Merge pull request #99 from useblacksmith/scaffold-multi-platform

src: add scaffolding for support multi-platform builds

jest.config.ts

@@ -9,16 +9,10 @@ const config: Config.InitialOptions = {
     '^.+\\.ts$': 'ts-jest',
     '^.+\\.js$': 'babel-jest'
   },
-  transformIgnorePatterns: [
+  transformIgnorePatterns: ['/node_modules/(?!(@buf|@connectrpc)/)'],
-    '/node_modules/(?!(@buf|@connectrpc)/)'
-  ],
   verbose: true,
   collectCoverage: true,
-  collectCoverageFrom: [
+  collectCoverageFrom: ['src/**/*.ts', '!src/**/*.d.ts', '!src/**/__tests__/**']
-    'src/**/*.ts',
-    '!src/**/*.d.ts',
-    '!src/**/__tests__/**'
-  ]
 };
 
 export default config;

package-lock.json

@@ -8,7 +8,7 @@
       "license": "Apache-2.0",
       "dependencies": {
         "@actions/core": "^1.10.1",
-        "@buf/blacksmith_vm-agent.connectrpc_es": "^1.6.1-20250209182455-7d83cfb8ddb1.2",
+        "@buf/blacksmith_vm-agent.connectrpc_es": "^1.6.1-20250211212423-70f4f1344c53.2",
         "@connectrpc/connect": "^1.6.1",
         "@connectrpc/connect-node": "^1.6.1",
         "@docker/actions-toolkit": "0.37.1",
@@ -2304,8 +2304,8 @@
       "dev": true
     },
     "node_modules/@buf/blacksmith_vm-agent.bufbuild_es": {
-      "version": "1.10.0-20250209182455-7d83cfb8ddb1.1",
+      "version": "1.10.0-20250211212423-70f4f1344c53.1",
-      "resolved": "https://buf.build/gen/npm/v1/@buf/blacksmith_vm-agent.bufbuild_es/-/blacksmith_vm-agent.bufbuild_es-1.10.0-20250209182455-7d83cfb8ddb1.1.tgz",
+      "resolved": "https://buf.build/gen/npm/v1/@buf/blacksmith_vm-agent.bufbuild_es/-/blacksmith_vm-agent.bufbuild_es-1.10.0-20250211212423-70f4f1344c53.1.tgz",
       "dependencies": {
         "@buf/googleapis_googleapis.bufbuild_es": "1.10.0-20250203201857-83c0f6c19b2f.1"
       },
@@ -2314,10 +2314,10 @@
       }
     },
     "node_modules/@buf/blacksmith_vm-agent.connectrpc_es": {
-      "version": "1.6.1-20250209182455-7d83cfb8ddb1.2",
+      "version": "1.6.1-20250211212423-70f4f1344c53.2",
-      "resolved": "https://buf.build/gen/npm/v1/@buf/blacksmith_vm-agent.connectrpc_es/-/blacksmith_vm-agent.connectrpc_es-1.6.1-20250209182455-7d83cfb8ddb1.2.tgz",
+      "resolved": "https://buf.build/gen/npm/v1/@buf/blacksmith_vm-agent.connectrpc_es/-/blacksmith_vm-agent.connectrpc_es-1.6.1-20250211212423-70f4f1344c53.2.tgz",
       "dependencies": {
-        "@buf/blacksmith_vm-agent.bufbuild_es": "1.10.0-20250209182455-7d83cfb8ddb1.1",
+        "@buf/blacksmith_vm-agent.bufbuild_es": "1.10.0-20250211212423-70f4f1344c53.1",
         "@buf/googleapis_googleapis.connectrpc_es": "1.6.1-20250203201857-83c0f6c19b2f.2"
       },
       "peerDependencies": {

package.json

@@ -27,7 +27,7 @@
   "packageManager": "yarn@3.6.3",
   "dependencies": {
     "@actions/core": "^1.10.1",
-    "@buf/blacksmith_vm-agent.connectrpc_es": "^1.6.1-20250209182455-7d83cfb8ddb1.2",
+    "@buf/blacksmith_vm-agent.connectrpc_es": "^1.6.1-20250211212423-70f4f1344c53.2",
     "@connectrpc/connect": "^1.6.1",
     "@connectrpc/connect-node": "^1.6.1",
     "@docker/actions-toolkit": "0.37.1",

