mirror of
https://github.com/game-ci/unity-builder.git
synced 2026-06-01 14:26:17 -07:00
Compare commits
5 Commits
feature/cl
...
feature/ho
| Author | SHA1 | Date | |
|---|---|---|---|
|
|
3cac1845e3 | ||
|
|
41f00bd1f9 | ||
|
|
7c0c4c2072 | ||
|
|
1bb31f3e98 | ||
|
|
49b37f7831 |
129
action.yml
129
action.yml
@@ -279,120 +279,35 @@ inputs:
|
||||
description:
|
||||
'[Orchestrator] Specifies the repo for the unity builder. Useful if you forked the repo for testing, features, or
|
||||
fixes.'
|
||||
gcpProject:
|
||||
|
||||
hotRunnerEnabled:
|
||||
description: '[HotRunner] Use persistent hot runner for builds (requires pre-registered runners)'
|
||||
required: false
|
||||
default: ''
|
||||
description:
|
||||
'[Orchestrator] [Experimental] Google Cloud project ID for Cloud Run Jobs provider. Falls back to
|
||||
GOOGLE_CLOUD_PROJECT env var.'
|
||||
gcpRegion:
|
||||
default: 'false'
|
||||
hotRunnerTransport:
|
||||
description: '[HotRunner] Transport protocol for hot runner communication: websocket, grpc, named-pipe'
|
||||
required: false
|
||||
default: ''
|
||||
description:
|
||||
'[Orchestrator] [Experimental] Google Cloud region for Cloud Run Jobs (e.g. us-central1). Defaults to the region
|
||||
input if empty.'
|
||||
gcpStorageType:
|
||||
default: 'websocket'
|
||||
hotRunnerHost:
|
||||
description: '[HotRunner] Hot runner host address'
|
||||
required: false
|
||||
default: 'gcs-fuse'
|
||||
description:
|
||||
'[Orchestrator] [Experimental] Storage type for Cloud Run Jobs. Options: gcs-fuse (mount GCS bucket as filesystem,
|
||||
unlimited size, best for large sequential I/O), gcs-copy (copy artifacts in/out via gsutil, simpler, no FUSE
|
||||
overhead), nfs (Filestore NFS mount, true POSIX, good random I/O, up to 100 TiB), in-memory (tmpfs, fastest but
|
||||
volatile, up to 32 GiB).'
|
||||
gcpBucket:
|
||||
default: 'localhost'
|
||||
hotRunnerPort:
|
||||
description: '[HotRunner] Hot runner port number'
|
||||
required: false
|
||||
default: ''
|
||||
description:
|
||||
'[Orchestrator] [Experimental] GCS bucket name for build artifact storage. Used by gcs-fuse and gcs-copy storage
|
||||
types.'
|
||||
gcpFilestoreIp:
|
||||
default: '9090'
|
||||
hotRunnerHealthInterval:
|
||||
description: '[HotRunner] Health check interval in seconds'
|
||||
required: false
|
||||
default: ''
|
||||
description:
|
||||
'[Orchestrator] [Experimental] Filestore instance IP address for NFS storage type. Required when gcpStorageType is
|
||||
nfs.'
|
||||
gcpFilestoreShare:
|
||||
default: '30'
|
||||
hotRunnerMaxIdle:
|
||||
description: '[HotRunner] Maximum idle time in seconds before recycling runner'
|
||||
required: false
|
||||
default: '/share1'
|
||||
description:
|
||||
'[Orchestrator] [Experimental] Filestore share name for NFS storage type. Defaults to /share1 (the Filestore
|
||||
default).'
|
||||
gcpMachineType:
|
||||
default: '3600'
|
||||
hotRunnerFallbackToCold:
|
||||
description: '[HotRunner] Fall back to cold build if no hot runner available'
|
||||
required: false
|
||||
default: 'e2-standard-4'
|
||||
description: '[Orchestrator] [Experimental] Machine type for Cloud Run Jobs (e.g. e2-standard-4, e2-highmem-8).'
|
||||
gcpDiskSizeGb:
|
||||
required: false
|
||||
default: '100'
|
||||
description:
|
||||
'[Orchestrator] [Experimental] Disk size in GB for Cloud Run Jobs in-memory volumes. Only applies to in-memory
|
||||
storage type (max 32).'
|
||||
gcpServiceAccount:
|
||||
required: false
|
||||
default: ''
|
||||
description: '[Orchestrator] [Experimental] Google Cloud service account email for Cloud Run Jobs execution.'
|
||||
gcpVpcConnector:
|
||||
required: false
|
||||
default: ''
|
||||
description: '[Orchestrator] [Experimental] VPC connector name for Cloud Run Jobs private networking.'
|
||||
azureResourceGroup:
|
||||
required: false
|
||||
default: ''
|
||||
description:
|
||||
'[Orchestrator] [Experimental] Azure resource group for Container Instances provider. Falls back to
|
||||
AZURE_RESOURCE_GROUP env var.'
|
||||
azureLocation:
|
||||
required: false
|
||||
default: ''
|
||||
description:
|
||||
'[Orchestrator] [Experimental] Azure region for Container Instances (e.g. eastus, westeurope). Defaults to the
|
||||
region input if empty.'
|
||||
azureStorageType:
|
||||
required: false
|
||||
default: 'azure-files'
|
||||
description:
|
||||
'[Orchestrator] [Experimental] Storage type for Azure Container Instances. Options: azure-files (SMB file share
|
||||
mount, up to 100 TiB, premium throughput), blob-copy (copy artifacts in/out via az storage blob, no mount
|
||||
overhead), azure-files-nfs (NFS 4.1 file share mount, true POSIX, no SMB lock overhead), in-memory (emptyDir
|
||||
tmpfs, fastest but volatile, size limited by container memory).'
|
||||
azureStorageAccount:
|
||||
required: false
|
||||
default: ''
|
||||
description:
|
||||
'[Orchestrator] [Experimental] Azure Storage Account name. Used by azure-files, azure-files-nfs, and blob-copy
|
||||
storage types.'
|
||||
azureFileShareName:
|
||||
required: false
|
||||
default: 'unity-builds'
|
||||
description:
|
||||
'[Orchestrator] [Experimental] Azure File Share name within the storage account. Used by azure-files and
|
||||
azure-files-nfs storage types. Supports up to 100 TiB per share.'
|
||||
azureBlobContainer:
|
||||
required: false
|
||||
default: 'unity-builds'
|
||||
description: '[Orchestrator] [Experimental] Azure Blob container name for blob-copy storage type.'
|
||||
azureSubscriptionId:
|
||||
required: false
|
||||
default: ''
|
||||
description: '[Orchestrator] [Experimental] Azure subscription ID. Falls back to AZURE_SUBSCRIPTION_ID env var.'
|
||||
azureCpu:
|
||||
required: false
|
||||
default: '4'
|
||||
description: '[Orchestrator] [Experimental] CPU cores for Azure Container Instances (1-16).'
|
||||
azureMemoryGb:
|
||||
required: false
|
||||
default: '16'
|
||||
description: '[Orchestrator] [Experimental] Memory in GB for Azure Container Instances (1-16).'
|
||||
azureDiskSizeGb:
|
||||
required: false
|
||||
default: '100'
|
||||
description:
|
||||
'[Orchestrator] [Experimental] File share quota in GB for Azure Container Instances. Premium shares support up to
|
||||
102400 GB (100 TiB).'
|
||||
azureSubnetId:
|
||||
required: false
|
||||
default: ''
|
||||
description: '[Orchestrator] [Experimental] Azure subnet resource ID for VNet-integrated Container Instances.'
|
||||
default: 'true'
|
||||
|
||||
outputs:
|
||||
volume:
|
||||
|
||||
1647
dist/index.js
generated
vendored
1647
dist/index.js
generated
vendored
File diff suppressed because it is too large
Load Diff
2
dist/index.js.map
generated
vendored
2
dist/index.js.map
generated
vendored
File diff suppressed because one or more lines are too long
75
src/index.ts
75
src/index.ts
@@ -3,6 +3,8 @@ import { Action, BuildParameters, Cache, Orchestrator, Docker, ImageTag, Output
|
||||
import { Cli } from './model/cli/cli';
|
||||
import MacBuilder from './model/mac-builder';
|
||||
import PlatformSetup from './model/platform-setup';
|
||||
import { HotRunnerService } from './model/orchestrator/services/hot-runner';
|
||||
import { HotRunnerConfig } from './model/orchestrator/services/hot-runner/hot-runner-types';
|
||||
|
||||
async function runMain() {
|
||||
try {
|
||||
@@ -21,17 +23,46 @@ async function runMain() {
|
||||
|
||||
let exitCode = -1;
|
||||
|
||||
if (buildParameters.providerStrategy === 'local') {
|
||||
// Hot runner path: attempt to use a persistent Unity editor instance
|
||||
if (buildParameters.hotRunnerEnabled) {
|
||||
core.info('[HotRunner] Hot runner mode enabled, attempting hot build...');
|
||||
|
||||
const hotRunnerConfig: HotRunnerConfig = {
|
||||
enabled: true,
|
||||
transport: buildParameters.hotRunnerTransport,
|
||||
host: buildParameters.hotRunnerHost,
|
||||
port: buildParameters.hotRunnerPort,
|
||||
healthCheckInterval: buildParameters.hotRunnerHealthInterval,
|
||||
maxIdleTime: buildParameters.hotRunnerMaxIdle,
|
||||
maxJobsBeforeRecycle: 0, // no automatic recycle by job count
|
||||
};
|
||||
|
||||
const hotRunnerService = new HotRunnerService();
|
||||
|
||||
try {
|
||||
await hotRunnerService.initialize(hotRunnerConfig);
|
||||
const result = await hotRunnerService.submitBuild(buildParameters, (output) => {
|
||||
core.info(output);
|
||||
});
|
||||
|
||||
exitCode = result.exitCode;
|
||||
core.info(`[HotRunner] Build completed with exit code ${exitCode}`);
|
||||
await hotRunnerService.shutdown();
|
||||
} catch (hotRunnerError) {
|
||||
await hotRunnerService.shutdown();
|
||||
|
||||
if (buildParameters.hotRunnerFallbackToCold) {
|
||||
core.warning(
|
||||
`[HotRunner] Hot runner failed: ${(hotRunnerError as Error).message}. Falling back to cold build.`,
|
||||
);
|
||||
exitCode = await runColdBuild(buildParameters, baseImage, workspace, actionFolder);
|
||||
} else {
|
||||
throw hotRunnerError;
|
||||
}
|
||||
}
|
||||
} else if (buildParameters.providerStrategy === 'local') {
|
||||
core.info('Building locally');
|
||||
await PlatformSetup.setup(buildParameters, actionFolder);
|
||||
exitCode =
|
||||
process.platform === 'darwin'
|
||||
? await MacBuilder.run(actionFolder)
|
||||
: await Docker.run(baseImage.toString(), {
|
||||
workspace,
|
||||
actionFolder,
|
||||
...buildParameters,
|
||||
});
|
||||
exitCode = await runColdBuild(buildParameters, baseImage, workspace, actionFolder);
|
||||
} else {
|
||||
await Orchestrator.run(buildParameters, baseImage.toString());
|
||||
exitCode = 0;
|
||||
@@ -50,4 +81,28 @@ async function runMain() {
|
||||
}
|
||||
}
|
||||
|
||||
async function runColdBuild(
|
||||
buildParameters: BuildParameters,
|
||||
baseImage: ImageTag,
|
||||
workspace: string,
|
||||
actionFolder: string,
|
||||
): Promise<number> {
|
||||
if (buildParameters.providerStrategy === 'local') {
|
||||
core.info('Building locally');
|
||||
await PlatformSetup.setup(buildParameters, actionFolder);
|
||||
|
||||
return process.platform === 'darwin'
|
||||
? await MacBuilder.run(actionFolder)
|
||||
: await Docker.run(baseImage.toString(), {
|
||||
workspace,
|
||||
actionFolder,
|
||||
...buildParameters,
|
||||
});
|
||||
} else {
|
||||
await Orchestrator.run(buildParameters, baseImage.toString());
|
||||
|
||||
return 0;
|
||||
}
|
||||
}
|
||||
|
||||
runMain();
|
||||
|
||||
@@ -106,31 +106,13 @@ class BuildParameters {
|
||||
public cacheUnityInstallationOnMac!: boolean;
|
||||
public unityHubVersionOnMac!: string;
|
||||
public dockerWorkspacePath!: string;
|
||||
|
||||
// GCP Cloud Run (Experimental)
|
||||
public gcpProject!: string;
|
||||
public gcpRegion!: string;
|
||||
public gcpStorageType!: string;
|
||||
public gcpBucket!: string;
|
||||
public gcpFilestoreIp!: string;
|
||||
public gcpFilestoreShare!: string;
|
||||
public gcpMachineType!: string;
|
||||
public gcpDiskSizeGb!: string;
|
||||
public gcpServiceAccount!: string;
|
||||
public gcpVpcConnector!: string;
|
||||
|
||||
// Azure Container Instances (Experimental)
|
||||
public azureResourceGroup!: string;
|
||||
public azureLocation!: string;
|
||||
public azureStorageType!: string;
|
||||
public azureStorageAccount!: string;
|
||||
public azureBlobContainer!: string;
|
||||
public azureFileShareName!: string;
|
||||
public azureSubscriptionId!: string;
|
||||
public azureCpu!: string;
|
||||
public azureMemoryGb!: string;
|
||||
public azureDiskSizeGb!: string;
|
||||
public azureSubnetId!: string;
|
||||
public hotRunnerEnabled!: boolean;
|
||||
public hotRunnerTransport!: 'websocket' | 'grpc' | 'named-pipe';
|
||||
public hotRunnerHost!: string;
|
||||
public hotRunnerPort!: number;
|
||||
public hotRunnerHealthInterval!: number;
|
||||
public hotRunnerMaxIdle!: number;
|
||||
public hotRunnerFallbackToCold!: boolean;
|
||||
|
||||
public static shouldUseRetainedWorkspaceMode(buildParameters: BuildParameters) {
|
||||
return buildParameters.maxRetainedWorkspaces > 0 && Orchestrator.lockedWorkspace !== ``;
|
||||
@@ -253,27 +235,6 @@ class BuildParameters {
|
||||
inputPullCommand: OrchestratorOptions.inputPullCommand,
|
||||
pullInputList: OrchestratorOptions.pullInputList,
|
||||
kubeStorageClass: OrchestratorOptions.kubeStorageClass,
|
||||
gcpProject: Input.gcpProject,
|
||||
gcpRegion: Input.gcpRegion,
|
||||
gcpStorageType: Input.gcpStorageType,
|
||||
gcpBucket: Input.gcpBucket,
|
||||
gcpFilestoreIp: Input.gcpFilestoreIp,
|
||||
gcpFilestoreShare: Input.gcpFilestoreShare,
|
||||
gcpMachineType: Input.gcpMachineType,
|
||||
gcpDiskSizeGb: Input.gcpDiskSizeGb,
|
||||
gcpServiceAccount: Input.gcpServiceAccount,
|
||||
gcpVpcConnector: Input.gcpVpcConnector,
|
||||
azureResourceGroup: Input.azureResourceGroup,
|
||||
azureLocation: Input.azureLocation,
|
||||
azureStorageType: Input.azureStorageType,
|
||||
azureStorageAccount: Input.azureStorageAccount,
|
||||
azureBlobContainer: Input.azureBlobContainer,
|
||||
azureFileShareName: Input.azureFileShareName,
|
||||
azureSubscriptionId: Input.azureSubscriptionId,
|
||||
azureCpu: Input.azureCpu,
|
||||
azureMemoryGb: Input.azureMemoryGb,
|
||||
azureDiskSizeGb: Input.azureDiskSizeGb,
|
||||
azureSubnetId: Input.azureSubnetId,
|
||||
cacheKey: OrchestratorOptions.cacheKey,
|
||||
maxRetainedWorkspaces: Number.parseInt(OrchestratorOptions.maxRetainedWorkspaces),
|
||||
useLargePackages: OrchestratorOptions.useLargePackages,
|
||||
@@ -288,6 +249,13 @@ class BuildParameters {
|
||||
cacheUnityInstallationOnMac: Input.cacheUnityInstallationOnMac,
|
||||
unityHubVersionOnMac: Input.unityHubVersionOnMac,
|
||||
dockerWorkspacePath: Input.dockerWorkspacePath,
|
||||
hotRunnerEnabled: Input.hotRunnerEnabled,
|
||||
hotRunnerTransport: Input.hotRunnerTransport,
|
||||
hotRunnerHost: Input.hotRunnerHost,
|
||||
hotRunnerPort: Input.hotRunnerPort,
|
||||
hotRunnerHealthInterval: Input.hotRunnerHealthInterval,
|
||||
hotRunnerMaxIdle: Input.hotRunnerMaxIdle,
|
||||
hotRunnerFallbackToCold: Input.hotRunnerFallbackToCold,
|
||||
};
|
||||
}
|
||||
|
||||
|
||||
@@ -282,90 +282,36 @@ class Input {
|
||||
return Input.getInput('skipActivation')?.toLowerCase() ?? 'false';
|
||||
}
|
||||
|
||||
