+    const data = await runpodGraphql<{ myself: { id?: string; email?: string } }>(
+      ctx,
+      config,
+      `query Myself {
+        myself {
+          id
+          email
+        }
+      }`,
+    );
+    return { accountId: data.myself.id ?? data.myself.email ?? 'runpod-account' };
   },
 
-  async quote(ctx, spec) {
-    ctx.log(`runpod quote · gpu=${spec.gpu?.model} x${spec.gpu?.count ?? 1} · spot=${spec.spotOk ?? false}`);
-    // TODO: GraphQL query gpuTypes { id, displayName, communityPrice, securePrice }
-    // Pick the cheapest SKU matching spec.gpu.model + spec.gpu.count.
-    return { hourly: 0, monthly: 0, currency: 'USD', provider: 'runpod', sku: 'stub', spot: !!spec.spotOk };
+  async quote(ctx, spec, config) {
+    const gpu = requireGpuSpec(spec);
+    const gpuTypeId = config.gpuTypeId ?? gpu.model;
+    const hourly = config.hourlyPrice !== undefined
+      ? nonNegativeNumber(config.hourlyPrice, 'hourlyPrice') * gpu.count
+      : await quoteFromApi(ctx, spec, config, gpuTypeId);
+
+    return {
+      hourly,
+      monthly: hourly * 730,
+      currency: 'USD',
+      provider: 'runpod',
+      sku: `${gpuTypeId} x${gpu.count}`,
+      spot: !!spec.spotOk,
+      availabilityZone: config.cloudType ?? 'ALL',
+    } satisfies Quote;
   },
 
-  async provision(ctx, spec, config): Promise<Instance> {
-    if (!spec.gpu) throw new Error('cloud-runpod: spec.gpu is required');
-    if (spec.maxHourlyPrice !== undefined) {
-      ctx.log(`maxHourlyPrice=${spec.maxHourlyPrice} — quote will be validated before launch`);
+  async provision(ctx, spec, config) {
+    const gpu = requireGpuSpec(spec);
+    const quote = await this.quote(ctx, spec, config);
+    if (spec.maxHourlyPrice !== undefined && quote.hourly > spec.maxHourlyPrice) {
+      throw new Error(`RunPod quote ${quote.hourly} USD/hr exceeds maxHourlyPrice ${spec.maxHourlyPrice}`);
     }
-    ctx.log(`runpod provision · ${spec.gpu.count}×${spec.gpu.model} · ${config.cloudType ?? 'COMMUNITY'}`);
+
+    const name = safeName(config.name ?? `sh1pt-runpod-${Date.now().toString(36)}`);
     if (ctx.dryRun) {
-      return stubInstance('dry-run', 'provisioning', spec);
+      return podInstance({
+        id: `dry-run-${name}`,
+        name,
+        desiredStatus: 'CREATED',
+        createdAt: new Date().toISOString(),
+        costPerHr: quote.hourly,
+        gpuCount: gpu.count,
+      }, quote);
     }
-    // TODO: GraphQL mutation podFindAndDeployOnDemand({ gpuCount, gpuTypeId, imageName, ... })
-    return stubInstance(`pod_${Date.now()}`, 'provisioning', spec);
+
+    if (!config.imageName) {
+      throw new Error('config.imageName is required for RunPod provisioning');
+    }
+
+    const input = stripUndefined({
+      cloudType: config.cloudType ?? 'ALL',
+      gpuCount: gpu.count,
+      gpuTypeId: config.gpuTypeId ?? gpu.model,
+      name,
+      imageName: config.imageName,
+      dockerArgs: config.dockerArgs,
+      ports: config.ports,
+      volumeInGb: nonNegativeNumber(config.volumeInGb ?? spec.storage ?? 40, 'volumeInGb'),
+      containerDiskInGb: nonNegativeNumber(config.containerDiskInGb ?? 40, 'containerDiskInGb'),
+      minVcpuCount: optionalPositiveNumber(config.minVcpuCount ?? spec.cpu, 'minVcpuCount'),
+      minMemoryInGb: optionalPositiveNumber(config.minMemoryInGb ?? spec.memory, 'minMemoryInGb'),
+      volumeMountPath: config.volumeMountPath,
+      networkVolumeId: config.networkVolumeId,
+      env: envInput(config.env),
+    });
+
+    const data = await runpodGraphql<{ podFindAndDeployOnDemand: RunpodPod }>(
+      ctx,
+      config,
+      `mutation DeployPod($input: PodFindAndDeployOnDemandInput) {
+        podFindAndDeployOnDemand(input: $input) {
+          ${POD_FIELDS}
+        }
+      }`,
+      { input },
+    );
+
+    return podInstance(data.podFindAndDeployOnDemand, quote);
   },
 
-  async list() {
-    return [];
+  async list(ctx, config) {
+    const data = await runpodGraphql<{ myself: { pods?: RunpodPod[] } }>(
+      ctx,
+      config,
+      `query Pods {
+        myself {
+          pods {
+            ${POD_FIELDS}
+          }
+        }
+      }`,
+    );
+    return (data.myself.pods ?? []).map((pod) => podInstance(pod));
   },
 