src/__tests__/blacksmith-builder.test.ts

@@ -3,7 +3,7 @@ import * as main from '../main';
 import * as reporter from '../reporter';
 import {getDockerfilePath} from '../context';
 import * as setupBuilder from '../setup_builder';
-import { Metric_MetricType } from "@buf/blacksmith_vm-agent.bufbuild_es/stickydisk/v1/stickydisk_pb";
+import {Metric_MetricType} from '@buf/blacksmith_vm-agent.bufbuild_es/stickydisk/v1/stickydisk_pb';
 
 jest.mock('@actions/core', () => ({
   debug: jest.fn(),
@@ -34,7 +34,9 @@ jest.mock('../setup_builder', () => ({
   ...jest.requireActual('../setup_builder'),
   startAndConfigureBuildkitd: jest.fn(),
   setupStickyDisk: jest.fn(),
-  getNumCPUs: jest.fn().mockResolvedValue(4)
+  getNumCPUs: jest.fn().mockResolvedValue(4),
+  leaveTailnet: jest.fn().mockResolvedValue(undefined),
+  getTailscaleIP: jest.fn()
 }));
 
 describe('startBlacksmithBuilder', () => {
@@ -42,7 +44,7 @@ describe('startBlacksmithBuilder', () => {
 
   beforeEach(() => {
     jest.clearAllMocks();
-    mockInputs = {nofallback: false};
+    mockInputs = {nofallback: false, platforms: []};
   });
 
   test('should handle missing dockerfile path with nofallback=false', async () => {
@@ -110,7 +112,7 @@ describe('startBlacksmithBuilder', () => {
       buildId: mockBuildId,
       exposeId: mockExposeId
     });
-    expect(setupBuilder.startAndConfigureBuildkitd).toHaveBeenCalledWith(mockParallelism);
+    expect(setupBuilder.startAndConfigureBuildkitd).toHaveBeenCalledWith(mockParallelism, []);
     expect(reporter.reportBuildPushActionFailure).not.toHaveBeenCalled();
   });
 

src/main.ts

@@ -20,37 +20,13 @@ import * as context from './context';
 import {promisify} from 'util';
 import {exec} from 'child_process';
 import * as reporter from './reporter';
-import {setupStickyDisk, startAndConfigureBuildkitd, getNumCPUs} from './setup_builder';
+import {setupStickyDisk, startAndConfigureBuildkitd, getNumCPUs, leaveTailnet} from './setup_builder';
 import {Metric_MetricType} from '@buf/blacksmith_vm-agent.bufbuild_es/stickydisk/v1/stickydisk_pb';
 
 const buildxVersion = 'v0.17.0';
 const mountPoint = '/var/lib/buildkit';
 const execAsync = promisify(exec);
 
-async function joinTailnet(): Promise<void> {
-  const token = process.env.BLACKSMITH_TAILSCALE_TOKEN;
-  if (!token || token === 'unset') {
-    core.debug('BLACKSMITH_TAILSCALE_TOKEN environment variable not set, skipping tailnet join');
-    return;
-  }
-
-  try {
-    await execAsync(`sudo tailscale up --authkey=${token} --hostname=${process.env.VM_ID}`);
-
-    core.info('Successfully joined tailnet');
-  } catch (error) {
-    throw new Error(`Failed to join tailnet: ${error.message}`);
-  }
-}
-
-async function leaveTailnet(): Promise<void> {
-  try {
-    await execAsync('sudo tailscale down');
-  } catch (error) {
-    core.warning(`Error leaving tailnet: ${error.message}`);
-  }
-}
-
 async function setupBuildx(version: string, toolkit: Toolkit): Promise<void> {
   let toolPath;
   const standalone = await toolkit.buildx.isStandalone();
@@ -94,8 +70,6 @@ async function setupBuildx(version: string, toolkit: Toolkit): Promise<void> {
  */
 export async function startBlacksmithBuilder(inputs: context.Inputs): Promise<{addr: string | null; buildId: string | null; exposeId: string}> {
   try {
-    await joinTailnet();
-
     const dockerfilePath = context.getDockerfilePath(inputs);
     if (!dockerfilePath) {
       throw new Error('Failed to resolve dockerfile path');
@@ -107,7 +81,7 @@ export async function startBlacksmithBuilder(inputs: context.Inputs): Promise<{a
     const parallelism = await getNumCPUs();
 
     const buildkitdStartTime = Date.now();
-    const buildkitdAddr = await startAndConfigureBuildkitd(parallelism);
+    const buildkitdAddr = await startAndConfigureBuildkitd(parallelism, inputs.platforms);
     const buildkitdDurationMs = Date.now() - buildkitdStartTime;
     await reporter.reportMetric(Metric_MetricType.BPA_BUILDKITD_READY_DURATION_MS, buildkitdDurationMs);
 