// GCP Cloud Run (Experimental)
|
||||
static get gcpProject(): string {
|
||||
return Input.getInput('gcpProject') ?? '';
|
||||
static get hotRunnerEnabled(): boolean {
|
||||
const input = Input.getInput('hotRunnerEnabled') ?? false;
|
||||
|
||||
return input === 'true';
|
||||
}
|
||||
|
||||
static get gcpRegion(): string {
|
||||
return Input.getInput('gcpRegion') ?? '';
|
||||
static get hotRunnerTransport(): 'websocket' | 'grpc' | 'named-pipe' {
|
||||
return (Input.getInput('hotRunnerTransport') ?? 'websocket') as 'websocket' | 'grpc' | 'named-pipe';
|
||||
}
|
||||
|
||||
static get gcpStorageType(): string {
|
||||
return Input.getInput('gcpStorageType') ?? 'gcs-fuse';
|
||||
static get hotRunnerHost(): string {
|
||||
return Input.getInput('hotRunnerHost') ?? 'localhost';
|
||||
}
|
||||
|
||||
static get gcpBucket(): string {
|
||||
return Input.getInput('gcpBucket') ?? '';
|
||||
static get hotRunnerPort(): number {
|
||||
return Number.parseInt(Input.getInput('hotRunnerPort') ?? '9090', 10);
|
||||
}
|
||||
|
||||
static get gcpFilestoreIp(): string {
|
||||
return Input.getInput('gcpFilestoreIp') ?? '';
|
||||
static get hotRunnerHealthInterval(): number {
|
||||
return Number.parseInt(Input.getInput('hotRunnerHealthInterval') ?? '30', 10);
|
||||
}
|
||||
|
||||
static get gcpFilestoreShare(): string {
|
||||
return Input.getInput('gcpFilestoreShare') ?? '/share1';
|
||||
static get hotRunnerMaxIdle(): number {
|
||||
return Number.parseInt(Input.getInput('hotRunnerMaxIdle') ?? '3600', 10);
|
||||
}
|
||||
|
||||
static get gcpMachineType(): string {
|
||||
return Input.getInput('gcpMachineType') ?? 'e2-standard-4';
|
||||
}
|
||||
static get hotRunnerFallbackToCold(): boolean {
|
||||
const input = Input.getInput('hotRunnerFallbackToCold') ?? 'true';
|
||||
|
||||
static get gcpDiskSizeGb(): string {
|
||||
return Input.getInput('gcpDiskSizeGb') ?? '100';
|
||||
}
|
||||
|
||||
static get gcpServiceAccount(): string {
|
||||
return Input.getInput('gcpServiceAccount') ?? '';
|
||||
}
|
||||
|
||||
static get gcpVpcConnector(): string {
|
||||
return Input.getInput('gcpVpcConnector') ?? '';
|
||||
}
|
||||
|
||||
// Azure Container Instances (Experimental)
|
||||
static get azureResourceGroup(): string {
|
||||
return Input.getInput('azureResourceGroup') ?? '';
|
||||
}
|
||||
|
||||
static get azureLocation(): string {
|
||||
return Input.getInput('azureLocation') ?? '';
|
||||
}
|
||||
|
||||
static get azureStorageType(): string {
|
||||
return Input.getInput('azureStorageType') ?? 'azure-files';
|
||||
}
|
||||
|
||||
static get azureStorageAccount(): string {
|
||||
return Input.getInput('azureStorageAccount') ?? '';
|
||||
}
|
||||
|
||||
static get azureBlobContainer(): string {
|
||||
return Input.getInput('azureBlobContainer') ?? 'unity-builds';
|
||||
}
|
||||
|
||||
static get azureFileShareName(): string {
|
||||
return Input.getInput('azureFileShareName') ?? 'unity-builds';
|
||||
}
|
||||
|
||||
static get azureSubscriptionId(): string {
|
||||
return Input.getInput('azureSubscriptionId') ?? '';
|
||||
}
|
||||
|
||||
static get azureCpu(): string {
|
||||
return Input.getInput('azureCpu') ?? '4';
|
||||
}
|
||||
|
||||
static get azureMemoryGb(): string {
|
||||
return Input.getInput('azureMemoryGb') ?? '16';
|
||||
}
|
||||
|
||||
static get azureDiskSizeGb(): string {
|
||||
return Input.getInput('azureDiskSizeGb') ?? '100';
|
||||
}
|
||||
|
||||
static get azureSubnetId(): string {
|
||||
return Input.getInput('azureSubnetId') ?? '';
|
||||
return input === 'true';
|
||||
}
|
||||
|
||||
public static ToEnvVarFormat(input: string) {
|
||||
|
||||
@@ -13,8 +13,6 @@ import OrchestratorEnvironmentVariable from './options/orchestrator-environment-
|
||||
import TestOrchestrator from './providers/test';
|
||||
import LocalOrchestrator from './providers/local';
|
||||
import LocalDockerOrchestrator from './providers/docker';
|
||||
import GcpCloudRunProvider from './providers/gcp-cloud-run';
|
||||
import AzureAciProvider from './providers/azure-aci';
|
||||
import loadProvider from './providers/provider-loader';
|
||||
import GitHub from '../github';
|
||||
import SharedWorkspaceLocking from './services/core/shared-workspace-locking';
|
||||
@@ -160,14 +158,6 @@ class Orchestrator {
|
||||
case 'local':
|
||||
Orchestrator.Provider = new LocalOrchestrator();
|
||||
break;
|
||||
case 'gcp-cloud-run':
|
||||
OrchestratorLogger.log('⚠ EXPERIMENTAL: GCP Cloud Run Jobs provider');
|
||||
Orchestrator.Provider = new GcpCloudRunProvider(Orchestrator.buildParameters);
|
||||
break;
|
||||
case 'azure-aci':
|
||||
OrchestratorLogger.log('⚠ EXPERIMENTAL: Azure Container Instances provider');
|
||||
Orchestrator.Provider = new AzureAciProvider(Orchestrator.buildParameters);
|
||||
break;
|
||||
default:
|
||||
// Try to load provider using the dynamic loader for unknown providers
|
||||
try {
|
||||
|
||||
@@ -1,536 +0,0 @@
|
||||
/**
|
||||
* Azure Container Instances (ACI) Provider (Experimental)
|
||||
*
|
||||
* Executes Unity builds as Azure Container Instances with configurable storage backends.
|
||||
*
|
||||
* Storage types:
|
||||
* - azure-files: SMB file share mount via Azure Files. Up to 100 TiB per share,
|
||||
* premium throughput. Default.
|
||||
* Requires: azureStorageAccount, azureFileShareName
|
||||
* - blob-copy: Copy artifacts in/out of Azure Blob Storage before/after the build.
|
||||
* No mount overhead, simpler.
|
||||
* Requires: azureStorageAccount, azureBlobContainer
|
||||
* - azure-files-nfs: NFS 4.1 file share mount. True POSIX semantics, no SMB lock overhead,
|
||||
* better for Unity Library caching (many small random reads).
|
||||
* Requires: azureStorageAccount, azureFileShareName, Premium FileStorage,
|
||||
* VNet integration (azureSubnetId)
|
||||
* - in-memory: emptyDir volume (tmpfs). Fastest I/O but volatile, size limited by
|
||||
* container memory allocation.
|
||||
*
|
||||
* Prerequisites:
|
||||
* - Azure CLI authenticated (az login or service principal)
|
||||
* - A resource group for build resources
|
||||
* - Contributor role on the resource group
|
||||
*
|
||||
* @experimental This provider is experimental. APIs and behavior may change.
|
||||
*/
|
||||
|
||||
import { ProviderInterface } from '../provider-interface';
|
||||
import BuildParameters from '../../../build-parameters';
|
||||
import OrchestratorLogger from '../../services/core/orchestrator-logger';
|
||||
import OrchestratorEnvironmentVariable from '../../options/orchestrator-environment-variable';
|
||||
import OrchestratorSecret from '../../options/orchestrator-secret';
|
||||
import { ProviderResource } from '../provider-resource';
|
||||
import { ProviderWorkflow } from '../provider-workflow';
|
||||
import { OrchestratorSystem } from '../../services/core/orchestrator-system';
|
||||
import { Input } from '../../..';
|
||||
import ResourceTracking from '../../services/core/resource-tracking';
|
||||
|
||||
type AzureStorageType = 'azure-files' | 'blob-copy' | 'azure-files-nfs' | 'in-memory';
|
||||
|
||||
class AzureAciProvider implements ProviderInterface {
|
||||
private readonly resourceGroup: string;
|
||||
private readonly location: string;
|
||||
private readonly storageType: AzureStorageType;
|
||||
private readonly storageAccount: string;
|
||||
private readonly blobContainer: string;
|
||||
private readonly fileShareName: string;
|
||||
private readonly subscriptionId: string;
|
||||
private readonly cpu: number;
|
||||
private readonly memoryGb: number;
|
||||
private readonly diskSizeGb: number;
|
||||
private readonly subnetId: string;
|
||||
private buildParameters: BuildParameters;
|
||||
|
||||
constructor(buildParameters: BuildParameters) {
|
||||
this.buildParameters = buildParameters;
|
||||
this.resourceGroup = buildParameters.azureResourceGroup || process.env.AZURE_RESOURCE_GROUP || '';
|
||||
this.location = buildParameters.azureLocation || Input.region || 'eastus';
|
||||
this.storageType = (buildParameters.azureStorageType || 'azure-files') as AzureStorageType;
|
||||
this.storageAccount = buildParameters.azureStorageAccount || process.env.AZURE_STORAGE_ACCOUNT || '';
|
||||
this.blobContainer = buildParameters.azureBlobContainer || 'unity-builds';
|
||||
this.fileShareName = buildParameters.azureFileShareName || 'unity-builds';
|
||||
this.subscriptionId = buildParameters.azureSubscriptionId || process.env.AZURE_SUBSCRIPTION_ID || '';
|
||||
this.cpu = Number.parseInt(buildParameters.azureCpu || '4', 10);
|
||||
this.memoryGb = Number.parseInt(buildParameters.azureMemoryGb || '16', 10);
|
||||
this.diskSizeGb = Number.parseInt(buildParameters.azureDiskSizeGb || '100', 10);
|
||||
this.subnetId = buildParameters.azureSubnetId || '';
|
||||
|
||||
OrchestratorLogger.log('[Azure ACI] Provider initialized (EXPERIMENTAL)');
|
||||
OrchestratorLogger.log(`[Azure ACI] Resource Group: ${this.resourceGroup || '(not set)'}`);
|
||||
OrchestratorLogger.log(`[Azure ACI] Location: ${this.location}`);
|
||||
OrchestratorLogger.log(`[Azure ACI] Storage: ${this.storageType}`);
|
||||
OrchestratorLogger.log(`[Azure ACI] Resources: ${this.cpu} CPU, ${this.memoryGb}GB RAM`);
|
||||
|
||||
this.validateStorageConfig();
|
||||
}
|
||||
|
||||
private validateStorageConfig(): void {
|
||||
switch (this.storageType) {
|
||||
case 'azure-files':
|
||||
if (!this.storageAccount) {
|
||||
OrchestratorLogger.logWarning(
|
||||
'[Azure ACI] Storage type "azure-files" requires azureStorageAccount to be set.',
|
||||
);
|
||||
} else {
|
||||
OrchestratorLogger.log(`[Azure ACI] File Share: ${this.storageAccount}/${this.fileShareName} (SMB)`);
|
||||
}
|
||||
break;
|
||||
case 'azure-files-nfs':
|
||||
if (!this.storageAccount) {
|
||||
OrchestratorLogger.logWarning(
|
||||
'[Azure ACI] Storage type "azure-files-nfs" requires azureStorageAccount (Premium FileStorage).',
|
||||
);
|
||||
}
|
||||
if (!this.subnetId) {
|
||||
OrchestratorLogger.logWarning('[Azure ACI] NFS file shares require VNet integration. Set azureSubnetId.');
|
||||
} else {
|
||||
OrchestratorLogger.log(`[Azure ACI] File Share: ${this.storageAccount}/${this.fileShareName} (NFS 4.1)`);
|
||||
}
|
||||
break;
|
||||
case 'blob-copy':
|
||||
if (!this.storageAccount) {
|
||||
OrchestratorLogger.logWarning('[Azure ACI] Storage type "blob-copy" requires azureStorageAccount to be set.');
|
||||
} else {
|
||||
OrchestratorLogger.log(`[Azure ACI] Blob container: ${this.storageAccount}/${this.blobContainer}`);
|
||||
}
|
||||
break;
|
||||
case 'in-memory':
|
||||
OrchestratorLogger.log(
|
||||
`[Azure ACI] In-memory volume (emptyDir): limited by ${this.memoryGb}GB container memory`,
|
||||
);
|
||||
break;
|
||||
default:
|
||||
OrchestratorLogger.logWarning(
|
||||
`[Azure ACI] Unknown storage type '${this.storageType}'. Valid: azure-files, blob-copy, azure-files-nfs, in-memory`,
|
||||
);
|
||||
}
|
||||
|
||||
if (!this.resourceGroup) {
|
||||
OrchestratorLogger.logWarning(
|
||||
'[Azure ACI] No resource group specified. Set azureResourceGroup input or AZURE_RESOURCE_GROUP env var.',
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
async setupWorkflow(
|
||||
buildGuid: string,
|
||||
buildParameters: BuildParameters,
|
||||
branchName: string,
|
||||
defaultSecretsArray: { ParameterKey: string; EnvironmentVariable: string; ParameterValue: string }[],
|
||||
) {
|
||||
OrchestratorLogger.log(`[Azure ACI] Setting up workflow for build ${buildGuid}`);
|
||||
ResourceTracking.logAllocationSummary('azure-aci setup');
|
||||
|
||||
// Verify Azure CLI is available
|
||||
try {
|
||||
await OrchestratorSystem.Run('az version --output json', false, true);
|
||||
OrchestratorLogger.log('[Azure ACI] Azure CLI detected');
|
||||
} catch {
|
||||
throw new Error(
|
||||
'[Azure ACI] Azure CLI not found. Install Azure CLI: https://learn.microsoft.com/en-us/cli/azure/install-azure-cli',
|
||||
);
|
||||
}
|
||||
|
||||
if (this.subscriptionId) {
|
||||
await OrchestratorSystem.Run(`az account set --subscription="${this.subscriptionId}"`);
|
||||
}
|
||||
|
||||
// Ensure resource group exists
|
||||
if (this.resourceGroup) {
|
||||
try {
|
||||
await OrchestratorSystem.Run(`az group show --name "${this.resourceGroup}" --output json`, false, true);
|
||||
OrchestratorLogger.log(`[Azure ACI] Resource group ${this.resourceGroup} exists`);
|
||||
} catch {
|
||||
OrchestratorLogger.log(`[Azure ACI] Creating resource group ${this.resourceGroup}`);
|
||||
await OrchestratorSystem.Run(`az group create --name "${this.resourceGroup}" --location "${this.location}"`);
|
||||
}
|
||||
}
|
||||
|
||||
// Storage-specific setup
|
||||
switch (this.storageType) {
|
||||
case 'azure-files':
|
||||
await this.setupStorageAccount('Standard_LRS', 'StorageV2');
|
||||
await this.setupFileShare();
|
||||
break;
|
||||
case 'azure-files-nfs':
|
||||
await this.setupStorageAccount('Premium_LRS', 'FileStorage');
|
||||
await this.setupNfsFileShare();
|
||||
break;
|
||||
case 'blob-copy':
|
||||
await this.setupStorageAccount('Standard_LRS', 'StorageV2');
|
||||
await this.setupBlobContainer();
|
||||
break;
|
||||
case 'in-memory':
|
||||
// No storage setup needed
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
private async setupStorageAccount(sku: string, kind: string): Promise<void> {
|
||||
if (!this.storageAccount || !this.resourceGroup) return;
|
||||
|
||||
try {
|
||||
await OrchestratorSystem.Run(
|
||||
`az storage account show --name "${this.storageAccount}" --resource-group "${this.resourceGroup}" --output json`,
|
||||
false,
|
||||
true,
|
||||
);
|
||||
OrchestratorLogger.log(`[Azure ACI] Storage account ${this.storageAccount} exists`);
|
||||
} catch {
|
||||
OrchestratorLogger.log(`[Azure ACI] Creating storage account ${this.storageAccount} (${sku}, ${kind})`);
|
||||
await OrchestratorSystem.Run(
|
||||
`az storage account create --name "${this.storageAccount}" --resource-group "${this.resourceGroup}" --location "${this.location}" --sku ${sku} --kind ${kind}`,
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
private async setupFileShare(): Promise<void> {
|
||||
if (!this.storageAccount || !this.resourceGroup) return;