-  async destroy(ctx, instanceId) {
-    ctx.log(`runpod destroy pod=${instanceId}`);
-    // TODO: GraphQL mutation podTerminate({ podId })
+  async destroy(ctx, instanceId, config) {
+    if (ctx.dryRun) {
+      ctx.log(`runpod dry-run terminate pod=${instanceId}`);
+      return;
+    }
+
+    await runpodGraphql<{ podTerminate: null }>(
+      ctx,
+      config,
+      `mutation TerminatePod($input: PodTerminateInput!) {
+        podTerminate(input: $input)
+      }`,
+      { input: { podId: instanceId } },
+    );
   },
 
-  async status(ctx, instanceId) {
-    ctx.log(`runpod status pod=${instanceId}`);
-    return stubInstance(instanceId, 'running', { kind: 'gpu' });
+  async status(ctx, instanceId, config) {
+    const data = await runpodGraphql<{ pod?: RunpodPod }>(
+      ctx,
+      config,
+      `query Pod($input: PodFilter) {
+        pod(input: $input) {
+          ${POD_FIELDS}
+        }
+      }`,
+      { input: { podId: instanceId } },
+    );
+    if (!data.pod) throw new Error(`RunPod pod not found: ${instanceId}`);
+    return podInstance(data.pod);
   },
 
   setup: tokenSetup({
@@ -60,21 +232,187 @@ export default defineCloud<Config>({
     label: 'RunPod',
     vendorDocUrl: 'https://www.runpod.io/console/user/settings',
     steps: [
-      'Open runpod.io → Settings → API Keys',
-      'Create API key (read + write scope)',
+      'Open runpod.io -> Settings -> API Keys',
+      'Create an API key with pod read/write scope',
       'Copy the key',
-      '⚠ GPU instances bill by the second — always use --max-hourly-price',
+      'Run: sh1pt secret set RUNPOD_API_KEY <key>',
+      'Set maxHourlyPrice when provisioning GPU pods',
     ],
   }),
 });
 