@@ -128,6 +102,8 @@ export async function startBlacksmithBuilder(inputs: context.Inputs): Promise<{a
 
     core.warning(`${errorMessage}. Falling back to a local build.`);
     return {addr: null, buildId: null, exposeId: ''};
+  } finally {
+    await leaveTailnet();
   }
 }
 

src/reporter.ts

@@ -1,20 +1,16 @@
 import * as core from '@actions/core';
-import axios, {AxiosError, AxiosInstance, AxiosResponse } from 'axios';
+import axios, {AxiosError, AxiosInstance, AxiosResponse} from 'axios';
 import axiosRetry from 'axios-retry';
 import {ExportRecordResponse} from '@docker/actions-toolkit/lib/types/buildx/history';
 import FormData from 'form-data';
-import { createClient } from "@connectrpc/connect";
+import {createClient} from '@connectrpc/connect';
-import { createGrpcTransport } from "@connectrpc/connect-node";
+import {createGrpcTransport} from '@connectrpc/connect-node';
-import { StickyDiskService } from "@buf/blacksmith_vm-agent.connectrpc_es/stickydisk/v1/stickydisk_connect";
+import {StickyDiskService} from '@buf/blacksmith_vm-agent.connectrpc_es/stickydisk/v1/stickydisk_connect';
-import { Metric, Metric_MetricType } from "@buf/blacksmith_vm-agent.bufbuild_es/stickydisk/v1/stickydisk_pb";
+import {Metric, Metric_MetricType} from '@buf/blacksmith_vm-agent.bufbuild_es/stickydisk/v1/stickydisk_pb';
 
 // Configure base axios instance for Blacksmith API.
 const createBlacksmithAPIClient = () => {
-  const apiUrl = process.env.BLACKSMITH_BACKEND_URL || (
+  const apiUrl = process.env.BLACKSMITH_BACKEND_URL || (process.env.BLACKSMITH_ENV?.includes('staging') ? 'https://stagingapi.blacksmith.sh' : 'https://api.blacksmith.sh');
-    process.env.BLACKSMITH_ENV?.includes('staging') 
-      ? 'https://stagingapi.blacksmith.sh' 
-      : 'https://api.blacksmith.sh'
-  );
   core.debug(`Using Blacksmith API URL: ${apiUrl}`);
 
   const client = axios.create({
@@ -30,8 +26,7 @@ const createBlacksmithAPIClient = () => {
     retries: 5,
     retryDelay: axiosRetry.exponentialDelay,
     retryCondition: (error: AxiosError) => {
-      return axiosRetry.isNetworkOrIdempotentRequestError(error) || 
+      return axiosRetry.isNetworkOrIdempotentRequestError(error) || (error.response?.status ? error.response.status >= 500 : false);
-             (error.response?.status ? error.response.status >= 500 : false);
     }
   });
 
@@ -41,7 +36,7 @@ const createBlacksmithAPIClient = () => {
 export function createBlacksmithAgentClient() {
   const transport = createGrpcTransport({
     baseUrl: 'http://192.168.127.1:5557',
-    httpVersion: '2',
+    httpVersion: '2'
   });
 
   return createClient(StickyDiskService, transport);
@@ -175,22 +170,19 @@ export async function post(client: AxiosInstance, url: string, formData: FormDat
   return await client.post(url, formData, {
     headers: {
       ...client.defaults.headers.common,
-      ...(formData && { 'Content-Type': 'multipart/form-data' }),
+      ...(formData && {'Content-Type': 'multipart/form-data'})
     },
     signal: options?.signal
   });
 }
 