|
||||
try {
|
||||
await OrchestratorSystem.Run(
|
||||
`az storage share-rm show --storage-account "${this.storageAccount}" --name "${this.fileShareName}" --resource-group "${this.resourceGroup}" --output json`,
|
||||
false,
|
||||
true,
|
||||
);
|
||||
} catch {
|
||||
OrchestratorLogger.log(`[Azure ACI] Creating file share ${this.fileShareName} (${this.diskSizeGb}GB)`);
|
||||
await OrchestratorSystem.Run(
|
||||
`az storage share-rm create --storage-account "${this.storageAccount}" --name "${this.fileShareName}" --resource-group "${this.resourceGroup}" --quota ${this.diskSizeGb}`,
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
private async setupNfsFileShare(): Promise<void> {
|
||||
if (!this.storageAccount || !this.resourceGroup) return;
|
||||
try {
|
||||
await OrchestratorSystem.Run(
|
||||
`az storage share-rm show --storage-account "${this.storageAccount}" --name "${this.fileShareName}" --resource-group "${this.resourceGroup}" --output json`,
|
||||
false,
|
||||
true,
|
||||
);
|
||||
} catch {
|
||||
OrchestratorLogger.log(`[Azure ACI] Creating NFS file share ${this.fileShareName} (${this.diskSizeGb}GB)`);
|
||||
await OrchestratorSystem.Run(
|
||||
`az storage share-rm create --storage-account "${this.storageAccount}" --name "${this.fileShareName}" --resource-group "${this.resourceGroup}" --quota ${this.diskSizeGb} --enabled-protocols NFS`,
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
private async setupBlobContainer(): Promise<void> {
|
||||
if (!this.storageAccount || !this.resourceGroup) return;
|
||||
try {
|
||||
await OrchestratorSystem.Run(
|
||||
`az storage container show --name "${this.blobContainer}" --account-name "${this.storageAccount}" --output json`,
|
||||
false,
|
||||
true,
|
||||
);
|
||||
} catch {
|
||||
OrchestratorLogger.log(`[Azure ACI] Creating blob container ${this.blobContainer}`);
|
||||
await OrchestratorSystem.Run(
|
||||
`az storage container create --name "${this.blobContainer}" --account-name "${this.storageAccount}"`,
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
private async getStorageKey(): Promise<string> {
|
||||
if (!this.storageAccount || !this.resourceGroup) return '';
|
||||
try {
|
||||
const keyJson = await OrchestratorSystem.Run(
|
||||
`az storage account keys list --account-name "${this.storageAccount}" --resource-group "${this.resourceGroup}" --output json`,
|
||||
false,
|
||||
true,
|
||||
);
|
||||
const keys = JSON.parse(keyJson);
|
||||
return keys[0]?.value || '';
|
||||
} catch (error: any) {
|
||||
OrchestratorLogger.logWarning(`[Azure ACI] Could not get storage key: ${error.message}`);
|
||||
return '';
|
||||
}
|
||||
}
|
||||
|
||||
private async buildVolumeFlags(mountdir: string): Promise<string> {
|
||||
switch (this.storageType) {
|
||||
case 'azure-files': {
|
||||
const storageKey = await this.getStorageKey();
|
||||
if (!storageKey) return '';
|
||||
return [
|
||||
`--azure-file-volume-account-name "${this.storageAccount}"`,
|
||||
`--azure-file-volume-account-key "${storageKey}"`,
|
||||
`--azure-file-volume-share-name "${this.fileShareName}"`,
|
||||
`--azure-file-volume-mount-path "${mountdir}"`,
|
||||
].join(' ');
|
||||
}
|
||||
|
||||
case 'azure-files-nfs': {
|
||||
// ACI NFS mount uses a YAML deployment template; for CLI we use the same
|
||||
// azure-file-volume flags but the share must be NFS-enabled and
|
||||
// the container must be in a VNet
|
||||
const storageKey = await this.getStorageKey();
|
||||
if (!storageKey) return '';
|
||||
return [
|
||||
`--azure-file-volume-account-name "${this.storageAccount}"`,
|
||||
`--azure-file-volume-account-key "${storageKey}"`,
|
||||
`--azure-file-volume-share-name "${this.fileShareName}"`,
|
||||
`--azure-file-volume-mount-path "${mountdir}"`,
|
||||
].join(' ');
|
||||
}
|
||||
|
||||
case 'in-memory':
|
||||
// ACI emptyDir volumes require YAML deployment; for simplicity we skip
|
||||
// the volume mount and let the container use its own filesystem
|
||||
OrchestratorLogger.log('[Azure ACI] In-memory mode: using container filesystem (no persistent mount)');
|
||||
return '';
|
||||
|
||||
case 'blob-copy':
|
||||
// No volume mount — artifacts are copied in/out via az storage blob commands
|
||||
return '';
|
||||
|
||||
default:
|
||||
return '';
|
||||
}
|
||||
}
|
||||
|
||||
async runTaskInWorkflow(
|
||||
buildGuid: string,
|
||||
image: string,
|
||||
commands: string,
|
||||
mountdir: string,
|
||||
workingdir: string,
|
||||
environment: OrchestratorEnvironmentVariable[],
|
||||
secrets: OrchestratorSecret[],
|
||||
): Promise<string> {
|
||||
OrchestratorLogger.log(`[Azure ACI] Running task for build ${buildGuid}`);
|
||||
ResourceTracking.logAllocationSummary('azure-aci task');
|
||||
|
||||
const containerName = `unity-build-${buildGuid}`
|
||||
.toLowerCase()
|
||||
.replace(/[^a-z0-9-]/g, '-')
|
||||
.slice(0, 63);
|
||||
|
||||
// Build environment variable flags
|
||||
const allEnvVars = [
|
||||
...environment.map((env) => `${env.name}=${env.value}`),
|
||||
...secrets.map((s) => `${s.EnvironmentVariable}=${s.ParameterValue}`),
|
||||
];
|
||||
const envFlag = allEnvVars.length > 0 ? `--environment-variables ${allEnvVars.map((e) => `"${e}"`).join(' ')}` : '';
|
||||
|
||||
// Build volume flags based on storage type
|
||||
const volumeFlags = await this.buildVolumeFlags(mountdir);
|
||||
|
||||
const subnetFlag = this.subnetId ? `--subnet "${this.subnetId}"` : '';
|
||||
|
||||
// For blob-copy, wrap the user command with copy-in/copy-out steps
|
||||
let effectiveCommands = commands;
|
||||
if (this.storageType === 'blob-copy' && this.storageAccount && commands) {
|
||||
effectiveCommands = [
|
||||
`az storage blob download-batch --destination "${mountdir}" --source "${this.blobContainer}" --account-name "${this.storageAccount}" 2>/dev/null || true`,
|
||||
commands,
|
||||
`az storage blob upload-batch --source "${mountdir}" --destination "${this.blobContainer}" --account-name "${this.storageAccount}" --overwrite`,
|
||||
].join(' && ');
|
||||
}
|
||||
|
||||
const commandFlag = effectiveCommands
|
||||
? `--command-line "/bin/sh -c '${effectiveCommands.replace(/'/g, "'\\''")}'"`
|
||||
: '';
|
||||
|
||||
const createCmd = [
|
||||
'az container create',
|
||||
`--resource-group "${this.resourceGroup}"`,
|
||||
`--name "${containerName}"`,
|
||||
`--image "${image}"`,
|
||||
`--location "${this.location}"`,
|
||||
`--cpu ${this.cpu}`,
|
||||
`--memory ${this.memoryGb}`,
|
||||
'--restart-policy Never',
|
||||
'--os-type Linux',
|
||||
volumeFlags,
|
||||
envFlag,
|
||||
subnetFlag,
|
||||
commandFlag,
|
||||
'--output json',
|
||||
]
|
||||
.filter(Boolean)
|
||||
.join(' ');
|
||||
|
||||
try {
|
||||
await OrchestratorSystem.Run(createCmd);
|
||||
OrchestratorLogger.log(
|
||||
`[Azure ACI] Container ${containerName} created (storage: ${this.storageType}), waiting for completion...`,
|
||||
);
|
||||
} catch (error: any) {
|
||||
throw new Error(`[Azure ACI] Failed to create container: ${error.message}`);
|
||||
}
|
||||
|
||||
const output = await this.waitForContainerCompletion(containerName);
|
||||
return output;
|
||||
}
|
||||
|
||||
private async waitForContainerCompletion(containerName: string): Promise<string> {
|
||||
const maxWaitMs = 24 * 60 * 60 * 1000;
|
||||
const pollIntervalMs = 15_000;
|
||||
const startTime = Date.now();
|
||||
let lastLogLength = 0;
|
||||
|
||||
while (Date.now() - startTime < maxWaitMs) {
|
||||
try {
|
||||
const stateJson = await OrchestratorSystem.Run(
|
||||
`az container show --resource-group "${this.resourceGroup}" --name "${containerName}" --output json`,
|
||||
false,
|
||||
true,
|
||||
);
|
||||
|
||||
const state = JSON.parse(stateJson);
|
||||
const containerState =
|
||||
state.containers?.[0]?.instanceView?.currentState?.state || state.instanceView?.state || 'Unknown';
|
||||
const provisioningState = state.provisioningState || 'Unknown';
|
||||
|
||||
// Stream logs incrementally
|
||||
try {
|
||||
const logs = await OrchestratorSystem.Run(
|
||||
`az container logs --resource-group "${this.resourceGroup}" --name "${containerName}"`,
|
||||
false,
|
||||
true,
|
||||
);
|
||||
if (logs && logs.length > lastLogLength) {
|
||||
const newLogs = logs.slice(lastLogLength);
|
||||
for (const line of newLogs.split('\n')) {
|
||||
if (line.trim()) {
|
||||
OrchestratorLogger.log(`[Build] ${line}`);
|
||||
}
|
||||
}
|
||||
lastLogLength = logs.length;
|
||||
}
|
||||
} catch {
|
||||
// Logs may not be available yet
|
||||
}
|
||||
|
||||
if (containerState === 'Terminated' || provisioningState === 'Succeeded') {
|
||||
const exitCode = state.containers?.[0]?.instanceView?.currentState?.exitCode;
|
||||
if (exitCode !== undefined && exitCode !== 0) {
|
||||
throw new Error(`[Azure ACI] Container exited with code ${exitCode}`);
|
||||
}
|
||||
OrchestratorLogger.log('[Azure ACI] Container completed successfully');
|
||||
try {
|
||||
return await OrchestratorSystem.Run(
|
||||
`az container logs --resource-group "${this.resourceGroup}" --name "${containerName}"`,
|
||||
false,
|
||||
true,
|
||||
);
|
||||
} catch {
|
||||
return '';
|
||||
}
|
||||
}
|
||||
|
||||
if (provisioningState === 'Failed') {
|
||||
const detail =
|
||||
state.containers?.[0]?.instanceView?.currentState?.detailStatus ||
|
||||
state.containers?.[0]?.instanceView?.events?.map((e: any) => e.message).join('; ') ||
|
||||
'Unknown error';
|
||||
throw new Error(`[Azure ACI] Container provisioning failed: ${detail}`);
|
||||
}
|
||||
} catch (error: any) {
|
||||
if (error.message?.includes('Container provisioning failed') || error.message?.includes('exited with code')) {
|
||||
throw error;
|
||||
}
|
||||
OrchestratorLogger.logWarning(`[Azure ACI] Polling error: ${error.message}`);
|
||||
}
|
||||
|
||||
await new Promise((resolve) => setTimeout(resolve, pollIntervalMs));
|
||||
}
|
||||
|
||||
throw new Error('[Azure ACI] Container execution timed out after 24 hours');
|
||||
}
|
||||
|
||||
async cleanupWorkflow(
|
||||
buildParameters: BuildParameters,
|
||||
branchName: string,
|
||||
defaultSecretsArray: { ParameterKey: string; EnvironmentVariable: string; ParameterValue: string }[],
|
||||
) {
|
||||
OrchestratorLogger.log('[Azure ACI] Cleaning up workflow');
|
||||
}
|
||||
|
||||
async garbageCollect(
|
||||
filter: string,
|
||||
previewOnly: boolean,
|
||||
olderThan: Number,
|
||||
fullCache: boolean,
|
||||
baseDependencies: boolean,
|
||||
): Promise<string> {
|
||||
OrchestratorLogger.log('[Azure ACI] Garbage collecting old container groups');
|
||||
|
||||
try {
|
||||
const containersJson = await OrchestratorSystem.Run(
|
||||
`az container list --resource-group "${this.resourceGroup}" --output json`,
|
||||
false,
|
||||
true,
|
||||
);
|
||||
|
||||
const containers = JSON.parse(containersJson || '[]');
|
||||
const cutoffDate = new Date();
|
||||
cutoffDate.setDate(cutoffDate.getDate() - Number(olderThan));
|
||||
|
||||
let deletedCount = 0;
|
||||
for (const container of containers) {
|
||||
const name = container.name || '';
|
||||
if (!name.startsWith('unity-build-')) continue;
|
||||
|
||||
const createdAt = new Date(container.tags?.createdAt || container.properties?.provisioningState || 0);
|
||||
const state = container.containers?.[0]?.instanceView?.currentState?.state || '';
|
||||
|
||||
if (state === 'Terminated' || createdAt < cutoffDate) {
|
||||
if (previewOnly) {
|
||||
OrchestratorLogger.log(`[Azure ACI] Would delete: ${name}`);
|
||||
} else {
|
||||
await OrchestratorSystem.Run(
|
||||
`az container delete --resource-group "${this.resourceGroup}" --name "${name}" --yes`,
|
||||
);
|
||||
deletedCount++;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return `Garbage collected ${deletedCount} Azure container instances`;
|
||||
} catch (error: any) {
|
||||
OrchestratorLogger.logWarning(`[Azure ACI] Garbage collection failed: ${error.message}`);
|
||||
return '';
|
||||
}
|
||||
}
|
||||
|
||||
async listResources(): Promise<ProviderResource[]> {
|
||||
try {
|
||||
const containersJson = await OrchestratorSystem.Run(
|
||||
`az container list --resource-group "${this.resourceGroup}" --output json`,
|
||||
false,
|
||||
true,
|
||||
);
|
||||
|
||||
const containers = JSON.parse(containersJson || '[]');
|
||||
return containers
|
||||
.filter((c: any) => (c.name || '').startsWith('unity-build-'))
|
||||
.map((c: any) => ({ Name: c.name || '' }));
|
||||
} catch {
|
||||
return [];
|
||||
}
|
||||
}
|
||||
|
||||
listWorkflow(): Promise<ProviderWorkflow[]> {
|
||||
throw new Error('[Azure ACI] listWorkflow not implemented for this experimental provider');
|
||||
}
|
||||
|
||||
async watchWorkflow(): Promise<string> {
|
||||
throw new Error('[Azure ACI] watchWorkflow not implemented for this experimental provider');
|
||||
}
|
||||
}
|
||||
|
||||
export default AzureAciProvider;
|
||||
@@ -1,435 +0,0 @@
|
||||
/**
|
||||
* Google Cloud Run Jobs Provider (Experimental)
|
||||
*
|
||||
* Executes Unity builds as Cloud Run Jobs with configurable storage backends.
|
||||
*
|
||||
* Storage types:
|
||||
* - gcs-fuse: Mount a GCS bucket as a POSIX filesystem via GCS FUSE sidecar.
|
||||
* Unlimited size, best for large sequential reads/writes.
|
||||
* Requires: gcpBucket
|
||||
* - gcs-copy: Copy artifacts in/out of GCS before/after the build via gsutil.
|
||||
* No mount overhead, simpler, works everywhere.
|
||||
* Requires: gcpBucket
|
||||
* - nfs: Mount a Filestore NFS share. True POSIX semantics, good random I/O,
|
||||
* up to 100 TiB. Best for Library caching (many small random reads).
|
||||
* Requires: gcpFilestoreIp, gcpFilestoreShare
|
||||
* - in-memory: tmpfs volume (emptyDir). Fastest I/O but volatile and limited to 32 GiB.
|
||||
* Good for scratch/temp space during builds.
|
||||
*
|
||||
* Prerequisites:
|
||||
* - Google Cloud SDK authenticated (GOOGLE_APPLICATION_CREDENTIALS or gcloud auth)
|
||||
* - Cloud Run Jobs API enabled
|
||||
* - Service account with roles: Cloud Run Admin, Storage Admin, Logs Viewer
|
||||
*
|
||||
* @experimental This provider is experimental. APIs and behavior may change.