-function stubInstance(id: string, status: Instance['status'], spec: Partial<InstanceSpec>): Instance {
+async function quoteFromApi(
+  ctx: CloudConnectContext,
+  spec: InstanceSpec,
+  config: Config,
+  gpuTypeId: string,
+): Promise<number> {
+  const data = await runpodGraphql<{ gpuTypes: RunpodGpuType[] }>(
+    ctx,
+    config,
+    `query GpuTypes($input: GpuTypeFilter) {
+      gpuTypes(input: $input) {
+        id
+        displayName
+        memoryInGb
+        communityPrice
+        securePrice
+        communitySpotPrice
+        secureSpotPrice
+      }
+    }`,
+    { input: { id: gpuTypeId } },
+  );
+  const selected = selectGpuType(data.gpuTypes, gpuTypeId);
+  const price = priceForGpu(selected, config.cloudType ?? 'ALL', !!spec.spotOk);
+  return price * (spec.gpu?.count ?? 1);
+}
+
+async function runpodGraphql<T>(
+  ctx: CloudConnectContext | ProvisionContext,
+  config: Config,
+  query: string,
+  variables?: Record<string, unknown>,
+): Promise<T> {
+  const response = await fetch(config.apiBaseUrl ?? API, {
+    method: 'POST',
+    headers: {
+      Authorization: `Bearer ${requireToken(ctx)}`,
+      'Content-Type': 'application/json',
+    },
+    body: JSON.stringify({ query, variables }),
+  });
+
+  const text = await response.text();
+  let payload: RunpodGraphqlResponse<T>;
+  try {
+    payload = text ? JSON.parse(text) : {};
+  } catch (error) {
+    if (response.ok) throw error;
+    throw new Error(`RunPod GraphQL failed: ${response.status} ${text || response.statusText}`);
+  }
+
+  if (!response.ok) {
+    throw new Error(`RunPod GraphQL failed: ${response.status} ${graphqlError(payload) || response.statusText}`);
+  }
+  if (payload.errors?.length) {
+    throw new Error(`RunPod GraphQL failed: ${graphqlError(payload)}`);
+  }
+  if (!payload.data) {
+    throw new Error('RunPod GraphQL response did not include data');
+  }
+  return payload.data;
+}
+
+function requireToken(ctx: CloudConnectContext | ProvisionContext): string {
+  const token = ctx.secret('RUNPOD_API_KEY');
+  if (!token) throw new Error('RUNPOD_API_KEY not set - run: sh1pt secret set RUNPOD_API_KEY <key>');
+  return token;
+}
+
+function requireGpuSpec(spec: InstanceSpec): NonNullable<InstanceSpec['gpu']> {
+  if (spec.kind !== 'gpu') throw new Error(`cloud-runpod supports gpu specs only, got ${spec.kind}`);
+  if (!spec.gpu?.model) throw new Error('cloud-runpod: spec.gpu.model is required');
+  if (!spec.gpu.count || spec.gpu.count < 1) throw new Error('cloud-runpod: spec.gpu.count must be >= 1');
+  return spec.gpu;
+}
+
+function selectGpuType(gpus: RunpodGpuType[], requested: string): RunpodGpuType {
+  const normalized = normalize(requested);
+  const selected = gpus.find((gpu) => normalize(gpu.id) === normalized || normalize(gpu.displayName) === normalized) ??
+    gpus.find((gpu) => normalize(gpu.id).includes(normalized) || normalize(gpu.displayName).includes(normalized)) ??
+    gpus[0];
+
+  if (!selected) throw new Error(`RunPod GPU type not found: ${requested}`);
+  return selected;
+}
+
+function priceForGpu(gpu: RunpodGpuType, cloudType: CloudType, spot: boolean): number {
+  const community = spot ? gpu.communitySpotPrice : gpu.communityPrice;
+  const secure = spot ? gpu.secureSpotPrice : gpu.securePrice;
+  const prices = cloudType === 'COMMUNITY'
+    ? [community]
+    : cloudType === 'SECURE'
+      ? [secure]
+      : [community, secure];
+  const price = prices.find((value): value is number => typeof value === 'number' && value >= 0);
+  if (price === undefined) {
+    throw new Error(`RunPod GPU price not available for ${gpu.id ?? gpu.displayName ?? 'selected GPU'}`);
+  }
+  return price;
+}
+
+function podInstance(pod: RunpodPod, quote?: Quote): Instance {
+  const id = pod.id;
+  if (!id) throw new Error('RunPod pod response did not include an id');
+  const publicPort = pod.runtime?.ports?.find((port) => port.isIpPublic && port.ip);
+  const hourlyRate = pod.costPerHr ?? pod.adjustedCostPerHr ?? quote?.hourly ?? 0;
+
   return {
     id,
-    kind: spec.kind ?? 'gpu',
-    status,
-    createdAt: new Date().toISOString(),
-    hourlyRate: 0,
-    currency: 'USD',
+    kind: 'gpu',
+    status: podStatus(pod.desiredStatus),
+    publicIp: publicPort?.ip,
+    createdAt: pod.createdAt ?? pod.lastStartedAt ?? new Date().toISOString(),
+    hourlyRate,
+    currency: quote?.currency ?? 'USD',
+    sku: pod.imageName ?? quote?.sku,
+    tags: pod.name ? [pod.name] : undefined,
   };
 }
+
+function podStatus(status: string | undefined): Instance['status'] {
+  switch (status) {
+    case 'RUNNING':
+      return 'running';
+    case 'EXITED':
+    case 'STOPPED':
+      return 'stopped';
+    case 'TERMINATED':
+      return 'destroyed';
+    case 'ERROR':
+    case 'FAILED':
+      return 'failed';
+    default:
+      return 'provisioning';
+  }
+}
+
+function envInput(env: Record<string, string> | undefined): Array<{ key: string; value: string }> | undefined {
+  if (!env) return undefined;
+  return Object.entries(env).map(([key, value]) => ({ key, value }));
+}
+
+function safeName(value: string): string {
+  return value
+    .toLowerCase()
+    .replace(/[^a-z0-9-]/g, '-')
+    .replace(/-+/g, '-')
+    .replace(/^-+|-+$/g, '')
+    .slice(0, 60) || 'sh1pt-runpod';
+}
+
+function optionalPositiveNumber(value: Numberish | undefined, label: string): number | undefined {
+  if (value === undefined) return undefined;
+  const number = Number(value);
+  if (!Number.isFinite(number) || number <= 0) throw new Error(`RunPod ${label} must be a positive number`);
+  return number;
+}
+
+function nonNegativeNumber(value: Numberish, label: string): number {
+  const number = Number(value);
+  if (!Number.isFinite(number) || number < 0) throw new Error(`RunPod ${label} must be a non-negative number`);
+  return number;
+}
+
+function graphqlError(payload: RunpodGraphqlResponse<unknown>): string {
+  return payload.errors?.map((error) => error.message).filter(Boolean).join('; ') ?? '';
+}
+
+function stripUndefined(value: Record<string, unknown>): Record<string, unknown> {
+  return Object.fromEntries(Object.entries(value).filter(([, item]) => item !== undefined));
+}
+
+function normalize(value: string | undefined): string {
+  return (value ?? '').toLowerCase().replace(/[^a-z0-9]+/g, '');
+}

From 061285b006cba153402ff8cf6f59f5ae691d23f3 Mon Sep 17 00:00:00 2001
From: caydyan <notevol@qq.com>
Date: Sun, 14 Jun 2026 11:16:51 +0800
Subject: [PATCH 2/6] Address RunPod adapter review feedback

---
 packages/cloud/runpod/README.md         |  2 +
 packages/cloud/runpod/src/index.test.ts | 49 ++++++++++++++++++++++++
 packages/cloud/runpod/src/index.ts      | 50 ++++++++++++++++++++-----
 3 files changed, 92 insertions(+), 9 deletions(-)

diff --git a/packages/cloud/runpod/README.md b/packages/cloud/runpod/README.md
index 1c4e9b73..84e798c5 100644
--- a/packages/cloud/runpod/README.md
+++ b/packages/cloud/runpod/README.md
@@ -6,9 +6,11 @@ Provides the RunPod (GPU) cloud provider adapter for sh1pt scale and deploy work
 
 - Connects to RunPod through the GraphQL API with `RUNPOD_API_KEY`.
 - Quotes GPU pods from either explicit `hourlyPrice` config or RunPod `gpuTypes` pricing.
+- Uses the highest available price when `cloudType` is `ALL` so `maxHourlyPrice` is not checked against a lower community-only estimate.
 - Provisions on-demand GPU pods with `podFindAndDeployOnDemand`.
 - Lists account pods, checks a pod by ID, and terminates pods with `podTerminate`.
 - Requires `imageName` for real provisioning and supports `maxHourlyPrice` guardrails before any pod is created.
+- Does not call RunPod during dry-run provisioning; set `hourlyPrice` to simulate a non-zero cost in dry-run output.
 