-export async function reportMetric(
+export async function reportMetric(metricType: Metric_MetricType, value: number): Promise<void> {
-  metricType: Metric_MetricType,
-  value: number
-): Promise<void> {
   try {
     const agentClient = createBlacksmithAgentClient();
 
     const metric = new Metric({
       type: metricType,
-      value: { case: "intValue", value: BigInt(value) }
+      value: {case: 'intValue', value: BigInt(value)}
     });
 
     await agentClient.reportMetric({

src/setup_builder.ts

@@ -10,6 +10,16 @@ const BUILDKIT_DAEMON_ADDR = 'tcp://127.0.0.1:1234';
 const mountPoint = '/var/lib/buildkit';
 const execAsync = promisify(exec);
 
+export async function getTailscaleIP(): Promise<string | null> {
+  try {
+    const {stdout} = await execAsync('tailscale ip -4');
+    return stdout.trim();
+  } catch (error) {
+    core.debug(`Error getting tailscale IP: ${error.message}`);
+    return null;
+  }
+}
+
 async function maybeFormatBlockDevice(device: string): Promise<string> {
   try {
     // Check if device is formatted with ext4
@@ -52,11 +62,11 @@ export async function getNumCPUs(): Promise<number> {
   }
 }
 
-async function writeBuildkitdTomlFile(parallelism: number): Promise<void> {
+async function writeBuildkitdTomlFile(parallelism: number, addr: string): Promise<void> {
   const jsonConfig: TOML.JsonMap = {
     root: '/var/lib/buildkit',
     grpc: {
-      address: [BUILDKIT_DAEMON_ADDR]
+      address: [addr]
     },
     registry: {
       'docker.io': {
@@ -95,13 +105,12 @@ async function writeBuildkitdTomlFile(parallelism: number): Promise<void> {
   }
 }
 
-async function startBuildkitd(parallelism: number): Promise<string> {
+export async function startBuildkitd(parallelism: number, addr: string): Promise<string> {
   try {
-    await writeBuildkitdTomlFile(parallelism);
+    await writeBuildkitdTomlFile(parallelism, addr);
-    const addr = BUILDKIT_DAEMON_ADDR;
 
     const logStream = fs.createWriteStream('buildkitd.log');
-    const buildkitd = spawn('sudo', ['buildkitd', '--debug', '--addr', addr, '--allow-insecure-entitlement', 'security.insecure', '--config=buildkitd.toml', '--allow-insecure-entitlement', 'network.host'], {
+    const buildkitd = spawn('sudo', ['buildkitd', '--debug', '--config=buildkitd.toml', '--allow-insecure-entitlement', 'security.insecure', '--allow-insecure-entitlement', 'network.host'], {
       stdio: ['ignore', 'pipe', 'pipe']
     });
 
@@ -138,20 +147,6 @@ async function startBuildkitd(parallelism: number): Promise<string> {
   }
 }
 
-async function getDiskSize(device: string): Promise<number> {
-  try {
-    const {stdout} = await execAsync(`sudo lsblk -b -n -o SIZE ${device}`);
-    const sizeInBytes = parseInt(stdout.trim(), 10);
-    if (isNaN(sizeInBytes)) {
-      throw new Error('Failed to parse disk size');
-    }
-    return sizeInBytes;
-  } catch (error) {
-    console.error(`Error getting disk size: ${error.message}`);
-    throw error;
-  }
-}
-
 export async function getStickyDisk(options?: {signal?: AbortSignal}): Promise<{expose_id: string; device: string}> {
   const client = await reporter.createBlacksmithAgentClient();
 
@@ -181,14 +176,65 @@ export async function getStickyDisk(options?: {signal?: AbortSignal}): Promise<{
   };
 }
 
+export async function joinTailnet(): Promise<void> {
+  const token = process.env.BLACKSMITH_TAILSCALE_TOKEN;
+  if (!token || token === 'unset') {
+    core.debug('BLACKSMITH_TAILSCALE_TOKEN environment variable not set, skipping tailnet join');
+    return;
+  }
+
+  try {
+    await execAsync(`sudo tailscale up --authkey=${token} --hostname=${process.env.VM_ID}`);
+
+    core.info('Successfully joined tailnet');
+  } catch (error) {
+    throw new Error(`Failed to join tailnet: ${error.message}`);
+  }
+}
+
+export async function leaveTailnet(): Promise<void> {
+  try {
+    // Check if we're part of a tailnet before trying to leave
+    const {stdout} = await execAsync('sudo tailscale status');
+    if (stdout.trim() !== '') {
+      await execAsync('sudo tailscale down');
+      core.debug('Successfully left tailnet');
+    } else {
+      core.debug('Not part of a tailnet, skipping leave');
+    }
+  } catch (error) {
+    core.warning(`Error leaving tailnet: ${error.message}`);
+  }
+}
+
 // buildkitdTimeoutMs states the max amount of time this action will wait for the buildkitd
 // daemon to start have its socket ready. It also additionally governs how long we will wait for
 // the buildkitd workers to be ready.
 const buildkitdTimeoutMs = 30000;
 