|
||||
*/
|
||||
|
||||
import { ProviderInterface } from '../provider-interface';
|
||||
import BuildParameters from '../../../build-parameters';
|
||||
import OrchestratorLogger from '../../services/core/orchestrator-logger';
|
||||
import OrchestratorEnvironmentVariable from '../../options/orchestrator-environment-variable';
|
||||
import OrchestratorSecret from '../../options/orchestrator-secret';
|
||||
import { ProviderResource } from '../provider-resource';
|
||||
import { ProviderWorkflow } from '../provider-workflow';
|
||||
import { OrchestratorSystem } from '../../services/core/orchestrator-system';
|
||||
import { Input } from '../../..';
|
||||
import ResourceTracking from '../../services/core/resource-tracking';
|
||||
|
||||
type GcpStorageType = 'gcs-fuse' | 'gcs-copy' | 'nfs' | 'in-memory';
|
||||
|
||||
class GcpCloudRunProvider implements ProviderInterface {
|
||||
private readonly project: string;
|
||||
private readonly region: string;
|
||||
private readonly storageType: GcpStorageType;
|
||||
private readonly bucket: string;
|
||||
private readonly filestoreIp: string;
|
||||
private readonly filestoreShare: string;
|
||||
private readonly machineType: string;
|
||||
private readonly diskSizeGb: number;
|
||||
private readonly serviceAccount: string;
|
||||
private readonly vpcConnector: string;
|
||||
private buildParameters: BuildParameters;
|
||||
|
||||
constructor(buildParameters: BuildParameters) {
|
||||
this.buildParameters = buildParameters;
|
||||
this.project = buildParameters.gcpProject || process.env.GOOGLE_CLOUD_PROJECT || process.env.GCLOUD_PROJECT || '';
|
||||
this.region = buildParameters.gcpRegion || Input.region || 'us-central1';
|
||||
this.storageType = (buildParameters.gcpStorageType || 'gcs-fuse') as GcpStorageType;
|
||||
this.bucket = buildParameters.gcpBucket || '';
|
||||
this.filestoreIp = buildParameters.gcpFilestoreIp || '';
|
||||
this.filestoreShare = buildParameters.gcpFilestoreShare || '/share1';
|
||||
this.machineType = buildParameters.gcpMachineType || 'e2-standard-4';
|
||||
this.diskSizeGb = Number.parseInt(buildParameters.gcpDiskSizeGb || '100', 10);
|
||||
this.serviceAccount = buildParameters.gcpServiceAccount || '';
|
||||
this.vpcConnector = buildParameters.gcpVpcConnector || '';
|
||||
|
||||
OrchestratorLogger.log('[GCP Cloud Run] Provider initialized (EXPERIMENTAL)');
|
||||
OrchestratorLogger.log(`[GCP Cloud Run] Project: ${this.project || '(auto-detect)'}`);
|
||||
OrchestratorLogger.log(`[GCP Cloud Run] Region: ${this.region}`);
|
||||
OrchestratorLogger.log(`[GCP Cloud Run] Storage: ${this.storageType}`);
|
||||
|
||||
this.validateStorageConfig();
|
||||
}
|
||||
|
||||
private validateStorageConfig(): void {
|
||||
switch (this.storageType) {
|
||||
case 'gcs-fuse':
|
||||
case 'gcs-copy':
|
||||
if (!this.bucket) {
|
||||
OrchestratorLogger.logWarning(
|
||||
`[GCP Cloud Run] Storage type '${this.storageType}' requires gcpBucket to be set.`,
|
||||
);
|
||||
} else {
|
||||
OrchestratorLogger.log(`[GCP Cloud Run] Bucket: gs://${this.bucket}`);
|
||||
}
|
||||
break;
|
||||
case 'nfs':
|
||||
if (!this.filestoreIp) {
|
||||
OrchestratorLogger.logWarning('[GCP Cloud Run] Storage type "nfs" requires gcpFilestoreIp to be set.');
|
||||
} else {
|
||||
OrchestratorLogger.log(`[GCP Cloud Run] Filestore: ${this.filestoreIp}:${this.filestoreShare}`);
|
||||
}
|
||||
if (!this.vpcConnector) {
|
||||
OrchestratorLogger.logWarning(
|
||||
'[GCP Cloud Run] NFS storage usually requires gcpVpcConnector for private network access to Filestore.',
|
||||
);
|
||||
}
|
||||
break;
|
||||
case 'in-memory':
|
||||
OrchestratorLogger.log(`[GCP Cloud Run] In-memory volume: ${Math.min(this.diskSizeGb, 32)} GiB (max 32)`);
|
||||
break;
|
||||
default:
|
||||
OrchestratorLogger.logWarning(
|
||||
`[GCP Cloud Run] Unknown storage type '${this.storageType}'. Valid: gcs-fuse, gcs-copy, nfs, in-memory`,
|
||||
);
|
||||
}
|
||||
|
||||
if (!this.project) {
|
||||
OrchestratorLogger.logWarning(
|
||||
'[GCP Cloud Run] No project specified. Set gcpProject input or GOOGLE_CLOUD_PROJECT env var.',
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
async setupWorkflow(
|
||||
buildGuid: string,
|
||||
buildParameters: BuildParameters,
|
||||
branchName: string,
|
||||
defaultSecretsArray: { ParameterKey: string; EnvironmentVariable: string; ParameterValue: string }[],
|
||||
) {
|
||||
OrchestratorLogger.log(`[GCP Cloud Run] Setting up workflow for build ${buildGuid}`);
|
||||
ResourceTracking.logAllocationSummary('gcp-cloud-run setup');
|
||||
|
||||
// Verify gcloud CLI is available
|
||||
try {
|
||||
await OrchestratorSystem.Run('gcloud --version', false, true);
|
||||
OrchestratorLogger.log('[GCP Cloud Run] gcloud CLI detected');
|
||||
} catch {
|
||||
throw new Error(
|
||||
'[GCP Cloud Run] gcloud CLI not found. Install Google Cloud SDK: https://cloud.google.com/sdk/docs/install',
|
||||
);
|
||||
}
|
||||
|
||||
// Verify Cloud Run Jobs API is enabled
|
||||
try {
|
||||
const projectFlag = this.project ? `--project=${this.project}` : '';
|
||||
await OrchestratorSystem.Run(
|
||||
`gcloud services list --enabled --filter="name:run.googleapis.com" ${projectFlag} --format="value(name)"`,
|
||||
false,
|
||||
true,
|
||||
);
|
||||
} catch {
|
||||
OrchestratorLogger.logWarning(
|
||||
'[GCP Cloud Run] Could not verify Cloud Run API status. Ensure run.googleapis.com is enabled.',
|
||||
);
|
||||
}
|
||||
|
||||
// Storage-specific setup
|
||||
if ((this.storageType === 'gcs-fuse' || this.storageType === 'gcs-copy') && this.bucket) {
|
||||
await this.ensureBucketExists();
|
||||
}
|
||||
}
|
||||
|
||||
private async ensureBucketExists(): Promise<void> {
|
||||
try {
|
||||
await OrchestratorSystem.Run(
|
||||
`gcloud storage buckets describe gs://${this.bucket} --format="value(name)"`,
|
||||
false,
|
||||
true,
|
||||
);
|
||||
OrchestratorLogger.log(`[GCP Cloud Run] Bucket gs://${this.bucket} exists`);
|
||||
} catch {
|
||||
OrchestratorLogger.log(`[GCP Cloud Run] Creating bucket gs://${this.bucket}`);
|
||||
const projectFlag = this.project ? `--project=${this.project}` : '';
|
||||
await OrchestratorSystem.Run(
|
||||
`gcloud storage buckets create gs://${this.bucket} --location=${this.region} ${projectFlag}`,
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
private buildVolumeFlags(mountdir: string): { volumeFlags: string; mountFlags: string } {
|
||||
switch (this.storageType) {
|
||||
case 'gcs-fuse':
|
||||
if (!this.bucket) return { volumeFlags: '', mountFlags: '' };
|
||||
return {
|
||||
volumeFlags: `--add-volume=name=gcs-fuse,type=cloud-storage,bucket=${this.bucket}`,
|
||||
mountFlags: `--add-volume-mount=volume=gcs-fuse,mount-path=${mountdir}`,
|
||||
};
|
||||
|
||||
case 'nfs':
|
||||
if (!this.filestoreIp) return { volumeFlags: '', mountFlags: '' };
|
||||
return {
|
||||
volumeFlags: `--add-volume=name=nfs-vol,type=nfs,location=${this.filestoreIp}:${this.filestoreShare}`,
|
||||
mountFlags: `--add-volume-mount=volume=nfs-vol,mount-path=${mountdir}`,
|
||||
};
|
||||
|
||||
case 'in-memory': {
|
||||
const sizeGib = Math.min(this.diskSizeGb, 32);
|
||||
return {
|
||||
volumeFlags: `--add-volume=name=tmpfs-vol,type=in-memory,size-limit=${sizeGib}Gi`,
|
||||
mountFlags: `--add-volume-mount=volume=tmpfs-vol,mount-path=${mountdir}`,
|
||||
};
|
||||
}
|
||||
|
||||
case 'gcs-copy':
|
||||
// No volume mount — artifacts are copied in/out via gsutil commands
|
||||
return { volumeFlags: '', mountFlags: '' };
|
||||
|
||||
default:
|
||||
return { volumeFlags: '', mountFlags: '' };
|
||||
}
|
||||
}
|
||||
|
||||
private async copyArtifactsIn(mountdir: string): Promise<void> {
|
||||
if (this.storageType !== 'gcs-copy' || !this.bucket) return;
|
||||
OrchestratorLogger.log(`[GCP Cloud Run] Copying artifacts from gs://${this.bucket} to ${mountdir}`);
|
||||
try {
|
||||
await OrchestratorSystem.Run(`gcloud storage cp -r "gs://${this.bucket}/*" "${mountdir}/" || true`, false, true);
|
||||
} catch {
|
||||
OrchestratorLogger.log('[GCP Cloud Run] No existing artifacts to restore (bucket may be empty)');
|
||||
}
|
||||
}
|
||||
|
||||
private async copyArtifactsOut(mountdir: string): Promise<void> {
|
||||
if (this.storageType !== 'gcs-copy' || !this.bucket) return;
|
||||
OrchestratorLogger.log(`[GCP Cloud Run] Uploading artifacts from ${mountdir} to gs://${this.bucket}`);
|
||||
await OrchestratorSystem.Run(`gcloud storage cp -r "${mountdir}/*" "gs://${this.bucket}/"`, false, true);
|
||||
}
|
||||
|
||||
async runTaskInWorkflow(
|
||||
buildGuid: string,
|
||||
image: string,
|
||||
commands: string,
|
||||
mountdir: string,
|
||||
workingdir: string,
|
||||
environment: OrchestratorEnvironmentVariable[],
|
||||
secrets: OrchestratorSecret[],
|
||||
): Promise<string> {
|
||||
OrchestratorLogger.log(`[GCP Cloud Run] Running task for build ${buildGuid}`);
|
||||
ResourceTracking.logAllocationSummary('gcp-cloud-run task');
|
||||
|
||||
const jobName = `unity-build-${buildGuid}`
|
||||
.toLowerCase()
|
||||
.replace(/[^a-z0-9-]/g, '-')
|
||||
.slice(0, 63);
|
||||
const projectFlag = this.project ? `--project=${this.project}` : '';
|
||||
|
||||
// Build environment variable flags
|
||||
const envFlags = environment
|
||||
.map((env) => `${env.name}=${env.value}`)
|
||||
.concat(secrets.map((s) => `${s.EnvironmentVariable}=${s.ParameterValue}`));
|
||||
const envString = envFlags.length > 0 ? `--set-env-vars="${envFlags.join(',')}"` : '';
|
||||
|
||||
// Build storage volume flags
|
||||
const { volumeFlags, mountFlags } = this.buildVolumeFlags(mountdir);
|
||||
|
||||
// For gcs-copy, wrap the user command with copy-in/copy-out steps
|
||||
let effectiveCommands = commands;
|
||||
if (this.storageType === 'gcs-copy' && this.bucket && commands) {
|
||||
effectiveCommands = [
|
||||
`gcloud storage cp -r "gs://${this.bucket}/*" "${mountdir}/" 2>/dev/null || true`,
|
||||
commands,
|
||||
`gcloud storage cp -r "${mountdir}/*" "gs://${this.bucket}/"`,
|
||||
].join(' && ');
|
||||
}
|
||||
|
||||
const saFlag = this.serviceAccount ? `--service-account=${this.serviceAccount}` : '';
|
||||
const vpcFlag = this.vpcConnector ? `--vpc-connector=${this.vpcConnector}` : '';
|
||||
|
||||
// Create the Cloud Run Job
|
||||
const createCmd = [
|
||||
'gcloud run jobs create',
|
||||
jobName,
|
||||
`--image=${image}`,
|
||||
`--region=${this.region}`,
|
||||
'--task-timeout=86400s',
|
||||
'--max-retries=0',
|
||||
'--cpu=4',
|
||||
'--memory=16Gi',
|
||||
volumeFlags,
|
||||
mountFlags,
|
||||
envString,
|
||||
saFlag,
|
||||
vpcFlag,
|
||||
projectFlag,
|
||||
'--format=json',
|
||||
'--quiet',
|
||||
]
|
||||
.filter(Boolean)
|
||||
.join(' ');
|
||||
|
||||
try {
|
||||
await OrchestratorSystem.Run(createCmd);
|
||||
OrchestratorLogger.log(`[GCP Cloud Run] Job ${jobName} created`);
|
||||
} catch (error: any) {
|
||||
if (error.message?.includes('already exists')) {
|
||||
OrchestratorLogger.log(`[GCP Cloud Run] Job ${jobName} already exists, updating...`);
|
||||
const updateCmd = createCmd.replace('jobs create', 'jobs update');
|
||||
await OrchestratorSystem.Run(updateCmd);
|
||||
} else {
|
||||
throw error;
|
||||
}
|
||||
}
|
||||
|
||||
// Override the command if provided
|
||||
if (effectiveCommands) {
|
||||
const updateCmd = [
|
||||
'gcloud run jobs update',
|
||||
jobName,
|
||||
`--region=${this.region}`,
|
||||
'--command="/bin/sh"',
|
||||
`--args="-c,${effectiveCommands}"`,
|
||||
projectFlag,
|
||||
'--quiet',
|
||||
]
|
||||
.filter(Boolean)
|
||||
.join(' ');
|
||||
|
||||
await OrchestratorSystem.Run(updateCmd);
|
||||
}
|
||||
|
||||
// Execute the job
|
||||
OrchestratorLogger.log(`[GCP Cloud Run] Executing job ${jobName} (storage: ${this.storageType})...`);
|
||||
const executeCmd = [
|
||||
'gcloud run jobs execute',
|
||||
jobName,
|
||||
`--region=${this.region}`,
|
||||
projectFlag,
|
||||
'--wait',
|
||||
'--format=json',
|
||||
'--quiet',
|
||||
]
|
||||
.filter(Boolean)
|
||||
.join(' ');
|
||||
|
||||
let output = '';
|
||||
try {
|
||||
output = await OrchestratorSystem.Run(executeCmd);
|
||||
OrchestratorLogger.log('[GCP Cloud Run] Job execution completed');
|
||||
} catch (error: any) {
|
||||
await this.streamJobLogs(jobName);
|
||||
throw new Error(`[GCP Cloud Run] Job execution failed: ${error.message}`);
|
||||
}
|
||||
|
||||
await this.streamJobLogs(jobName);
|
||||
return output;
|
||||
}
|
||||
|
||||
private async streamJobLogs(jobName: string): Promise<void> {
|
||||
const projectFlag = this.project ? `--project=${this.project}` : '';
|
||||
try {
|
||||
const logs = await OrchestratorSystem.Run(
|
||||
`gcloud logging read "resource.type=cloud_run_job AND resource.labels.job_name=${jobName}" ${projectFlag} --limit=1000 --format="value(textPayload)" --order=asc`,
|
||||
false,
|
||||
true,
|
||||
);
|
||||
if (logs) {
|
||||
for (const line of logs.split('\n')) {
|
||||
if (line.trim()) {
|
||||
OrchestratorLogger.log(`[Build] ${line}`);
|
||||
}
|
||||
}
|
||||
}
|
||||
} catch {
|
||||
OrchestratorLogger.logWarning('[GCP Cloud Run] Could not retrieve job logs');
|
||||
}
|
||||
}
|
||||
|
||||
async cleanupWorkflow(
|
||||
buildParameters: BuildParameters,
|
||||
branchName: string,
|
||||
defaultSecretsArray: { ParameterKey: string; EnvironmentVariable: string; ParameterValue: string }[],
|
||||
) {
|
||||
OrchestratorLogger.log('[GCP Cloud Run] Cleaning up workflow');
|
||||
}
|
||||
|
||||
async garbageCollect(
|
||||
filter: string,
|
||||
previewOnly: boolean,
|
||||
olderThan: Number,
|
||||
fullCache: boolean,
|
||||
baseDependencies: boolean,
|
||||
): Promise<string> {
|
||||
OrchestratorLogger.log('[GCP Cloud Run] Garbage collecting old jobs');
|
||||
const projectFlag = this.project ? `--project=${this.project}` : '';
|
||||
|
||||
try {
|
||||
const jobsJson = await OrchestratorSystem.Run(
|
||||
`gcloud run jobs list --region=${this.region} ${projectFlag} --filter="metadata.name~unity-build-" --format="json(metadata.name,metadata.creationTimestamp)"`,
|
||||
false,
|
||||
true,
|
||||
);
|
||||
|
||||
const jobs = JSON.parse(jobsJson || '[]');
|
||||
const cutoffDate = new Date();
|
||||
cutoffDate.setDate(cutoffDate.getDate() - Number(olderThan));
|
||||
|
||||
let deletedCount = 0;
|
||||
for (const job of jobs) {
|
||||
const createdAt = new Date(job.metadata?.creationTimestamp || 0);
|
||||
if (createdAt < cutoffDate) {
|
||||
const name = job.metadata?.name;
|
||||
if (previewOnly) {
|
||||
OrchestratorLogger.log(`[GCP Cloud Run] Would delete: ${name}`);
|
||||
} else {
|
||||
await OrchestratorSystem.Run(
|
||||
`gcloud run jobs delete ${name} --region=${this.region} ${projectFlag} --quiet`,
|
||||
);
|
||||
deletedCount++;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return `Garbage collected ${deletedCount} Cloud Run jobs`;
|
||||
} catch (error: any) {
|
||||
OrchestratorLogger.logWarning(`[GCP Cloud Run] Garbage collection failed: ${error.message}`);
|
||||
return '';
|
||||
}
|
||||
}
|
||||
|
||||
async listResources(): Promise<ProviderResource[]> {
|
||||
const projectFlag = this.project ? `--project=${this.project}` : '';
|
||||
try {
|
||||
const jobsJson = await OrchestratorSystem.Run(
|
||||
`gcloud run jobs list --region=${this.region} ${projectFlag} --filter="metadata.name~unity-build-" --format="json(metadata.name)"`,
|
||||
false,
|
||||
true,
|
||||
);
|
||||
|
||||
const jobs = JSON.parse(jobsJson || '[]');
|
||||
return jobs.map((job: any) => ({ Name: job.metadata?.name || '' }));
|
||||
} catch {
|
||||
return [];
|
||||
}
|
||||
}
|
||||
|
||||
listWorkflow(): Promise<ProviderWorkflow[]> {
|
||||
throw new Error('[GCP Cloud Run] listWorkflow not implemented for this experimental provider');
|
||||
}
|
||||
|
||||
async watchWorkflow(): Promise<string> {
|
||||
throw new Error('[GCP Cloud Run] watchWorkflow not implemented for this experimental provider');
|
||||
}
|
||||
}
|
||||
|
||||
export default GcpCloudRunProvider;
|
||||
@@ -62,8 +62,6 @@ export default async function loadProvider(
|
||||
'local-docker': './docker',
|
||||
'local-system': './local',
|
||||
local: './local',
|
||||
'gcp-cloud-run': './gcp-cloud-run',
|
||||
'azure-aci': './azure-aci',
|
||||
};
|
||||
|
||||
modulePath = providerModuleMap[providerSource] || providerSource;
|
||||
@@ -138,7 +136,7 @@ export class ProviderLoader {
|
||||
* @returns string[] - Array of available provider names
|
||||
*/
|
||||
static getAvailableProviders(): string[] {
|
||||
return ['aws', 'k8s', 'test', 'local-docker', 'local-system', 'local', 'gcp-cloud-run', 'azure-aci'];
|
||||
return ['aws', 'k8s', 'test', 'local-docker', 'local-system', 'local'];
|
||||
}
|
||||
|
||||
/**
|
||||
|
||||
5
src/model/orchestrator/runners/README.md
Normal file
5
src/model/orchestrator/runners/README.md
Normal file
@@ -0,0 +1,5 @@
|
||||
# Hot Runner Protocol
|
||||
|
||||
Extensible runner registration and persistent Unity editor provider protocol.