 Common config fields:
 
diff --git a/packages/cloud/runpod/src/index.test.ts b/packages/cloud/runpod/src/index.test.ts
index 346f3e2c..1d5a626d 100644
--- a/packages/cloud/runpod/src/index.test.ts
+++ b/packages/cloud/runpod/src/index.test.ts
@@ -25,6 +25,12 @@ describe('RunPod cloud adapter', () => {
     await expect(adapter.connect(connectCtx(), {})).resolves.toEqual({ accountId: 'user-1' });
   });
 
+  it('reports a scoped account error when RunPod returns no account', async () => {
+    vi.stubGlobal('fetch', vi.fn(async () => graphql({ myself: null })));
+
+    await expect(adapter.connect(connectCtx(), {})).rejects.toThrow('RunPod account not available');
+  });
+
   it('quotes from configured hourly pricing without calling RunPod', async () => {
     const fetchMock = vi.fn();
     vi.stubGlobal('fetch', fetchMock);
@@ -67,6 +73,25 @@ describe('RunPod cloud adapter', () => {
     expect(quote.spot).toBe(true);
   });
 
+  it('uses the highest available price for ALL cloud type quotes', async () => {
+    vi.stubGlobal('fetch', vi.fn(async () => graphql({
+      gpuTypes: [{
+        id: 'NVIDIA RTX A6000',
+        displayName: 'RTX A6000',
+        communityPrice: 0.44,
+        securePrice: 0.79,
+      }],
+    })));
+
+    const quote = await adapter.quote(
+      connectCtx(),
+      { kind: 'gpu', gpu: { model: 'RTX A6000', count: 1 } },
+      {},
+    );
+
+    expect(quote.hourly).toBe(0.79);
+  });
+
   it('creates a RunPod pod through GraphQL', async () => {
     const fetchMock = vi.fn(async (_url: string, init: RequestInit) => {
       const body = JSON.parse(String(init.body));
@@ -133,6 +158,24 @@ describe('RunPod cloud adapter', () => {
     expect(fetchMock).not.toHaveBeenCalled();
   });
 
+  it('does not call RunPod for dry-run provisioning without hourlyPrice', async () => {
+    const fetchMock = vi.fn();
+    vi.stubGlobal('fetch', fetchMock);
+
+    const instance = await adapter.provision(
+      provisionCtx(true),
+      { kind: 'gpu', gpu: { model: 'NVIDIA RTX A6000', count: 1 } },
+      { name: 'preview' },
+    );
+
+    expect(instance).toMatchObject({
+      id: 'dry-run-preview',
+      status: 'provisioning',
+      hourlyRate: 0,
+    });
+    expect(fetchMock).not.toHaveBeenCalled();
+  });
+
   it('honors maxHourlyPrice before provisioning', async () => {
     const fetchMock = vi.fn();
     vi.stubGlobal('fetch', fetchMock);
@@ -163,6 +206,12 @@ describe('RunPod cloud adapter', () => {
     ]);
   });
 
+  it('reports a scoped account error when listing has no account', async () => {
+    vi.stubGlobal('fetch', vi.fn(async () => graphql({ myself: null })));
+
+    await expect(adapter.list(connectCtx(), {})).rejects.toThrow('RunPod account not available');
+  });
+
   it('checks status for a single pod', async () => {
     const fetchMock = vi.fn(async (_url: string, init: RequestInit) => {
       const body = JSON.parse(String(init.body));
diff --git a/packages/cloud/runpod/src/index.ts b/packages/cloud/runpod/src/index.ts
index 54e366d1..7be0493e 100644
--- a/packages/cloud/runpod/src/index.ts
+++ b/packages/cloud/runpod/src/index.ts
@@ -96,7 +96,7 @@ export default defineCloud<Config>({
   supports: ['gpu'],
 
   async connect(ctx, config) {
-    const data = await runpodGraphql<{ myself: { id?: string; email?: string } }>(
+    const data = await runpodGraphql<{ myself?: { id?: string; email?: string } | null }>(
       ctx,
       config,
       `query Myself {
@@ -106,7 +106,8 @@ export default defineCloud<Config>({
         }
       }`,
     );
-    return { accountId: data.myself.id ?? data.myself.email ?? 'runpod-account' };
+    const account = requireAccount(data.myself);
+    return { accountId: account.id ?? account.email ?? 'runpod-account' };
   },
 
   async quote(ctx, spec, config) {
@@ -129,7 +130,9 @@ export default defineCloud<Config>({
 
   async provision(ctx, spec, config) {
     const gpu = requireGpuSpec(spec);
-    const quote = await this.quote(ctx, spec, config);
+    const quote = ctx.dryRun && config.hourlyPrice === undefined
+      ? quoteFromHourly(spec, config, gpu, 0)
+      : await this.quote(ctx, spec, config);
     if (spec.maxHourlyPrice !== undefined && quote.hourly > spec.maxHourlyPrice) {
       throw new Error(`RunPod quote ${quote.hourly} USD/hr exceeds maxHourlyPrice ${spec.maxHourlyPrice}`);
     }
@@ -182,7 +185,7 @@ export default defineCloud<Config>({
   },
 
   async list(ctx, config) {
-    const data = await runpodGraphql<{ myself: { pods?: RunpodPod[] } }>(
+    const data = await runpodGraphql<{ myself?: { pods?: RunpodPod[] } | null }>(
       ctx,
       config,
       `query Pods {
@@ -193,7 +196,7 @@ export default defineCloud<Config>({
         }
       }`,
     );
-    return (data.myself.pods ?? []).map((pod) => podInstance(pod));
+    return (requireAccount(data.myself).pods ?? []).map((pod) => podInstance(pod));
   },
 
   async destroy(ctx, instanceId, config) {
@@ -268,6 +271,24 @@ async function quoteFromApi(
   return price * (spec.gpu?.count ?? 1);
 }
 