-export async function startAndConfigureBuildkitd(parallelism: number): Promise<string> {
+export async function startAndConfigureBuildkitd(parallelism: number, platforms?: string[]): Promise<string> {
-  const buildkitdAddr = await startBuildkitd(parallelism);
+  // For multi-platform builds, we need to use the tailscale IP
-  core.debug(`buildkitd daemon started at addr ${buildkitdAddr}`);
+  let buildkitdAddr = BUILDKIT_DAEMON_ADDR;
+
+  // If we are doing a multi-platform build, we need to join the tailnet and bind buildkitd to the tailscale IP.
+  // We do this so that the remote VM can join the same buildkitd cluster as a worker.
+  if (platforms && platforms.length > 1) {
+    await joinTailnet();
+    const tailscaleIP = await getTailscaleIP();
+    if (!tailscaleIP) {
+      throw new Error('Failed to get tailscale IP for multi-platform build');
+    }
+    buildkitdAddr = `tcp://${tailscaleIP}:1234`;
+    core.info(`Using tailscale IP for multi-platform build: ${buildkitdAddr}`);
+  }
+
+  const addr = await startBuildkitd(parallelism, buildkitdAddr);
+  core.debug(`buildkitd daemon started at addr ${addr}`);
+
+  if (platforms && platforms.length > 1) {
+    // TODO(adityamaru): Queue docker job for multi-platform build with a well known tailscale hostname.
+    // TODO(adityamaru): Wait until the VM joins the tailnet.
+  }
 
   // Check that buildkit instance is ready by querying workers for up to 30s
   const startTimeBuildkitReady = Date.now();
@@ -196,10 +242,12 @@ export async function startAndConfigureBuildkitd(parallelism: number): Promise<s
 
   while (Date.now() - startTimeBuildkitReady < timeoutBuildkitReady) {
     try {
-      const {stdout} = await execAsync(`sudo buildctl --addr ${BUILDKIT_DAEMON_ADDR} debug workers`);
+      const {stdout} = await execAsync(`sudo buildctl --addr ${addr} debug workers`);
       const lines = stdout.trim().split('\n');
-      if (lines.length > 1) {
+      // For multi-platform builds, we need at least 2 workers
-        // Check if we have output lines beyond the header
+      const requiredWorkers = platforms && platforms.length > 1 ? 2 : 1;
+      if (lines.length > requiredWorkers) {
+        core.info(`Found ${lines.length - 1} workers, required ${requiredWorkers}`);
         break;
       }
     } catch (error) {
@@ -210,10 +258,11 @@ export async function startAndConfigureBuildkitd(parallelism: number): Promise<s
 
   // Final check after timeout.
   try {
-    const {stdout} = await execAsync(`sudo buildctl --addr ${BUILDKIT_DAEMON_ADDR} debug workers`);
+    const {stdout} = await execAsync(`sudo buildctl --addr ${addr} debug workers`);
     const lines = stdout.trim().split('\n');
-    if (lines.length <= 1) {
+    const requiredWorkers = platforms && platforms.length > 1 ? 2 : 1;
-      throw new Error('buildkit workers not ready after 15s timeout');
+    if (lines.length <= requiredWorkers) {
+      throw new Error(`buildkit workers not ready after ${buildkitdTimeoutMs}ms timeout. Found ${lines.length - 1} workers, required ${requiredWorkers}`);
     }
   } catch (error) {
     core.warning(`Error checking buildkit workers: ${error.message}`);
@@ -225,7 +274,7 @@ export async function startAndConfigureBuildkitd(parallelism: number): Promise<s
     core.warning(`Background cache pruning failed: ${error.message}`);
   });
 
-  return buildkitdAddr;
+  return addr;
 }
 
 /**