|
||||
|
||||
See GitHub Issue for full specification.
|
||||
@@ -0,0 +1,159 @@
|
||||
import OrchestratorLogger from '../core/orchestrator-logger';
|
||||
import { HotRunnerRegistry } from './hot-runner-registry';
|
||||
import { HotRunnerJobRequest, HotRunnerJobResult, HotRunnerStatus, HotRunnerTransport } from './hot-runner-types';
|
||||
|
||||
const POLL_INTERVAL_MS = 1000;
|
||||
|
||||
// eslint-disable-next-line no-unused-vars
|
||||
export type OutputCallback = (output: string) => void;
|
||||
|
||||
export class HotRunnerDispatcher {
|
||||
private transports: Map<string, HotRunnerTransport>;
|
||||
|
||||
constructor(transports: Map<string, HotRunnerTransport>) {
|
||||
this.transports = transports;
|
||||
}
|
||||
|
||||
/**
|
||||
* Dispatch a job to an available hot runner matching the request's build target.
|
||||
* If no runner is immediately available, waits up to the request timeout.
|
||||
* Returns the job result, or throws if no runner becomes available in time.
|
||||
*/
|
||||
async dispatchJob(
|
||||
request: HotRunnerJobRequest,
|
||||
registry: HotRunnerRegistry,
|
||||
unityVersion: string,
|
||||
onOutput?: OutputCallback,
|
||||
): Promise<HotRunnerJobResult> {
|
||||
OrchestratorLogger.log(`[HotRunner] Dispatching job ${request.jobId} (target: ${request.buildTarget})`);
|
||||
|
||||
// Find or wait for an available runner
|
||||
let runner = registry.findAvailableRunner({
|
||||
unityVersion,
|
||||
platform: request.buildTarget,
|
||||
});
|
||||
|
||||
if (!runner) {
|
||||
OrchestratorLogger.log(
|
||||
`[HotRunner] No idle runner available for ${unityVersion}/${request.buildTarget}, waiting...`,
|
||||
);
|
||||
runner = await this.waitForRunner({ unityVersion, platform: request.buildTarget }, request.timeout, registry);
|
||||
}
|
||||
|
||||
// Mark runner as busy
|
||||
registry.updateRunner(runner.id, {
|
||||
state: 'busy',
|
||||
currentJob: request.jobId,
|
||||
});
|
||||
|
||||
const transport = this.transports.get(runner.id);
|
||||
if (!transport) {
|
||||
registry.updateRunner(runner.id, { state: 'idle', currentJob: undefined });
|
||||
throw new Error(`[HotRunner] No transport available for runner ${runner.id}`);
|
||||
}
|
||||
|
||||
OrchestratorLogger.log(`[HotRunner] Sending job ${request.jobId} to runner ${runner.id}`);
|
||||
|
||||
const startTime = Date.now();
|
||||
|
||||
try {
|
||||
const result = await this.executeWithTimeout(transport, request);
|
||||
|
||||
const duration = Date.now() - startTime;
|
||||
OrchestratorLogger.log(
|
||||
`[HotRunner] Job ${request.jobId} completed on runner ${runner.id} in ${duration}ms (exit: ${result.exitCode})`,
|
||||
);
|
||||
|
||||
if (onOutput && result.output) {
|
||||
onOutput(result.output);
|
||||
}
|
||||
|
||||
// Mark runner as idle and increment job count
|
||||
const currentStatus = registry.getRunner(runner.id);
|
||||
registry.updateRunner(runner.id, {
|
||||
state: 'idle',
|
||||
currentJob: undefined,
|
||||
lastJobCompleted: request.jobId,
|
||||
jobsCompleted: (currentStatus?.jobsCompleted ?? 0) + 1,
|
||||
});
|
||||
|
||||
return result;
|
||||
} catch (error: any) {
|
||||
OrchestratorLogger.logWarning(`[HotRunner] Job ${request.jobId} failed on runner ${runner.id}: ${error.message}`);
|
||||
|
||||
// Mark runner as idle despite failure -- the health monitor will recycle if needed
|
||||
registry.updateRunner(runner.id, {
|
||||
state: 'idle',
|
||||
currentJob: undefined,
|
||||
});
|
||||
|
||||
throw error;
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Wait for an available runner matching the requirements.
|
||||
* Polls the registry at a fixed interval until one becomes available or timeout expires.
|
||||
*/
|
||||
async waitForRunner(
|
||||
requirements: { unityVersion: string; platform: string },
|
||||
timeoutMs: number,
|
||||
registry: HotRunnerRegistry,
|
||||
): Promise<HotRunnerStatus> {
|
||||
const deadline = Date.now() + timeoutMs;
|
||||
|
||||
while (Date.now() < deadline) {
|
||||
const runner = registry.findAvailableRunner(requirements);
|
||||
if (runner) {
|
||||
OrchestratorLogger.log(`[HotRunner] Runner ${runner.id} became available`);
|
||||
|
||||
return runner;
|
||||
}
|
||||
|
||||
await this.sleep(Math.min(POLL_INTERVAL_MS, deadline - Date.now()));
|
||||
}
|
||||
|
||||
throw new Error(
|
||||
`[HotRunner] Timed out waiting for available runner (${requirements.unityVersion}/${requirements.platform}) after ${timeoutMs}ms`,
|
||||
);
|
||||
}
|
||||
|
||||
/**
|
||||
* Execute a job on a transport with a timeout guard.
|
||||
* On timeout, disconnects the transport to release the connection
|
||||
* and prevent the orphaned sendJob promise from holding resources.
|
||||
*/
|
||||
private async executeWithTimeout(
|
||||
transport: HotRunnerTransport,
|
||||
request: HotRunnerJobRequest,
|
||||
): Promise<HotRunnerJobResult> {
|
||||
const TIMEOUT_SENTINEL = Symbol('timeout');
|
||||
|
||||
const timeoutPromise = new Promise<typeof TIMEOUT_SENTINEL>((resolve) => {
|
||||
setTimeout(() => {
|
||||
resolve(TIMEOUT_SENTINEL);
|
||||
}, request.timeout);
|
||||
});
|
||||
|
||||
const result = await Promise.race([transport.sendJob(request), timeoutPromise]);
|
||||
|
||||
if (result === TIMEOUT_SENTINEL) {
|
||||
// Disconnect the transport to clean up the orphaned sendJob call
|
||||
try {
|
||||
await transport.disconnect();
|
||||
} catch (disconnectError: any) {
|
||||
OrchestratorLogger.logWarning(
|
||||
`[HotRunner] Error disconnecting transport after timeout for job ${request.jobId}: ${disconnectError.message}`,
|
||||
);
|
||||
}
|
||||
|
||||
throw new Error(`[HotRunner] Job ${request.jobId} timed out after ${request.timeout}ms`);
|
||||
}
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
private sleep(ms: number): Promise<void> {
|
||||
return new Promise((resolve) => setTimeout(resolve, ms));
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,186 @@
|
||||
import OrchestratorLogger from '../core/orchestrator-logger';
|
||||
import { HotRunnerRegistry } from './hot-runner-registry';
|
||||
import { HotRunnerTransport } from './hot-runner-types';
|
||||
|
||||
export class HotRunnerHealthMonitor {
|
||||
private intervalHandle: ReturnType<typeof setInterval> | undefined;
|
||||
private registry: HotRunnerRegistry | undefined;
|
||||
private transports: Map<string, HotRunnerTransport> = new Map();
|
||||
|
||||
/**
|
||||
* Start periodic health monitoring for all registered runners.
|
||||
*/
|
||||
startMonitoring(registry: HotRunnerRegistry, interval: number, transports: Map<string, HotRunnerTransport>): void {
|
||||
if (this.intervalHandle) {
|
||||
this.stopMonitoring();
|
||||
}
|
||||
|
||||
this.registry = registry;
|
||||
this.transports = transports;
|
||||
|
||||
OrchestratorLogger.log(`[HotRunner] Starting health monitoring (interval: ${interval}s)`);
|
||||
|
||||
this.intervalHandle = setInterval(() => {
|
||||
this.runHealthChecks().catch((error: any) => {
|
||||
OrchestratorLogger.logWarning(`[HotRunner] Health check cycle failed: ${error.message}`);
|
||||
});
|
||||
}, interval * 1000);
|
||||
}
|
||||
|
||||
/**
|
||||
* Stop periodic health monitoring.
|
||||
*/
|
||||
stopMonitoring(): void {
|
||||
if (this.intervalHandle) {
|
||||
clearInterval(this.intervalHandle);
|
||||
this.intervalHandle = undefined;
|
||||
OrchestratorLogger.log(`[HotRunner] Health monitoring stopped`);
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Check health of a specific runner by ID. Returns true if healthy.
|
||||
*/
|
||||
async checkHealth(runnerId: string): Promise<boolean> {
|
||||
if (!this.registry) {
|
||||
return false;
|
||||
}
|
||||
|
||||
const transport = this.transports.get(runnerId);
|
||||
if (!transport) {
|
||||
OrchestratorLogger.logWarning(`[HotRunner] No transport for runner ${runnerId}`);
|
||||
this.registry.updateRunner(runnerId, {
|
||||
state: 'unhealthy',
|
||||
lastHealthCheck: new Date().toISOString(),
|
||||
});
|
||||
|
||||
return false;
|
||||
}
|
||||
|
||||
try {
|
||||
const healthy = await transport.healthCheck();
|
||||
if (healthy) {
|
||||
const status = await transport.getStatus();
|
||||
this.registry.updateRunner(runnerId, {
|
||||
lastHealthCheck: new Date().toISOString(),
|
||||
memoryUsageMB: status.memoryUsageMB,
|
||||
uptime: status.uptime,
|
||||
libraryHash: status.libraryHash,
|
||||
});
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
OrchestratorLogger.logWarning(`[HotRunner] Runner ${runnerId} health check returned false`);
|
||||
this.registry.updateRunner(runnerId, {
|
||||
state: 'unhealthy',
|
||||
lastHealthCheck: new Date().toISOString(),
|
||||
});
|
||||
|
||||
return false;
|
||||
} catch (error: any) {
|
||||
OrchestratorLogger.logWarning(`[HotRunner] Runner ${runnerId} health check failed: ${error.message}`);
|
||||
this.registry.updateRunner(runnerId, {
|
||||
state: 'unhealthy',
|
||||
lastHealthCheck: new Date().toISOString(),
|
||||
});
|
||||
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Mark an unhealthy runner for cleanup and disconnect its transport.
|
||||
*/
|
||||
async recycleUnhealthyRunner(runnerId: string): Promise<void> {
|
||||
if (!this.registry) {
|
||||
return;
|
||||
}
|
||||
|
||||
OrchestratorLogger.log(`[HotRunner] Recycling unhealthy runner ${runnerId}`);
|
||||
this.registry.updateRunner(runnerId, { state: 'stopping' });
|
||||
|
||||
const transport = this.transports.get(runnerId);
|
||||
if (transport) {
|
||||
try {
|
||||
await transport.disconnect();
|
||||
} catch (error: any) {
|
||||
OrchestratorLogger.logWarning(`[HotRunner] Error disconnecting runner ${runnerId}: ${error.message}`);
|
||||
}
|
||||
this.transports.delete(runnerId);
|
||||
}
|
||||
|
||||
this.registry.unregisterRunner(runnerId);
|
||||
OrchestratorLogger.log(`[HotRunner] Runner ${runnerId} recycled and removed`);
|
||||
}
|
||||
|
||||
/**
|
||||
* Recycle a runner that has been idle longer than the maximum idle time.
|
||||
*/
|
||||
async recycleIdleRunner(runnerId: string, maxIdleTime: number): Promise<void> {
|
||||
if (!this.registry) {
|
||||
return;
|
||||
}
|
||||
|
||||
const runner = this.registry.getRunner(runnerId);
|
||||
if (!runner || runner.state !== 'idle') {
|
||||
return;
|
||||
}
|
||||
|
||||
const lastCheckTime = new Date(runner.lastHealthCheck).getTime();
|
||||
const now = Date.now();
|
||||
const idleSeconds = (now - lastCheckTime) / 1000;
|
||||
|
||||
if (idleSeconds >= maxIdleTime) {
|
||||
OrchestratorLogger.log(
|
||||
`[HotRunner] Runner ${runnerId} idle for ${Math.floor(idleSeconds)}s (max: ${maxIdleTime}s), recycling`,
|
||||
);
|
||||
await this.recycleUnhealthyRunner(runnerId);
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Run health checks and idle-recycle checks for all registered runners.
|
||||
*/
|
||||
private async runHealthChecks(): Promise<void> {
|
||||
if (!this.registry) {
|
||||
return;
|
||||
}
|
||||
|
||||
const runners = this.registry.listRunners();
|
||||
|
||||
for (const runner of runners) {
|
||||
if (runner.state === 'stopping') {
|
||||
continue;
|
||||
}
|
||||
|
||||
const healthy = await this.checkHealth(runner.id);
|
||||
|
||||
if (!healthy && runner.state !== 'starting') {
|
||||
await this.recycleUnhealthyRunner(runner.id);
|
||||
continue;
|
||||
}
|
||||
|
||||
// Check for idle timeout
|
||||
const config = this.registry.getConfig(runner.id);
|
||||
if (config && runner.state === 'idle') {
|
||||
await this.recycleIdleRunner(runner.id, config.maxIdleTime);
|
||||
}
|
||||
|
||||
// Check for max jobs before recycle
|
||||
if (config && config.maxJobsBeforeRecycle > 0 && runner.jobsCompleted >= config.maxJobsBeforeRecycle) {
|
||||
OrchestratorLogger.log(
|
||||
`[HotRunner] Runner ${runner.id} reached max jobs (${runner.jobsCompleted}/${config.maxJobsBeforeRecycle}), recycling`,
|
||||
);
|
||||
await this.recycleUnhealthyRunner(runner.id);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Whether health monitoring is currently active.
|
||||
*/
|
||||
get isMonitoring(): boolean {
|
||||
return this.intervalHandle !== undefined;
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,315 @@
|
||||
import fs from 'node:fs';
|
||||
import path from 'node:path';
|
||||
import { customAlphabet } from 'nanoid';
|
||||
import OrchestratorLogger from '../core/orchestrator-logger';
|
||||
import { HotRunnerConfig, HotRunnerStatus } from './hot-runner-types';
|
||||
|
||||
const generateId = customAlphabet('abcdefghijklmnopqrstuvwxyz0123456789', 12);
|
||||
|
||||
const PERSISTENCE_FILENAME = 'hot-runners.json';
|
||||
|
||||
const VALID_RUNNER_STATES: ReadonlySet<string> = new Set(['idle', 'busy', 'starting', 'stopping', 'unhealthy']);
|
||||
|
||||
export interface HotRunnerFilter {
|
||||
platform?: string;
|
||||
state?: string;
|
||||
unityVersion?: string;
|
||||
}
|
||||
|
||||
/**
|
||||
* Validate that a restored runner entry has all required fields with correct types.
|
||||
* Returns true if the entry is a valid HotRunnerStatus, false otherwise.
|
||||
*/
|
||||
function isValidRunnerStatus(entry: unknown): entry is HotRunnerStatus {
|
||||
if (typeof entry !== 'object' || entry === null) {
|
||||
return false;
|
||||
}
|
||||
|
||||
const record = entry as Record<string, unknown>;
|
||||
|
||||
return (
|
||||
typeof record.id === 'string' &&
|
||||
record.id.length > 0 &&
|
||||
typeof record.state === 'string' &&
|
||||
VALID_RUNNER_STATES.has(record.state) &&
|
||||
typeof record.unityVersion === 'string' &&
|
||||
typeof record.platform === 'string' &&
|
||||
typeof record.uptime === 'number' &&
|
||||
typeof record.jobsCompleted === 'number' &&
|
||||
typeof record.lastHealthCheck === 'string' &&
|
||||
typeof record.memoryUsageMB === 'number'
|
||||
);
|
||||
}
|
||||
|
||||
/**
|
||||
* Validate that a restored config entry has all required fields with correct types.
|
||||
* Returns true if the entry is a valid HotRunnerConfig, false otherwise.