+function quoteFromHourly(
+  spec: InstanceSpec,
+  config: Config,
+  gpu: NonNullable<InstanceSpec['gpu']>,
+  hourly: number,
+): Quote {
+  const gpuTypeId = config.gpuTypeId ?? gpu.model;
+  return {
+    hourly,
+    monthly: hourly * 730,
+    currency: 'USD',
+    provider: 'runpod',
+    sku: `${gpuTypeId} x${gpu.count}`,
+    spot: !!spec.spotOk,
+    availabilityZone: config.cloudType ?? 'ALL',
+  };
+}
+
 async function runpodGraphql<T>(
   ctx: CloudConnectContext | ProvisionContext,
   config: Config,
@@ -331,17 +352,28 @@ function priceForGpu(gpu: RunpodGpuType, cloudType: CloudType, spot: boolean): n
   const community = spot ? gpu.communitySpotPrice : gpu.communityPrice;
   const secure = spot ? gpu.secureSpotPrice : gpu.securePrice;
   const prices = cloudType === 'COMMUNITY'
-    ? [community]
+    ? validPrices([community])
     : cloudType === 'SECURE'
-      ? [secure]
-      : [community, secure];
-  const price = prices.find((value): value is number => typeof value === 'number' && value >= 0);
+      ? validPrices([secure])
+      : validPrices([community, secure]);
+  const price = cloudType === 'ALL' && prices.length > 0 ? Math.max(...prices) : prices[0];
   if (price === undefined) {
     throw new Error(`RunPod GPU price not available for ${gpu.id ?? gpu.displayName ?? 'selected GPU'}`);
   }
   return price;
 }
 
+function validPrices(values: Array<number | undefined>): number[] {
+  return values.filter((value): value is number => typeof value === 'number' && value >= 0);
+}
+
+function requireAccount<T extends object>(account: T | null | undefined): T {
+  if (!account) {
+    throw new Error('RunPod account not available; check RUNPOD_API_KEY permissions');
+  }
+  return account;
+}
+
 function podInstance(pod: RunpodPod, quote?: Quote): Instance {
   const id = pod.id;
   if (!id) throw new Error('RunPod pod response did not include an id');

From 01d613814a4d9f9ae4ede40ff8b08aca2a27e390 Mon Sep 17 00:00:00 2001
From: caydyan <notevol@qq.com>
Date: Sun, 14 Jun 2026 11:21:50 +0800
Subject: [PATCH 3/6] Avoid default RunPod network volume

---
 packages/cloud/runpod/README.md         |  3 ++-
 packages/cloud/runpod/src/index.test.ts | 33 +++++++++++++++++++++++++
 packages/cloud/runpod/src/index.ts      |  2 +-
 3 files changed, 36 insertions(+), 2 deletions(-)

diff --git a/packages/cloud/runpod/README.md b/packages/cloud/runpod/README.md
index 84e798c5..25cdcc81 100644
--- a/packages/cloud/runpod/README.md
+++ b/packages/cloud/runpod/README.md
@@ -18,7 +18,8 @@ Common config fields:
 - `imageName`: container image for real pod creation, such as `runpod/pytorch`.
 - `cloudType`: `ALL`, `COMMUNITY`, or `SECURE`.
 - `hourlyPrice`: optional explicit offline quote value per GPU.
-- `ports`, `dockerArgs`, `volumeInGb`, `containerDiskInGb`, `minVcpuCount`, `minMemoryInGb`, `volumeMountPath`, `networkVolumeId`, and `env`.
+- `volumeInGb`: optional network volume size; omitted unless explicitly requested through config or the spec storage field.
+- `ports`, `dockerArgs`, `containerDiskInGb`, `minVcpuCount`, `minMemoryInGb`, `volumeMountPath`, `networkVolumeId`, and `env`.
 
 ## Package
 
diff --git a/packages/cloud/runpod/src/index.test.ts b/packages/cloud/runpod/src/index.test.ts
index 1d5a626d..253adf9e 100644
--- a/packages/cloud/runpod/src/index.test.ts
+++ b/packages/cloud/runpod/src/index.test.ts
@@ -146,6 +146,39 @@ describe('RunPod cloud adapter', () => {
     });
   });
 