|
||||
*/
|
||||
function isValidRunnerConfig(entry: unknown): entry is HotRunnerConfig {
|
||||
if (typeof entry !== 'object' || entry === null) {
|
||||
return false;
|
||||
}
|
||||
|
||||
const record = entry as Record<string, unknown>;
|
||||
|
||||
return (
|
||||
typeof record.enabled === 'boolean' &&
|
||||
typeof record.transport === 'string' &&
|
||||
['websocket', 'grpc', 'named-pipe'].includes(record.transport) &&
|
||||
typeof record.host === 'string' &&
|
||||
typeof record.port === 'number' &&
|
||||
typeof record.healthCheckInterval === 'number' &&
|
||||
typeof record.maxIdleTime === 'number' &&
|
||||
typeof record.maxJobsBeforeRecycle === 'number'
|
||||
);
|
||||
}
|
||||
|
||||
export class HotRunnerRegistry {
|
||||
private runners: Map<string, HotRunnerStatus> = new Map();
|
||||
private configs: Map<string, HotRunnerConfig> = new Map();
|
||||
private persistencePath: string;
|
||||
|
||||
constructor(persistenceDirectory?: string) {
|
||||
this.persistencePath = persistenceDirectory ? path.join(persistenceDirectory, PERSISTENCE_FILENAME) : '';
|
||||
}
|
||||
|
||||
/**
|
||||
* Register a new hot runner. Returns the generated runner ID.
|
||||
*/
|
||||
registerRunner(config: HotRunnerConfig): string {
|
||||
const id = `hr-${generateId()}`;
|
||||
|
||||
const status: HotRunnerStatus = {
|
||||
id,
|
||||
state: 'starting',
|
||||
unityVersion: config.unityVersion ?? 'unknown',
|
||||
platform: config.platform ?? 'unknown',
|
||||
uptime: 0,
|
||||
jobsCompleted: 0,
|
||||
lastHealthCheck: new Date().toISOString(),
|
||||
memoryUsageMB: 0,
|
||||
};
|
||||
|
||||
this.runners.set(id, status);
|
||||
this.configs.set(id, config);
|
||||
OrchestratorLogger.log(`[HotRunner] Registered runner ${id} (${status.unityVersion}/${status.platform})`);
|
||||
|
||||
this.persist();
|
||||
|
||||
return id;
|
||||
}
|
||||
|
||||
/**
|
||||
* Remove a runner from the registry.
|
||||
*/
|
||||
unregisterRunner(id: string): void {
|
||||
const existed = this.runners.delete(id);
|
||||
this.configs.delete(id);
|
||||
|
||||
if (existed) {
|
||||
OrchestratorLogger.log(`[HotRunner] Unregistered runner ${id}`);
|
||||
this.persist();
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Get a runner's current status by ID.
|
||||
*/
|
||||
getRunner(id: string): HotRunnerStatus | undefined {
|
||||
return this.runners.get(id);
|
||||
}
|
||||
|
||||
/**
|
||||
* Get a runner's config by ID.
|
||||
*/
|
||||
getConfig(id: string): HotRunnerConfig | undefined {
|
||||
return this.configs.get(id);
|
||||
}
|
||||
|
||||
/**
|
||||
* List all runners, optionally filtered by platform, state, or Unity version.
|
||||
*/
|
||||
listRunners(filter?: HotRunnerFilter): HotRunnerStatus[] {
|
||||
let results = [...this.runners.values()];
|
||||
|
||||
if (filter?.platform) {
|
||||
results = results.filter((runner) => runner.platform === filter.platform);
|
||||
}
|
||||
|
||||
if (filter?.state) {
|
||||
results = results.filter((runner) => runner.state === filter.state);
|
||||
}
|
||||
|
||||
if (filter?.unityVersion) {
|
||||
results = results.filter((runner) => runner.unityVersion === filter.unityVersion);
|
||||
}
|
||||
|
||||
return results;
|
||||
}
|
||||
|
||||
/**
|
||||
* Find an idle runner matching the given Unity version and platform requirements.
|
||||
*/
|
||||
findAvailableRunner(requirements: { unityVersion: string; platform: string }): HotRunnerStatus | undefined {
|
||||
return this.listRunners({
|
||||
state: 'idle',
|
||||
unityVersion: requirements.unityVersion,
|
||||
platform: requirements.platform,
|
||||
})[0];
|
||||
}
|
||||
|
||||
/**
|
||||
* Update a runner's status fields. Merges partial updates into existing status.
|
||||
*/
|
||||
updateRunner(id: string, update: Partial<HotRunnerStatus>): void {
|
||||
const existing = this.runners.get(id);
|
||||
if (!existing) {
|
||||
return;
|
||||
}
|
||||
|
||||
this.runners.set(id, { ...existing, ...update, id });
|
||||
this.persist();
|
||||
}
|
||||
|
||||
/**
|
||||
* Get the total number of registered runners.
|
||||
*/
|
||||
get size(): number {
|
||||
return this.runners.size;
|
||||
}
|
||||
|
||||
/**
|
||||
* Validate all runners in the registry and reset invalid ones to 'unhealthy'.
|
||||
* Returns the number of runners that were repaired.
|
||||
*/
|
||||
validateAndRepair(): number {
|
||||
let repaired = 0;
|
||||
|
||||
for (const [id, status] of this.runners) {
|
||||
// Cast to unknown to bypass the type guard narrowing to 'never',
|
||||
// since the Map is typed as HotRunnerStatus but entries may have
|
||||
// been corrupted via direct deserialization or unsafe casts.
|
||||
const entry = status as unknown as Record<string, unknown>;
|
||||
|
||||
if (!isValidRunnerStatus(entry)) {
|
||||
OrchestratorLogger.logWarning(`[HotRunner] Runner ${id} has invalid state, marking as unhealthy`);
|
||||
this.runners.set(id, {
|
||||
id,
|
||||
state: 'unhealthy',
|
||||
unityVersion: typeof entry.unityVersion === 'string' ? entry.unityVersion : 'unknown',
|
||||
platform: typeof entry.platform === 'string' ? entry.platform : 'unknown',
|
||||
uptime: typeof entry.uptime === 'number' ? entry.uptime : 0,
|
||||
jobsCompleted: typeof entry.jobsCompleted === 'number' ? entry.jobsCompleted : 0,
|
||||
lastHealthCheck: typeof entry.lastHealthCheck === 'string' ? entry.lastHealthCheck : new Date().toISOString(),
|
||||
memoryUsageMB: typeof entry.memoryUsageMB === 'number' ? entry.memoryUsageMB : 0,
|
||||
});
|
||||
repaired++;
|
||||
}
|
||||
}
|
||||
|
||||
if (repaired > 0) {
|
||||
this.persist();
|
||||
}
|
||||
|
||||
return repaired;
|
||||
}
|
||||
|
||||
/**
|
||||
* Persist current registry state to disk for crash recovery.
|
||||
* Validates data before writing to prevent persisting corrupt state.
|
||||
*/
|
||||
private persist(): void {
|
||||
if (!this.persistencePath) {
|
||||
return;
|
||||
}
|
||||
|
||||
try {
|
||||
// Validate data before persisting
|
||||
for (const [id, status] of this.runners) {
|
||||
if (!isValidRunnerStatus(status)) {
|
||||
OrchestratorLogger.logWarning(`[HotRunner] Skipping persistence -- runner ${id} has invalid state`);
|
||||
|
||||
return;
|
||||
}
|
||||
}
|
||||
|
||||
const data = {
|
||||
runners: Object.fromEntries(this.runners),
|
||||
configs: Object.fromEntries(this.configs),
|
||||
};
|
||||
const directory = path.dirname(this.persistencePath);
|
||||
if (!fs.existsSync(directory)) {
|
||||
fs.mkdirSync(directory, { recursive: true });
|
||||
}
|
||||
|
||||
fs.writeFileSync(this.persistencePath, JSON.stringify(data, undefined, 2));
|
||||
} catch (error: any) {
|
||||
OrchestratorLogger.logWarning(`[HotRunner] Failed to persist registry: ${error.message}`);
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Load registry state from disk. Returns the number of runners restored.
|
||||
* Validates each restored entry and discards corrupt entries with warnings.
|
||||
* If the persistence file itself is corrupt (invalid JSON), starts with
|
||||
* an empty registry.
|
||||
*/
|
||||
loadFromDisk(): number {
|
||||
if (!this.persistencePath || !fs.existsSync(this.persistencePath)) {
|
||||
return 0;
|
||||
}
|
||||
|
||||
let data: any;
|
||||
|
||||
try {
|
||||
const raw = fs.readFileSync(this.persistencePath, 'utf8');
|
||||
data = JSON.parse(raw);
|
||||
} catch (error: any) {
|
||||
OrchestratorLogger.logWarning(
|
||||
`[HotRunner] Persistence file is corrupt, starting with empty registry: ${error.message}`,
|
||||
);
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
if (typeof data !== 'object' || data === null) {
|
||||
OrchestratorLogger.logWarning('[HotRunner] Persistence file has invalid structure, starting with empty registry');
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
let discarded = 0;
|
||||
|
||||
if (data.runners && typeof data.runners === 'object') {
|
||||
for (const [id, status] of Object.entries(data.runners)) {
|
||||
if (isValidRunnerStatus(status)) {
|
||||
this.runners.set(id, status);
|
||||
} else {
|
||||
OrchestratorLogger.logWarning(`[HotRunner] Discarding invalid runner entry '${id}' from persistence file`);
|
||||
discarded++;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (data.configs && typeof data.configs === 'object') {
|
||||
for (const [id, config] of Object.entries(data.configs)) {
|
||||
// Only restore configs for runners that were successfully restored
|
||||
if (this.runners.has(id)) {
|
||||
if (isValidRunnerConfig(config)) {
|
||||
this.configs.set(id, config);
|
||||
} else {
|
||||
OrchestratorLogger.logWarning(`[HotRunner] Discarding invalid config entry '${id}' from persistence file`);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (discarded > 0) {
|
||||
OrchestratorLogger.logWarning(`[HotRunner] Discarded ${discarded} invalid runner(s) from persistence file`);
|
||||
}
|
||||
|
||||
OrchestratorLogger.log(`[HotRunner] Restored ${this.runners.size} runner(s) from disk`);
|
||||
|
||||
return this.runners.size;
|
||||
}
|
||||
}
|
||||
166
src/model/orchestrator/services/hot-runner/hot-runner-service.ts
Normal file
166
src/model/orchestrator/services/hot-runner/hot-runner-service.ts
Normal file
@@ -0,0 +1,166 @@
|
||||
import OrchestratorLogger from '../core/orchestrator-logger';
|
||||
import { HotRunnerRegistry } from './hot-runner-registry';
|
||||
import { HotRunnerHealthMonitor } from './hot-runner-health-monitor';
|
||||
import { HotRunnerDispatcher, OutputCallback } from './hot-runner-dispatcher';
|
||||
import {
|
||||
HotRunnerConfig,
|
||||
HotRunnerJobRequest,
|
||||
HotRunnerJobResult,
|
||||
HotRunnerStatus,
|
||||
HotRunnerTransport,
|
||||
} from './hot-runner-types';
|
||||
import BuildParameters from '../../../build-parameters';
|
||||
|
||||
export class HotRunnerService {
|
||||
private registry: HotRunnerRegistry;
|
||||
private healthMonitor: HotRunnerHealthMonitor;
|
||||
private dispatcher: HotRunnerDispatcher;
|
||||
private transports: Map<string, HotRunnerTransport> = new Map();
|
||||
private config: HotRunnerConfig | undefined;
|
||||
|
||||
constructor(persistenceDirectory?: string) {
|
||||
this.registry = new HotRunnerRegistry(persistenceDirectory);
|
||||
this.healthMonitor = new HotRunnerHealthMonitor();
|
||||
this.dispatcher = new HotRunnerDispatcher(this.transports);
|
||||
}
|
||||
|
||||
/**
|
||||
* Initialize the hot runner service: load persisted state, start health monitoring.
|
||||
*/
|
||||
async initialize(config: HotRunnerConfig): Promise<void> {
|
||||
this.config = config;
|
||||
|
||||
OrchestratorLogger.log(
|
||||
`[HotRunner] Initializing service (transport: ${config.transport}, ${config.host}:${config.port})`,
|
||||
);
|
||||
|
||||
// Attempt to restore previously registered runners from disk
|
||||
const restored = this.registry.loadFromDisk();
|
||||
if (restored > 0) {
|
||||
OrchestratorLogger.log(`[HotRunner] Restored ${restored} runner(s) from persistence`);
|
||||
}
|
||||
|
||||
// Start health monitoring
|
||||
this.healthMonitor.startMonitoring(this.registry, config.healthCheckInterval, this.transports);
|
||||
|
||||
OrchestratorLogger.log(`[HotRunner] Service initialized`);
|
||||
}
|
||||
|
||||
/**
|
||||
* Register a runner with a transport implementation.
|
||||
* Returns the runner ID.
|
||||
*/
|
||||
registerRunner(config: HotRunnerConfig, transport: HotRunnerTransport): string {
|
||||
const id = this.registry.registerRunner(config);
|
||||
this.transports.set(id, transport);
|
||||
|
||||
return id;
|
||||
}
|
||||
|
||||
/**
|
||||
* Submit a build job to an available hot runner.
|
||||
* Converts BuildParameters to a HotRunnerJobRequest and dispatches.
|
||||
*/
|
||||
async submitBuild(params: BuildParameters, onOutput?: OutputCallback): Promise<HotRunnerJobResult> {
|
||||
const request: HotRunnerJobRequest = {
|
||||
jobId: params.buildGuid || `build-${Date.now()}`,
|
||||
buildMethod: params.buildMethod || undefined,
|
||||
buildTarget: params.targetPlatform,
|
||||
buildPath: params.buildPath,
|
||||
customParameters: params.customParameters ? this.parseCustomParameters(params.customParameters) : undefined,
|
||||
timeout: 30 * 60 * 1000, // 30 minutes default
|
||||
};
|
||||
|
||||
OrchestratorLogger.log(`[HotRunner] Submitting build: ${request.jobId} (target: ${request.buildTarget})`);
|
||||
|
||||
return this.dispatcher.dispatchJob(request, this.registry, params.editorVersion, onOutput);
|
||||
}
|
||||
|
||||
/**
|
||||
* Submit a test job to an available hot runner.
|
||||
* Converts BuildParameters and optional suite config to a test-mode HotRunnerJobRequest.
|
||||
*/
|
||||
async submitTest(
|
||||
params: BuildParameters,
|
||||
suiteConfig?: { testMode?: 'editmode' | 'playmode'; testSuitePath?: string },
|
||||
onOutput?: OutputCallback,
|
||||
): Promise<HotRunnerJobResult> {
|
||||
const request: HotRunnerJobRequest = {
|
||||
jobId: params.buildGuid || `test-${Date.now()}`,
|
||||
buildTarget: params.targetPlatform,
|
||||
customParameters: params.customParameters ? this.parseCustomParameters(params.customParameters) : undefined,
|
||||
timeout: 30 * 60 * 1000, // 30 minutes default
|
||||
testMode: suiteConfig?.testMode ?? 'editmode',
|
||||
testSuitePath: suiteConfig?.testSuitePath,
|
||||
};
|
||||
|
||||
OrchestratorLogger.log(`[HotRunner] Submitting test: ${request.jobId} (mode: ${request.testMode})`);
|
||||
|
||||
return this.dispatcher.dispatchJob(request, this.registry, params.editorVersion, onOutput);
|
||||
}
|
||||
|
||||
/**
|
||||
* Shut down the service: stop health monitoring, disconnect all transports,
|
||||
* and unregister all runners.
|
||||
*/
|
||||
async shutdown(): Promise<void> {
|
||||
OrchestratorLogger.log(`[HotRunner] Shutting down service`);
|
||||
|
||||
this.healthMonitor.stopMonitoring();
|
||||
|
||||
const disconnectPromises: Promise<void>[] = [];
|
||||
for (const [id, transport] of this.transports.entries()) {
|
||||
disconnectPromises.push(
|
||||
transport.disconnect().catch((error: any) => {
|
||||
OrchestratorLogger.logWarning(`[HotRunner] Error disconnecting runner ${id}: ${error.message}`);
|
||||
}),
|
||||
);
|
||||
}
|
||||
await Promise.all(disconnectPromises);
|
||||
|
||||
this.transports.clear();
|
||||
|
||||
OrchestratorLogger.log(`[HotRunner] Service shut down`);
|
||||
}
|
||||
|
||||
/**
|
||||
* Get the status of all registered runners.
|
||||
*/
|
||||
getStatus(): HotRunnerStatus[] {
|
||||
return this.registry.listRunners();
|
||||
}
|
||||
|
||||
/**
|
||||
* Get the underlying registry (for testing or advanced use).
|
||||
*/
|
||||
getRegistry(): HotRunnerRegistry {
|
||||
return this.registry;
|
||||
}
|
||||
|
||||
/**
|
||||
* Parse a space-separated custom parameters string into a key-value map.
|
||||
* Handles `-key value` and `-key=value` formats.
|
||||
*/
|
||||
private parseCustomParameters(raw: string): Record<string, string> {
|
||||
const result: Record<string, string> = {};
|
||||
const parts = raw.trim().split(/\s+/);
|
||||
|
||||
for (let i = 0; i < parts.length; i++) {
|
||||
const part = parts[i];
|
||||
if (part.startsWith('-')) {
|
||||
const key = part.replace(/^-+/, '');
|
||||
if (key.includes('=')) {
|
||||
const [k, ...v] = key.split('=');
|
||||
result[k] = v.join('=');
|
||||
} else if (i + 1 < parts.length && !parts[i + 1].startsWith('-')) {
|
||||
result[key] = parts[i + 1];
|
||||
i++;
|
||||
} else {
|
||||
result[key] = 'true';
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return result;
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,54 @@
|
||||
export interface HotRunnerConfig {
|
||||
enabled: boolean;
|
||||
transport: 'websocket' | 'grpc' | 'named-pipe';
|
||||
host: string;
|
||||
port: number;
|
||||
healthCheckInterval: number; // seconds
|
||||
maxIdleTime: number; // seconds before recycling
|
||||
maxJobsBeforeRecycle: number;
|
||||
unityVersion?: string;
|
||||
platform?: string;
|
||||
}
|
||||
|
||||
export interface HotRunnerStatus {
|
||||
id: string;
|
||||
state: 'idle' | 'busy' | 'starting' | 'stopping' | 'unhealthy';
|
||||
unityVersion: string;
|
||||
platform: string;
|
||||
currentJob?: string;
|
||||
lastJobCompleted?: string;
|
||||
uptime: number;
|
||||
jobsCompleted: number;
|
||||
lastHealthCheck: string;
|
||||
memoryUsageMB: number;
|
||||
libraryHash?: string;
|
||||
}
|
||||
|
||||
export interface HotRunnerJobRequest {
|
||||
jobId: string;
|
||||
buildMethod?: string;
|
||||
buildTarget: string;
|
||||
buildPath?: string;
|
||||
customParameters?: Record<string, string>;
|
||||
timeout: number;
|
||||
testMode?: 'editmode' | 'playmode';
|
||||
testSuitePath?: string;
|
||||
}
|
||||
|
||||
export interface HotRunnerJobResult {
|
||||
jobId: string;
|
||||
success: boolean;
|
||||
exitCode: number;
|
||||
duration: number;
|
||||
output: string;
|
||||
artifacts?: string[];
|
||||
testResults?: string; // path to test result file
|
||||
}
|
||||
|
||||
export interface HotRunnerTransport {
|
||||
connect(config: HotRunnerConfig): Promise<void>;
|
||||
disconnect(): Promise<void>;
|
||||
sendJob(request: HotRunnerJobRequest): Promise<HotRunnerJobResult>;
|
||||
getStatus(): Promise<HotRunnerStatus>;
|
||||
healthCheck(): Promise<boolean>;
|
||||
}
|
||||
740
src/model/orchestrator/services/hot-runner/hot-runner.test.ts
Normal file
740
src/model/orchestrator/services/hot-runner/hot-runner.test.ts
Normal file
@@ -0,0 +1,740 @@
|
||||
import fs from 'node:fs';
|
||||
import { HotRunnerRegistry } from './hot-runner-registry';
|
||||
import { HotRunnerHealthMonitor } from './hot-runner-health-monitor';
|
||||
import { HotRunnerDispatcher } from './hot-runner-dispatcher';
|
||||
import { HotRunnerService } from './hot-runner-service';
|
||||
import {
|
||||
HotRunnerConfig,
|
||||
HotRunnerJobRequest,
|
||||
HotRunnerJobResult,
|
||||
HotRunnerStatus,
|
||||
HotRunnerTransport,
|
||||
} from './hot-runner-types';
|
||||
|
||||
// Mock dependencies
|
||||
jest.mock('node:fs');
|
||||
jest.mock('../core/orchestrator-logger');
|
||||
|
||||
const mockFs = fs as jest.Mocked<typeof fs>;
|
||||
|
||||
function createMockConfig(overrides?: Partial<HotRunnerConfig>): HotRunnerConfig {
|
||||
return {
|
||||
enabled: true,
|
||||
transport: 'websocket',
|
||||
host: 'localhost',
|
||||
port: 9090,
|
||||
healthCheckInterval: 30,
|
||||
maxIdleTime: 3600,
|
||||
maxJobsBeforeRecycle: 100,
|
||||
unityVersion: '2022.3.0f1',
|
||||
platform: 'StandaloneWindows64',
|
||||
...overrides,
|
||||
};
|
||||
}
|
||||
|
||||
function createMockTransport(overrides?: Partial<HotRunnerTransport>): HotRunnerTransport {
|
||||
return {
|
||||
// eslint-disable-next-line unicorn/no-useless-undefined
|
||||
connect: jest.fn().mockResolvedValue(undefined),
|
||||
// eslint-disable-next-line unicorn/no-useless-undefined
|
||||
disconnect: jest.fn().mockResolvedValue(undefined),
|
||||
sendJob: jest.fn().mockResolvedValue({
|
||||
jobId: 'test-job',
|
||||
success: true,
|
||||
exitCode: 0,
|
||||
duration: 5000,
|
||||
output: 'Build succeeded',
|
||||
artifacts: ['build/output.exe'],
|
||||
} as HotRunnerJobResult),
|
||||
getStatus: jest.fn().mockResolvedValue({
|
||||
id: 'mock-runner',
|
||||
state: 'idle',
|
||||
unityVersion: '2022.3.0f1',
|
||||
platform: 'StandaloneWindows64',
|
||||
uptime: 3600,
|
||||
jobsCompleted: 5,
|
||||
lastHealthCheck: new Date().toISOString(),
|
||||
memoryUsageMB: 1024,
|
||||
} as HotRunnerStatus),
|
||||
healthCheck: jest.fn().mockResolvedValue(true),
|
||||
...overrides,
|
||||
};
|
||||
}
|
||||
|
||||
function createMockJobRequest(overrides?: Partial<HotRunnerJobRequest>): HotRunnerJobRequest {
|
||||
return {
|
||||
jobId: 'job-001',
|
||||
buildTarget: 'StandaloneWindows64',
|
||||
timeout: 60000,
|
||||
...overrides,
|
||||
};
|
||||
}
|
||||
|
||||
// --- Registry Tests ---
|
||||
|
||||
describe('HotRunnerRegistry', () => {
|
||||
let registry: HotRunnerRegistry;
|
||||
|
||||
beforeEach(() => {
|
||||
jest.clearAllMocks();
|
||||
registry = new HotRunnerRegistry();
|
||||
});
|
||||
|
||||
it('should register a runner and return an ID', () => {
|
||||
const config = createMockConfig();
|
||||
const id = registry.registerRunner(config);
|
||||
|
||||
expect(id).toMatch(/^hr-/);
|
||||
expect(registry.size).toBe(1);
|
||||
});
|
||||
|
||||
it('should retrieve a registered runner by ID', () => {
|
||||
const config = createMockConfig();
|
||||
const id = registry.registerRunner(config);
|
||||
const runner = registry.getRunner(id);
|
||||
|
||||
expect(runner).toBeDefined();
|
||||
expect(runner!.id).toBe(id);
|
||||
expect(runner!.state).toBe('starting');
|
||||
expect(runner!.unityVersion).toBe('2022.3.0f1');
|
||||
expect(runner!.platform).toBe('StandaloneWindows64');
|
||||
});
|
||||
|
||||
it('should return undefined for unknown runner ID', () => {
|
||||
const runner = registry.getRunner('nonexistent');
|
||||
expect(runner).toBeUndefined();
|
||||
});
|
||||
|
||||
it('should unregister a runner', () => {
|
||||
const id = registry.registerRunner(createMockConfig());
|
||||
expect(registry.size).toBe(1);
|
||||
|
||||
registry.unregisterRunner(id);
|
||||
expect(registry.size).toBe(0);
|
||||
expect(registry.getRunner(id)).toBeUndefined();
|
||||
});
|
||||
|
||||
it('should handle unregistering a nonexistent runner gracefully', () => {
|
||||
registry.unregisterRunner('nonexistent');
|
||||
expect(registry.size).toBe(0);
|
||||
});
|
||||
|
||||
it('should list all runners without filter', () => {
|
||||
registry.registerRunner(createMockConfig({ platform: 'StandaloneWindows64' }));
|
||||
registry.registerRunner(createMockConfig({ platform: 'StandaloneLinux64' }));
|
||||
registry.registerRunner(createMockConfig({ platform: 'StandaloneOSX' }));
|
||||
|
||||
const all = registry.listRunners();
|
||||
expect(all).toHaveLength(3);
|
||||
});
|
||||
|
||||
it('should filter runners by platform', () => {
|
||||
registry.registerRunner(createMockConfig({ platform: 'StandaloneWindows64' }));
|
||||
registry.registerRunner(createMockConfig({ platform: 'StandaloneLinux64' }));
|
||||
registry.registerRunner(createMockConfig({ platform: 'StandaloneWindows64' }));
|
||||
|
||||
const windows = registry.listRunners({ platform: 'StandaloneWindows64' });
|
||||
expect(windows).toHaveLength(2);
|
||||
|
||||
const linux = registry.listRunners({ platform: 'StandaloneLinux64' });
|
||||
expect(linux).toHaveLength(1);
|
||||
});
|
||||
|
||||
it('should filter runners by state', () => {
|
||||
const id1 = registry.registerRunner(createMockConfig());
|
||||
registry.registerRunner(createMockConfig());
|
||||
|
||||
registry.updateRunner(id1, { state: 'idle' });
|
||||
|
||||
// second runner remains in 'starting' state
|
||||
|
||||
const idle = registry.listRunners({ state: 'idle' });
|
||||
expect(idle).toHaveLength(1);
|
||||
expect(idle[0].id).toBe(id1);
|
||||
});
|
||||
|
||||
it('should filter runners by Unity version', () => {
|
||||
registry.registerRunner(createMockConfig({ unityVersion: '2022.3.0f1' }));
|
||||
registry.registerRunner(createMockConfig({ unityVersion: '2023.1.0f1' }));
|
||||
registry.registerRunner(createMockConfig({ unityVersion: '2022.3.0f1' }));
|
||||
|
||||
const v2022 = registry.listRunners({ unityVersion: '2022.3.0f1' });
|
||||
expect(v2022).toHaveLength(2);
|
||||
});
|
||||
|
||||
it('should find an available idle runner matching requirements', () => {
|
||||
const id1 = registry.registerRunner(
|
||||
createMockConfig({ unityVersion: '2022.3.0f1', platform: 'StandaloneWindows64' }),
|
||||
);
|
||||
registry.updateRunner(id1, { state: 'idle' });
|
||||
|
||||
const id2 = registry.registerRunner(
|
||||
createMockConfig({ unityVersion: '2023.1.0f1', platform: 'StandaloneLinux64' }),
|
||||
);
|
||||
registry.updateRunner(id2, { state: 'idle' });
|
||||
|
||||
const found = registry.findAvailableRunner({
|
||||
unityVersion: '2022.3.0f1',
|
||||
platform: 'StandaloneWindows64',
|
||||
});
|
||||
|
||||
expect(found).toBeDefined();
|
||||
expect(found!.id).toBe(id1);
|
||||
});
|
||||
|
||||
it('should return undefined when no runner matches requirements', () => {
|
||||
const id = registry.registerRunner(
|
||||
createMockConfig({ unityVersion: '2022.3.0f1', platform: 'StandaloneWindows64' }),
|
||||
);
|
||||
registry.updateRunner(id, { state: 'idle' });
|
||||
|
||||
const found = registry.findAvailableRunner({
|
||||
unityVersion: '2023.1.0f1',
|
||||
platform: 'StandaloneLinux64',
|
||||
});
|
||||
|
||||
expect(found).toBeUndefined();
|
||||
});
|
||||
|
||||
it('should update runner status fields', () => {
|
||||
const id = registry.registerRunner(createMockConfig());
|
||||
registry.updateRunner(id, { state: 'idle', memoryUsageMB: 2048 });
|
||||
|
||||
const runner = registry.getRunner(id);
|
||||
expect(runner!.state).toBe('idle');
|
||||
expect(runner!.memoryUsageMB).toBe(2048);
|
||||
|
||||
// ID should not be overridden by the update
|
||||
expect(runner!.id).toBe(id);
|
||||
});
|
||||
|
||||
it('should persist and load registry from disk', () => {
|
||||
const persistenceRegistry = new HotRunnerRegistry('/tmp/test');
|
||||
mockFs.existsSync.mockReturnValue(true);
|
||||
mockFs.writeFileSync.mockImplementation(() => {});
|
||||
mockFs.mkdirSync.mockImplementation(() => '' as any);
|
||||
|
||||
const id = persistenceRegistry.registerRunner(createMockConfig());
|
||||
|
||||
// Verify writeFileSync was called for persistence
|
||||
expect(mockFs.writeFileSync).toHaveBeenCalled();
|
||||
const writtenData = JSON.parse((mockFs.writeFileSync as jest.Mock).mock.calls[0][1] as string);
|
||||
expect(writtenData.runners).toBeDefined();
|
||||
expect(writtenData.runners[id]).toBeDefined();
|
||||
});
|
||||
|
||||
it('should load runners from disk on loadFromDisk', () => {
|
||||
const persistenceRegistry = new HotRunnerRegistry('/tmp/test');
|
||||
const storedData = {
|
||||
runners: {
|
||||
'hr-restored': {
|
||||
id: 'hr-restored',
|
||||
state: 'idle',
|
||||
unityVersion: '2022.3.0f1',
|
||||
platform: 'StandaloneWindows64',
|
||||
uptime: 100,
|
||||
jobsCompleted: 3,
|
||||
lastHealthCheck: new Date().toISOString(),
|
||||
memoryUsageMB: 512,
|
||||
},
|
||||
},
|
||||
configs: {
|
||||
'hr-restored': createMockConfig(),
|
||||
},
|
||||
};
|
||||
|
||||
mockFs.existsSync.mockReturnValue(true);
|
||||
mockFs.readFileSync.mockReturnValue(JSON.stringify(storedData));
|
||||
|
||||
const count = persistenceRegistry.loadFromDisk();
|
||||
expect(count).toBe(1);
|
||||
expect(persistenceRegistry.getRunner('hr-restored')).toBeDefined();
|
||||
});
|
||||
|
||||
it('should discard invalid runner entries when loading from disk', () => {
|
||||
const persistenceRegistry = new HotRunnerRegistry('/tmp/test');
|
||||
const storedData = {
|
||||
runners: {
|
||||
'hr-valid': {
|
||||
id: 'hr-valid',
|
||||
state: 'idle',
|
||||
unityVersion: '2022.3.0f1',
|
||||
platform: 'StandaloneWindows64',
|
||||
uptime: 100,
|
||||
jobsCompleted: 3,
|
||||
lastHealthCheck: new Date().toISOString(),
|
||||
memoryUsageMB: 512,
|
||||
},
|
||||
'hr-invalid': {
|
||||
// Missing required fields like state, unityVersion
|
||||
id: 'hr-invalid',
|
||||
},
|
||||
'hr-bad-state': {
|
||||
id: 'hr-bad-state',
|
||||
state: 'nonexistent-state',
|
||||
unityVersion: '2022.3.0f1',
|
||||
platform: 'StandaloneWindows64',
|
||||
uptime: 0,
|
||||
jobsCompleted: 0,
|
||||
lastHealthCheck: new Date().toISOString(),
|
||||
memoryUsageMB: 0,
|
||||
},
|
||||
},
|
||||
configs: {
|
||||
'hr-valid': createMockConfig(),
|
||||
'hr-invalid': createMockConfig(),
|
||||
'hr-bad-state': createMockConfig(),
|
||||
},
|
||||
};
|
||||
|
||||
mockFs.existsSync.mockReturnValue(true);
|
||||
mockFs.readFileSync.mockReturnValue(JSON.stringify(storedData));
|
||||
|
||||
const count = persistenceRegistry.loadFromDisk();
|
||||
expect(count).toBe(1);
|
||||
expect(persistenceRegistry.getRunner('hr-valid')).toBeDefined();
|
||||
expect(persistenceRegistry.getRunner('hr-invalid')).toBeUndefined();
|
||||
expect(persistenceRegistry.getRunner('hr-bad-state')).toBeUndefined();
|
||||
});
|
||||
|
||||
it('should handle corrupt JSON persistence file gracefully', () => {
|
||||
const persistenceRegistry = new HotRunnerRegistry('/tmp/test');
|
||||
|
||||
mockFs.existsSync.mockReturnValue(true);
|
||||
mockFs.readFileSync.mockReturnValue('{ invalid json !!!');
|
||||
|
||||
const count = persistenceRegistry.loadFromDisk();
|
||||
expect(count).toBe(0);
|
||||
expect(persistenceRegistry.size).toBe(0);
|
||||
});
|
||||
|
||||
it('should handle persistence file with invalid top-level structure', () => {
|
||||
const persistenceRegistry = new HotRunnerRegistry('/tmp/test');
|
||||
|
||||
mockFs.existsSync.mockReturnValue(true);
|
||||
mockFs.readFileSync.mockReturnValue('"just a string"');
|
||||
|
||||
const count = persistenceRegistry.loadFromDisk();
|
||||
expect(count).toBe(0);
|
||||
});
|
||||
|
||||
it('should handle persistence file with null runners', () => {
|
||||
const persistenceRegistry = new HotRunnerRegistry('/tmp/test');
|
||||
|
||||
mockFs.existsSync.mockReturnValue(true);
|
||||
mockFs.readFileSync.mockReturnValue('{"runners": null, "configs": null}');
|
||||
|
||||
const count = persistenceRegistry.loadFromDisk();
|
||||
expect(count).toBe(0);
|
||||
});
|
||||
|
||||
it('should validate and repair invalid runners', () => {
|
||||
const persistenceRegistry = new HotRunnerRegistry('/tmp/test');
|
||||
mockFs.existsSync.mockReturnValue(true);
|
||||
mockFs.writeFileSync.mockImplementation(() => {});
|
||||
mockFs.mkdirSync.mockImplementation(() => '' as any);
|
||||
|
||||
// Register a valid runner first
|
||||
const id = persistenceRegistry.registerRunner(createMockConfig());
|
||||
persistenceRegistry.updateRunner(id, { state: 'idle' });
|
||||
|
||||
// Manually corrupt the runner's state by setting an invalid state