+  it('omits network volume storage unless explicitly requested', async () => {
+    const fetchMock = vi.fn(async (_url: string, init: RequestInit) => {
+      const body = JSON.parse(String(init.body));
+      expect(body.query).toContain('podFindAndDeployOnDemand');
+      expect(body.variables.input).toEqual(expect.objectContaining({
+        gpuTypeId: 'NVIDIA RTX A6000',
+        imageName: 'runpod/pytorch',
+        containerDiskInGb: 40,
+      }));
+      expect(body.variables.input).not.toHaveProperty('volumeInGb');
+      return graphql({
+        podFindAndDeployOnDemand: pod({
+          id: 'pod-1',
+          imageName: 'runpod/pytorch',
+          costPerHr: 0.5,
+        }),
+      });
+    });
+    vi.stubGlobal('fetch', fetchMock);
+
+    await adapter.provision(
+      provisionCtx(),
+      { kind: 'gpu', gpu: { model: 'NVIDIA RTX A6000', count: 1 } },
+      {
+        gpuTypeId: 'NVIDIA RTX A6000',
+        imageName: 'runpod/pytorch',
+        hourlyPrice: 0.5,
+      },
+    );
+
+    expect(fetchMock).toHaveBeenCalledTimes(1);
+  });
+
   it('requires imageName before creating a real pod', async () => {
     const fetchMock = vi.fn();
     vi.stubGlobal('fetch', fetchMock);
diff --git a/packages/cloud/runpod/src/index.ts b/packages/cloud/runpod/src/index.ts
index 7be0493e..3d21394e 100644
--- a/packages/cloud/runpod/src/index.ts
+++ b/packages/cloud/runpod/src/index.ts
@@ -161,7 +161,7 @@ export default defineCloud<Config>({
       imageName: config.imageName,
       dockerArgs: config.dockerArgs,
       ports: config.ports,
-      volumeInGb: nonNegativeNumber(config.volumeInGb ?? spec.storage ?? 40, 'volumeInGb'),
+      volumeInGb: optionalPositiveNumber(config.volumeInGb ?? spec.storage, 'volumeInGb'),
       containerDiskInGb: nonNegativeNumber(config.containerDiskInGb ?? 40, 'containerDiskInGb'),
       minVcpuCount: optionalPositiveNumber(config.minVcpuCount ?? spec.cpu, 'minVcpuCount'),
       minMemoryInGb: optionalPositiveNumber(config.minMemoryInGb ?? spec.memory, 'minMemoryInGb'),

From 2013f9a7b520ebbd4f1c09088deac27024c5876a Mon Sep 17 00:00:00 2001
From: caydyan <notevol@qq.com>
Date: Sun, 14 Jun 2026 11:24:19 +0800
Subject: [PATCH 4/6] Handle missing RunPod GPU type results

---
 packages/cloud/runpod/README.md         |  1 +
 packages/cloud/runpod/src/index.test.ts | 22 ++++++++++++++++++++++
 packages/cloud/runpod/src/index.ts      |  7 +++----
 3 files changed, 26 insertions(+), 4 deletions(-)

diff --git a/packages/cloud/runpod/README.md b/packages/cloud/runpod/README.md
index 25cdcc81..cc419d13 100644
--- a/packages/cloud/runpod/README.md
+++ b/packages/cloud/runpod/README.md
@@ -7,6 +7,7 @@ Provides the RunPod (GPU) cloud provider adapter for sh1pt scale and deploy work
 - Connects to RunPod through the GraphQL API with `RUNPOD_API_KEY`.
 - Quotes GPU pods from either explicit `hourlyPrice` config or RunPod `gpuTypes` pricing.
 - Uses the highest available price when `cloudType` is `ALL` so `maxHourlyPrice` is not checked against a lower community-only estimate.
+- Fails clearly when the requested GPU type is not returned instead of silently selecting another GPU.
 - Provisions on-demand GPU pods with `podFindAndDeployOnDemand`.
 - Lists account pods, checks a pod by ID, and terminates pods with `podTerminate`.
 - Requires `imageName` for real provisioning and supports `maxHourlyPrice` guardrails before any pod is created.
diff --git a/packages/cloud/runpod/src/index.test.ts b/packages/cloud/runpod/src/index.test.ts
index 253adf9e..4a96d100 100644
--- a/packages/cloud/runpod/src/index.test.ts
+++ b/packages/cloud/runpod/src/index.test.ts
@@ -92,6 +92,28 @@ describe('RunPod cloud adapter', () => {
     expect(quote.hourly).toBe(0.79);
   });
 
+  it('reports a clear error when RunPod returns no GPU types', async () => {
+    vi.stubGlobal('fetch', vi.fn(async () => graphql({ gpuTypes: null })));
+
+    await expect(adapter.quote(
+      connectCtx(),
+      { kind: 'gpu', gpu: { model: 'RTX A6000', count: 1 } },
+      {},
+    )).rejects.toThrow('RunPod GPU type not found: RTX A6000');
+  });
+
+  it('does not silently fall back to a different GPU type', async () => {
+    vi.stubGlobal('fetch', vi.fn(async () => graphql({
+      gpuTypes: [{ id: 'NVIDIA A100', displayName: 'A100', communityPrice: 1.25 }],
+    })));
+
+    await expect(adapter.quote(
+      connectCtx(),
+      { kind: 'gpu', gpu: { model: 'RTX A6000', count: 1 } },
+      {},
+    )).rejects.toThrow('RunPod GPU type not found: RTX A6000');
+  });
+
   it('creates a RunPod pod through GraphQL', async () => {
     const fetchMock = vi.fn(async (_url: string, init: RequestInit) => {
       const body = JSON.parse(String(init.body));
diff --git a/packages/cloud/runpod/src/index.ts b/packages/cloud/runpod/src/index.ts
index 3d21394e..7f98f56d 100644
--- a/packages/cloud/runpod/src/index.ts
+++ b/packages/cloud/runpod/src/index.ts
@@ -250,7 +250,7 @@ async function quoteFromApi(
   config: Config,
   gpuTypeId: string,
 ): Promise<number> {
-  const data = await runpodGraphql<{ gpuTypes: RunpodGpuType[] }>(
+  const data = await runpodGraphql<{ gpuTypes?: RunpodGpuType[] | null }>(
     ctx,
     config,
     `query GpuTypes($input: GpuTypeFilter) {
@@ -266,7 +266,7 @@ async function quoteFromApi(
     }`,
     { input: { id: gpuTypeId } },
   );
-  const selected = selectGpuType(data.gpuTypes, gpuTypeId);
+  const selected = selectGpuType(data.gpuTypes ?? [], gpuTypeId);
   const price = priceForGpu(selected, config.cloudType ?? 'ALL', !!spec.spotOk);
   return price * (spec.gpu?.count ?? 1);
 }
@@ -341,8 +341,7 @@ function requireGpuSpec(spec: InstanceSpec): NonNullable<InstanceSpec['gpu']> {
 function selectGpuType(gpus: RunpodGpuType[], requested: string): RunpodGpuType {
   const normalized = normalize(requested);
   const selected = gpus.find((gpu) => normalize(gpu.id) === normalized || normalize(gpu.displayName) === normalized) ??
-    gpus.find((gpu) => normalize(gpu.id).includes(normalized) || normalize(gpu.displayName).includes(normalized)) ??
-    gpus[0];
+    gpus.find((gpu) => normalize(gpu.id).includes(normalized) || normalize(gpu.displayName).includes(normalized));
 
   if (!selected) throw new Error(`RunPod GPU type not found: ${requested}`);
   return selected;

From 0477848e574b641626c6c627869b45e5835e2838 Mon Sep 17 00:00:00 2001
From: caydyan <notevol@qq.com>
Date: Sun, 14 Jun 2026 11:31:31 +0800
Subject: [PATCH 5/6] Use on-demand RunPod pricing for guardrails

---
 packages/cloud/runpod/README.md         |  1 +
 packages/cloud/runpod/src/index.test.ts | 29 ++++++++++++++++++++++---
 packages/cloud/runpod/src/index.ts      | 12 +++++-----
 3 files changed, 33 insertions(+), 9 deletions(-)

diff --git a/packages/cloud/runpod/README.md b/packages/cloud/runpod/README.md
index cc419d13..605e7056 100644
--- a/packages/cloud/runpod/README.md
+++ b/packages/cloud/runpod/README.md
@@ -7,6 +7,7 @@ Provides the RunPod (GPU) cloud provider adapter for sh1pt scale and deploy work
 - Connects to RunPod through the GraphQL API with `RUNPOD_API_KEY`.
 - Quotes GPU pods from either explicit `hourlyPrice` config or RunPod `gpuTypes` pricing.
 - Uses the highest available price when `cloudType` is `ALL` so `maxHourlyPrice` is not checked against a lower community-only estimate.
+- Uses on-demand GPU prices for guardrails; spot bids are not sent by this adapter.
 - Fails clearly when the requested GPU type is not returned instead of silently selecting another GPU.
 - Provisions on-demand GPU pods with `podFindAndDeployOnDemand`.
 - Lists account pods, checks a pod by ID, and terminates pods with `podTerminate`.
diff --git a/packages/cloud/runpod/src/index.test.ts b/packages/cloud/runpod/src/index.test.ts
index 4a96d100..b43084d9 100644
--- a/packages/cloud/runpod/src/index.test.ts
+++ b/packages/cloud/runpod/src/index.test.ts
@@ -52,7 +52,7 @@ describe('RunPod cloud adapter', () => {
     expect(fetchMock).not.toHaveBeenCalled();
   });
 
-  it('quotes from RunPod GPU type pricing when hourlyPrice is omitted', async () => {
+  it('quotes from RunPod on-demand GPU type pricing when hourlyPrice is omitted', async () => {
     vi.stubGlobal('fetch', vi.fn(async () => graphql({
       gpuTypes: [{
         id: 'NVIDIA RTX A6000',
@@ -69,8 +69,8 @@ describe('RunPod cloud adapter', () => {
       { cloudType: 'COMMUNITY' },
     );
 
-    expect(quote.hourly).toBe(0.44);
-    expect(quote.spot).toBe(true);
+    expect(quote.hourly).toBe(0.88);
+    expect(quote.spot).toBe(false);
   });
 
   it('uses the highest available price for ALL cloud type quotes', async () => {
@@ -243,6 +243,29 @@ describe('RunPod cloud adapter', () => {
     expect(fetchMock).not.toHaveBeenCalled();
   });
 