|
||||
// (we access via the public API -- updateRunner with a cast)
|
||||
persistenceRegistry.updateRunner(id, { state: 'invalid-state' as any });
|
||||
|
||||
const repaired = persistenceRegistry.validateAndRepair();
|
||||
expect(repaired).toBe(1);
|
||||
|
||||
const runner = persistenceRegistry.getRunner(id);
|
||||
expect(runner!.state).toBe('unhealthy');
|
||||
});
|
||||
|
||||
it('should not discard configs for valid runners when loading from disk', () => {
|
||||
const persistenceRegistry = new HotRunnerRegistry('/tmp/test');
|
||||
const storedData = {
|
||||
runners: {
|
||||
'hr-valid': {
|
||||
id: 'hr-valid',
|
||||
state: 'idle',
|
||||
unityVersion: '2022.3.0f1',
|
||||
platform: 'StandaloneWindows64',
|
||||
uptime: 100,
|
||||
jobsCompleted: 3,
|
||||
lastHealthCheck: new Date().toISOString(),
|
||||
memoryUsageMB: 512,
|
||||
},
|
||||
},
|
||||
configs: {
|
||||
'hr-valid': createMockConfig(),
|
||||
},
|
||||
};
|
||||
|
||||
mockFs.existsSync.mockReturnValue(true);
|
||||
mockFs.readFileSync.mockReturnValue(JSON.stringify(storedData));
|
||||
|
||||
persistenceRegistry.loadFromDisk();
|
||||
expect(persistenceRegistry.getConfig('hr-valid')).toBeDefined();
|
||||
});
|
||||
});
|
||||
|
||||
// --- Health Monitor Tests ---
|
||||
|
||||
describe('HotRunnerHealthMonitor', () => {
|
||||
let monitor: HotRunnerHealthMonitor;
|
||||
let registry: HotRunnerRegistry;
|
||||
let transports: Map<string, HotRunnerTransport>;
|
||||
|
||||
beforeEach(() => {
|
||||
jest.clearAllMocks();
|
||||
jest.useFakeTimers();
|
||||
monitor = new HotRunnerHealthMonitor();
|
||||
registry = new HotRunnerRegistry();
|
||||
transports = new Map();
|
||||
});
|
||||
|
||||
afterEach(() => {
|
||||
monitor.stopMonitoring();
|
||||
jest.useRealTimers();
|
||||
});
|
||||
|
||||
it('should start and stop monitoring', () => {
|
||||
monitor.startMonitoring(registry, 30, transports);
|
||||
expect(monitor.isMonitoring).toBe(true);
|
||||
|
||||
monitor.stopMonitoring();
|
||||
expect(monitor.isMonitoring).toBe(false);
|
||||
});
|
||||
|
||||
it('should report healthy when transport health check passes', async () => {
|
||||
const id = registry.registerRunner(createMockConfig());
|
||||
registry.updateRunner(id, { state: 'idle' });
|
||||
|
||||
const transport = createMockTransport();
|
||||
transports.set(id, transport);
|
||||
monitor.startMonitoring(registry, 30, transports);
|
||||
|
||||
const healthy = await monitor.checkHealth(id);
|
||||
expect(healthy).toBe(true);
|
||||
expect(transport.healthCheck).toHaveBeenCalled();
|
||||
});
|
||||
|
||||
it('should mark runner as unhealthy when health check fails', async () => {
|
||||
const id = registry.registerRunner(createMockConfig());
|
||||
registry.updateRunner(id, { state: 'idle' });
|
||||
|
||||
const transport = createMockTransport({
|
||||
healthCheck: jest.fn().mockResolvedValue(false),
|
||||
});
|
||||
transports.set(id, transport);
|
||||
monitor.startMonitoring(registry, 30, transports);
|
||||
|
||||
const healthy = await monitor.checkHealth(id);
|
||||
expect(healthy).toBe(false);
|
||||
|
||||
const runner = registry.getRunner(id);
|
||||
expect(runner!.state).toBe('unhealthy');
|
||||
});
|
||||
|
||||
it('should mark runner as unhealthy when health check throws', async () => {
|
||||
const id = registry.registerRunner(createMockConfig());
|
||||
registry.updateRunner(id, { state: 'idle' });
|
||||
|
||||
const transport = createMockTransport({
|
||||
healthCheck: jest.fn().mockRejectedValue(new Error('Connection refused')),
|
||||
});
|
||||
transports.set(id, transport);
|
||||
monitor.startMonitoring(registry, 30, transports);
|
||||
|
||||
const healthy = await monitor.checkHealth(id);
|
||||
expect(healthy).toBe(false);
|
||||
});
|
||||
|
||||
it('should recycle unhealthy runner and remove from registry', async () => {
|
||||
const id = registry.registerRunner(createMockConfig());
|
||||
const transport = createMockTransport();
|
||||
transports.set(id, transport);
|
||||
monitor.startMonitoring(registry, 30, transports);
|
||||
|
||||
await monitor.recycleUnhealthyRunner(id);
|
||||
|
||||
expect(registry.getRunner(id)).toBeUndefined();
|
||||
expect(transport.disconnect).toHaveBeenCalled();
|
||||
expect(transports.has(id)).toBe(false);
|
||||
});
|
||||
|
||||
it('should recycle idle runner when max idle time exceeded', async () => {
|
||||
const id = registry.registerRunner(createMockConfig({ maxIdleTime: 60 }));
|
||||
|
||||
// Set lastHealthCheck to 120 seconds ago
|
||||
const oldDate = new Date(Date.now() - 120 * 1000).toISOString();
|
||||
registry.updateRunner(id, { state: 'idle', lastHealthCheck: oldDate });
|
||||
|
||||
const transport = createMockTransport();
|
||||
transports.set(id, transport);
|
||||
monitor.startMonitoring(registry, 30, transports);
|
||||
|
||||
await monitor.recycleIdleRunner(id, 60);
|
||||
|
||||
expect(registry.getRunner(id)).toBeUndefined();
|
||||
});
|
||||
|
||||
it('should not recycle idle runner when within max idle time', async () => {
|
||||
const id = registry.registerRunner(createMockConfig({ maxIdleTime: 3600 }));
|
||||
registry.updateRunner(id, {
|
||||
state: 'idle',
|
||||
lastHealthCheck: new Date().toISOString(),
|
||||
});
|
||||
|
||||
const transport = createMockTransport();
|
||||
transports.set(id, transport);
|
||||
monitor.startMonitoring(registry, 30, transports);
|
||||
|
||||
await monitor.recycleIdleRunner(id, 3600);
|
||||
|
||||
// Runner should still exist
|
||||
expect(registry.getRunner(id)).toBeDefined();
|
||||
});
|
||||
|
||||
it('should return false when no transport exists for runner', async () => {
|
||||
const id = registry.registerRunner(createMockConfig());
|
||||
|
||||
// Do not set any transport for this runner
|
||||
monitor.startMonitoring(registry, 30, transports);
|
||||
|
||||
const healthy = await monitor.checkHealth(id);
|
||||
expect(healthy).toBe(false);
|
||||
});
|
||||
});
|
||||
|
||||
// --- Dispatcher Tests ---
|
||||
|
||||
describe('HotRunnerDispatcher', () => {
|
||||
let registry: HotRunnerRegistry;
|
||||
let transports: Map<string, HotRunnerTransport>;
|
||||
let dispatcher: HotRunnerDispatcher;
|
||||
|
||||
beforeEach(() => {
|
||||
jest.clearAllMocks();
|
||||
registry = new HotRunnerRegistry();
|
||||
transports = new Map();
|
||||
dispatcher = new HotRunnerDispatcher(transports);
|
||||
});
|
||||
|
||||
it('should dispatch a job to an available runner', async () => {
|
||||
const id = registry.registerRunner(createMockConfig());
|
||||
registry.updateRunner(id, { state: 'idle' });
|
||||
|
||||
const transport = createMockTransport();
|
||||
transports.set(id, transport);
|
||||
|
||||
const request = createMockJobRequest();
|
||||
const result = await dispatcher.dispatchJob(request, registry, '2022.3.0f1');
|
||||
|
||||
expect(result.success).toBe(true);
|
||||
expect(result.exitCode).toBe(0);
|
||||
expect(transport.sendJob).toHaveBeenCalledWith(request);
|
||||
});
|
||||
|
||||
it('should mark runner as busy during job execution', async () => {
|
||||
const id = registry.registerRunner(createMockConfig());
|
||||
registry.updateRunner(id, { state: 'idle' });
|
||||
|
||||
const statesDuringJob: string[] = [];
|
||||
const transport = createMockTransport({
|
||||
sendJob: jest.fn().mockImplementation(async () => {
|
||||
const runner = registry.getRunner(id);
|
||||
if (runner) statesDuringJob.push(runner.state);
|
||||
|
||||
return {
|
||||
jobId: 'job-001',
|
||||
success: true,
|
||||
exitCode: 0,
|
||||
duration: 1000,
|
||||
output: 'ok',
|
||||
};
|
||||
}),
|
||||
});
|
||||
transports.set(id, transport);
|
||||
|
||||
await dispatcher.dispatchJob(createMockJobRequest(), registry, '2022.3.0f1');
|
||||
|
||||
expect(statesDuringJob).toContain('busy');
|
||||
|
||||
// After completion, should be idle again
|
||||
const runner = registry.getRunner(id);
|
||||
expect(runner!.state).toBe('idle');
|
||||
});
|
||||
|
||||
it('should increment jobsCompleted after successful dispatch', async () => {
|
||||
const id = registry.registerRunner(createMockConfig());
|
||||
registry.updateRunner(id, { state: 'idle', jobsCompleted: 5 });
|
||||
|
||||
const transport = createMockTransport();
|
||||
transports.set(id, transport);
|
||||
|
||||
await dispatcher.dispatchJob(createMockJobRequest(), registry, '2022.3.0f1');
|
||||
|
||||
const runner = registry.getRunner(id);
|
||||
expect(runner!.jobsCompleted).toBe(6);
|
||||
});
|
||||
|
||||
it('should throw when no runner is available and wait times out', async () => {
|
||||
// No runners registered at all
|
||||
const request = createMockJobRequest({ timeout: 100 });
|
||||
|
||||
await expect(dispatcher.dispatchJob(request, registry, '2022.3.0f1')).rejects.toThrow(/Timed out waiting/);
|
||||
});
|
||||
|
||||
it('should throw when runner has no transport', async () => {
|
||||
const id = registry.registerRunner(createMockConfig());
|
||||
registry.updateRunner(id, { state: 'idle' });
|
||||
|
||||
// No transport set for this runner
|
||||
|
||||
const request = createMockJobRequest();
|
||||
|
||||
await expect(dispatcher.dispatchJob(request, registry, '2022.3.0f1')).rejects.toThrow(/No transport available/);
|
||||
});
|
||||
|
||||
it('should handle job failure and return runner to idle', async () => {
|
||||
const id = registry.registerRunner(createMockConfig());
|
||||
registry.updateRunner(id, { state: 'idle' });
|
||||
|
||||
const transport = createMockTransport({
|
||||
sendJob: jest.fn().mockRejectedValue(new Error('Unity crashed')),
|
||||
});
|
||||
transports.set(id, transport);
|
||||
|
||||
await expect(dispatcher.dispatchJob(createMockJobRequest(), registry, '2022.3.0f1')).rejects.toThrow(
|
||||
'Unity crashed',
|
||||
);
|
||||
|
||||
// Runner should be back to idle despite failure
|
||||
const runner = registry.getRunner(id);
|
||||
expect(runner!.state).toBe('idle');
|
||||
});
|
||||
|
||||
it('should handle job timeout', async () => {
|
||||
const id = registry.registerRunner(createMockConfig());
|
||||
registry.updateRunner(id, { state: 'idle' });
|
||||
|
||||
const transport = createMockTransport({
|
||||
sendJob: jest.fn().mockImplementation(
|
||||
() => new Promise((resolve) => setTimeout(resolve, 60000)), // never resolves within timeout
|
||||
),
|
||||
});
|
||||
transports.set(id, transport);
|
||||
|
||||
const request = createMockJobRequest({ timeout: 50 });
|
||||
|
||||
await expect(dispatcher.dispatchJob(request, registry, '2022.3.0f1')).rejects.toThrow(/timed out/);
|
||||
});
|
||||
|
||||
it('should disconnect transport on job timeout', async () => {
|
||||
const id = registry.registerRunner(createMockConfig());
|
||||
registry.updateRunner(id, { state: 'idle' });
|
||||
|
||||
const transport = createMockTransport({
|
||||
sendJob: jest.fn().mockImplementation(
|
||||
() => new Promise((resolve) => setTimeout(resolve, 60000)), // never resolves within timeout
|
||||
),
|
||||
});
|
||||
transports.set(id, transport);
|
||||
|
||||
const request = createMockJobRequest({ timeout: 50 });
|
||||
|
||||
await expect(dispatcher.dispatchJob(request, registry, '2022.3.0f1')).rejects.toThrow(/timed out/);
|
||||
|
||||
// Transport should have been disconnected to clean up orphaned connection
|
||||
expect(transport.disconnect).toHaveBeenCalled();
|
||||
});
|
||||
|
||||
it('should call output callback with job output', async () => {
|
||||
const id = registry.registerRunner(createMockConfig());
|
||||
registry.updateRunner(id, { state: 'idle' });
|
||||
|
||||
const transport = createMockTransport();
|
||||
transports.set(id, transport);
|
||||
|
||||
const outputCallback = jest.fn();
|
||||
await dispatcher.dispatchJob(createMockJobRequest(), registry, '2022.3.0f1', outputCallback);
|
||||
|
||||
expect(outputCallback).toHaveBeenCalledWith('Build succeeded');
|
||||
});
|
||||
|
||||
it('should wait for runner to become available', async () => {
|
||||
const id = registry.registerRunner(createMockConfig());
|
||||
|
||||
// Runner starts in 'starting' state, not idle
|
||||
|
||||
const transport = createMockTransport();
|
||||
transports.set(id, transport);
|
||||
|
||||
// Simulate runner becoming idle after a short delay
|
||||
setTimeout(() => {
|
||||
registry.updateRunner(id, { state: 'idle' });
|
||||
}, 50);
|
||||
|
||||
const request = createMockJobRequest({ timeout: 5000 });
|
||||
const result = await dispatcher.dispatchJob(request, registry, '2022.3.0f1');
|
||||
|
||||
expect(result.success).toBe(true);
|
||||
});
|
||||
});
|
||||
|
||||
// --- Service Integration Tests ---
|
||||
|
||||
describe('HotRunnerService', () => {
|
||||
let service: HotRunnerService;
|
||||
|
||||
beforeEach(() => {
|
||||
jest.clearAllMocks();
|
||||
mockFs.existsSync.mockReturnValue(false);
|
||||
service = new HotRunnerService();
|
||||
});
|
||||
|
||||
afterEach(async () => {
|
||||
await service.shutdown();
|
||||
});
|
||||
|
||||
it('should initialize and shut down cleanly', async () => {
|
||||
const config = createMockConfig();
|
||||
await service.initialize(config);
|
||||
|
||||
const status = service.getStatus();
|
||||
expect(status).toEqual([]);
|
||||
|
||||
await service.shutdown();
|
||||
});
|
||||
|
||||
it('should register a runner with transport', async () => {
|
||||
await service.initialize(createMockConfig());
|
||||
|
||||
const transport = createMockTransport();
|
||||
const id = service.registerRunner(createMockConfig(), transport);
|
||||
|
||||
expect(id).toMatch(/^hr-/);
|
||||
expect(service.getStatus()).toHaveLength(1);
|
||||
});
|
||||
|
||||
it('should disconnect all transports on shutdown', async () => {
|
||||
await service.initialize(createMockConfig());
|
||||
|
||||
const transport1 = createMockTransport();
|
||||
const transport2 = createMockTransport();
|
||||
service.registerRunner(createMockConfig(), transport1);
|
||||
service.registerRunner(createMockConfig(), transport2);
|
||||
|
||||
await service.shutdown();
|
||||
|
||||
expect(transport1.disconnect).toHaveBeenCalled();
|
||||
expect(transport2.disconnect).toHaveBeenCalled();
|
||||
});
|
||||
|
||||
it('should expose the underlying registry', async () => {
|
||||
await service.initialize(createMockConfig());
|
||||
const registry = service.getRegistry();
|
||||
|
||||
expect(registry).toBeInstanceOf(HotRunnerRegistry);
|
||||
});
|
||||
});
|
||||
11
src/model/orchestrator/services/hot-runner/index.ts
Normal file
11
src/model/orchestrator/services/hot-runner/index.ts
Normal file
@@ -0,0 +1,11 @@
|
||||
export { HotRunnerService } from './hot-runner-service';
|
||||
export { HotRunnerRegistry } from './hot-runner-registry';
|
||||
export { HotRunnerHealthMonitor } from './hot-runner-health-monitor';
|
||||
export { HotRunnerDispatcher } from './hot-runner-dispatcher';
|
||||
export type {
|
||||
HotRunnerConfig,
|
||||
HotRunnerStatus,
|
||||
HotRunnerJobRequest,
|
||||
HotRunnerJobResult,
|
||||
HotRunnerTransport,
|
||||
} from './hot-runner-types';
|
||||
Reference in New Issue
Block a user