+  it('uses on-demand pricing for maxHourlyPrice even when spot is allowed', async () => {
+    const fetchMock = vi.fn(async (_url: string, init: RequestInit) => {
+      const body = JSON.parse(String(init.body));
+      expect(body.query).toContain('GpuTypes');
+      return graphql({
+        gpuTypes: [{
+          id: 'NVIDIA RTX A6000',
+          displayName: 'RTX A6000',
+          communityPrice: 0.44,
+          communitySpotPrice: 0.22,
+        }],
+      });
+    });
+    vi.stubGlobal('fetch', fetchMock);
+
+    await expect(adapter.provision(
+      provisionCtx(),
+      { kind: 'gpu', gpu: { model: 'NVIDIA RTX A6000', count: 1 }, spotOk: true, maxHourlyPrice: 0.3 },
+      { imageName: 'runpod/pytorch', cloudType: 'COMMUNITY' },
+    )).rejects.toThrow('exceeds maxHourlyPrice');
+    expect(fetchMock).toHaveBeenCalledTimes(1);
+  });
+
   it('lists pods from the authenticated account', async () => {
     vi.stubGlobal('fetch', vi.fn(async () => graphql({
       myself: {
diff --git a/packages/cloud/runpod/src/index.ts b/packages/cloud/runpod/src/index.ts
index 7f98f56d..b3e9d838 100644
--- a/packages/cloud/runpod/src/index.ts
+++ b/packages/cloud/runpod/src/index.ts
@@ -123,7 +123,7 @@ export default defineCloud<Config>({
       currency: 'USD',
       provider: 'runpod',
       sku: `${gpuTypeId} x${gpu.count}`,
-      spot: !!spec.spotOk,
+      spot: false,
       availabilityZone: config.cloudType ?? 'ALL',
     } satisfies Quote;
   },
@@ -267,7 +267,7 @@ async function quoteFromApi(
     { input: { id: gpuTypeId } },
   );
   const selected = selectGpuType(data.gpuTypes ?? [], gpuTypeId);
-  const price = priceForGpu(selected, config.cloudType ?? 'ALL', !!spec.spotOk);
+  const price = priceForGpu(selected, config.cloudType ?? 'ALL');
   return price * (spec.gpu?.count ?? 1);
 }
 
@@ -284,7 +284,7 @@ function quoteFromHourly(
     currency: 'USD',
     provider: 'runpod',
     sku: `${gpuTypeId} x${gpu.count}`,
-    spot: !!spec.spotOk,
+    spot: false,
     availabilityZone: config.cloudType ?? 'ALL',
   };
 }
@@ -347,9 +347,9 @@ function selectGpuType(gpus: RunpodGpuType[], requested: string): RunpodGpuType
   return selected;
 }
 
-function priceForGpu(gpu: RunpodGpuType, cloudType: CloudType, spot: boolean): number {
-  const community = spot ? gpu.communitySpotPrice : gpu.communityPrice;
-  const secure = spot ? gpu.secureSpotPrice : gpu.securePrice;
+function priceForGpu(gpu: RunpodGpuType, cloudType: CloudType): number {
+  const community = gpu.communityPrice;
+  const secure = gpu.securePrice;
   const prices = cloudType === 'COMMUNITY'
     ? validPrices([community])
     : cloudType === 'SECURE'

From 5465626db76592a1913012adcad806d6712cca29 Mon Sep 17 00:00:00 2001
From: caydyan <notevol@qq.com>
Date: Sun, 14 Jun 2026 11:37:56 +0800
Subject: [PATCH 6/6] Guard empty RunPod provision response

---
 packages/cloud/runpod/src/index.test.ts | 16 ++++++++++++++++
 packages/cloud/runpod/src/index.ts      |  5 ++++-
 2 files changed, 20 insertions(+), 1 deletion(-)

diff --git a/packages/cloud/runpod/src/index.test.ts b/packages/cloud/runpod/src/index.test.ts
index b43084d9..90bd06e8 100644
--- a/packages/cloud/runpod/src/index.test.ts
+++ b/packages/cloud/runpod/src/index.test.ts
@@ -168,6 +168,22 @@ describe('RunPod cloud adapter', () => {
     });
   });
 
+  it('reports unavailable capacity when RunPod returns no provisioned pod', async () => {
+    const fetchMock = vi.fn(async (_url: string, init: RequestInit) => {
+      const body = JSON.parse(String(init.body));
+      expect(body.query).toContain('podFindAndDeployOnDemand');
+      return graphql({ podFindAndDeployOnDemand: null });
+    });
+    vi.stubGlobal('fetch', fetchMock);
+
+    await expect(adapter.provision(
+      provisionCtx(),
+      { kind: 'gpu', gpu: { model: 'NVIDIA RTX A6000', count: 1 } },
+      { gpuTypeId: 'NVIDIA RTX A6000', imageName: 'runpod/pytorch', hourlyPrice: 0.5 },
+    )).rejects.toThrow('RunPod pod was not provisioned');
+    expect(fetchMock).toHaveBeenCalledTimes(1);
+  });
+
   it('omits network volume storage unless explicitly requested', async () => {
     const fetchMock = vi.fn(async (_url: string, init: RequestInit) => {
       const body = JSON.parse(String(init.body));
diff --git a/packages/cloud/runpod/src/index.ts b/packages/cloud/runpod/src/index.ts
index b3e9d838..6fb11a4f 100644
--- a/packages/cloud/runpod/src/index.ts
+++ b/packages/cloud/runpod/src/index.ts
@@ -170,7 +170,7 @@ export default defineCloud<Config>({
       env: envInput(config.env),
     });
 
-    const data = await runpodGraphql<{ podFindAndDeployOnDemand: RunpodPod }>(
+    const data = await runpodGraphql<{ podFindAndDeployOnDemand?: RunpodPod | null }>(
       ctx,
       config,
       `mutation DeployPod($input: PodFindAndDeployOnDemandInput) {
@@ -181,6 +181,9 @@ export default defineCloud<Config>({
       { input },
     );
 
+    if (!data.podFindAndDeployOnDemand) {
+      throw new Error('RunPod pod was not provisioned; requested GPU capacity may be unavailable');
+    }
     return podInstance(data.podFindAndDeployOnDemand, quote);
   },