feat(ql3): compose cluster copilot diagnosis

This commit is contained in:
whyour
2026-08-15 21:59:59 +08:00
parent c3bd6d40bb
commit 5fc70010f2
27 changed files with 2378 additions and 24 deletions
@@ -35,6 +35,11 @@
"require": "./dist/application-runtime/aiProductionApplication.js",
"default": "./dist/application-runtime/aiProductionApplication.js"
},
"./copilot-production": {
"types": "./dist/application-runtime/copilot/failureDiagnosisComposition.d.ts",
"require": "./dist/application-runtime/copilot/failureDiagnosisComposition.js",
"default": "./dist/application-runtime/copilot/failureDiagnosisComposition.js"
},
"./failure-diagnosis-output-keyring": {
"types": "./dist/copilot/failure-diagnosis/outputProjectedKeyring.d.ts",
"require": "./dist/copilot/failure-diagnosis/outputProjectedKeyring.js",
@@ -1,4 +1,7 @@
import type { ModelGatewayProfileAudit } from '@qinglong/ai/profile';
import type { DurableModelInvocationCoordinator } from '@qinglong/ai/durable-model-invocation';
import { CopilotFailureDiagnosisModelCompletionCoordinator } from '@qinglong/ai/failure-diagnosis-model-execution';
import type { CopilotFailureDiagnosisApplicationService } from '@qinglong/ai/failure-diagnosis-application';
import { BoundModelProviderCredentialProvider } from '@qinglong/ai/provider-credential';
import { PostgresModelProviderCredentialReader } from '@qinglong/ai/postgres-model-provider-credential-storage';
import { loadProjectedModelGatewayProviderAuthority } from '@qinglong/ai/projected-model-gateway-authority';
@@ -22,12 +25,19 @@ import {
startProductionClusterControlApplication,
type ProductionClusterControlApplicationOptions,
} from './productionApplication';
import {
createProductionClusterCopilotFailureDiagnosis,
prepareProductionClusterCopilotFailureDiagnosisProjection,
type ClusterCopilotFailureDiagnosisProjection,
type CreateProductionClusterCopilotFailureDiagnosisOptions,
} from './copilot/failureDiagnosisComposition';
export interface EnabledProductionClusterAiConfig {
readonly enabled: true;
readonly providerAuthorityFile: string;
readonly secretRootDirectory: string;
readonly promptOutputKeyringRootDirectory?: string;
readonly copilot?: Readonly<ClusterCopilotFailureDiagnosisProjection>;
readonly maxConcurrent: number;
readonly recoveryLimit: number;
readonly databaseMaxConnections: number;
@@ -41,8 +51,18 @@ export interface ProductionClusterAiControlApplicationOptions {
) => void | Promise<void>;
readonly startControl?: typeof startProductionClusterControlApplication;
readonly bootstrapPrompt?: typeof bootstrapPostgresPluginPackagePromptApplication;
readonly createCopilot?: (
options: CreateProductionClusterCopilotFailureDiagnosisOptions,
) => Promise<Readonly<CopilotFailureDiagnosisApplicationService>>;
readonly openAiDatabase?: ReturnType<typeof createPostgresDatabaseOpener>;
}
export type ProductionClusterAiControlApplicationResult = Extract<
ClusterControlApplicationResult,
{ readonly status: 'active' }
> &
Readonly<{ copilot?: Readonly<CopilotFailureDiagnosisApplicationService> }>;
export class ProductionClusterAiConfigError extends TypeError {
readonly code = 'QL3_CLUSTER_AI_CONFIG_INVALID';
@@ -121,6 +141,11 @@ export function loadProductionClusterAiConfig(
'QL3_CLUSTER_AI_PROMPT_OUTPUT_ENABLED',
false,
);
const copilotEnabled = booleanValue(
environment,
'QL3_CLUSTER_AI_COPILOT_ENABLED',
false,
);
return Object.freeze({
enabled: true,
providerAuthorityFile: requiredPath(
@@ -139,6 +164,28 @@ export function loadProductionClusterAiConfig(
),
}
: {}),
...(copilotEnabled
? {
copilot: Object.freeze({
configFile: requiredPath(
environment,
'QL3_CLUSTER_AI_COPILOT_CONFIG_FILE',
),
invocationKeyringRootDirectory: requiredPath(
environment,
'QL3_CLUSTER_AI_COPILOT_INVOCATION_KEYRING_ROOT',
),
resultKeyringRootDirectory: requiredPath(
environment,
'QL3_CLUSTER_AI_COPILOT_RESULT_KEYRING_ROOT',
),
outputKeyringRootDirectory: requiredPath(
environment,
'QL3_CLUSTER_AI_COPILOT_OUTPUT_KEYRING_ROOT',
),
}),
}
: {}),
maxConcurrent: boundedInteger(
environment,
'QL3_CLUSTER_AI_MAX_CONCURRENT',
@@ -187,7 +234,7 @@ function aiDatabaseOpener(
*/
export async function startProductionClusterAiControlApplication(
options: ProductionClusterAiControlApplicationOptions,
): Promise<ClusterControlApplicationResult> {
): Promise<ProductionClusterAiControlApplicationResult> {
if (
!options ||
typeof options !== 'object' ||
@@ -200,9 +247,26 @@ export async function startProductionClusterAiControlApplication(
options.startControl ?? startProductionClusterControlApplication;
const bootstrapPrompt =
options.bootstrapPrompt ?? bootstrapPostgresPluginPackagePromptApplication;
if (typeof startControl !== 'function' || typeof bootstrapPrompt !== 'function') {
const createCopilot =
options.createCopilot ?? createProductionClusterCopilotFailureDiagnosis;
if (
typeof startControl !== 'function' ||
typeof bootstrapPrompt !== 'function' ||
typeof createCopilot !== 'function' ||
(options.openAiDatabase !== undefined &&
typeof options.openAiDatabase !== 'function')
) {
throw new TypeError('Production Cluster AI application factories are invalid');
}
const copilotArtifactStore = options.control.workerIngress?.artifactStore;
if (
options.ai.copilot !== undefined &&
typeof copilotArtifactStore?.readLogRange !== 'function'
) {
throw new TypeError(
'Cluster Copilot requires the bounded Worker log Artifact read capability',
);
}
const secretMaterial =
await createProjectedModelProviderSecretMaterialProvider({
rootDirectory: options.ai.secretRootDirectory,
@@ -213,6 +277,12 @@ export async function startProductionClusterAiControlApplication(
: await createPluginPackagePromptOutputProjectedKeyring({
rootDirectory: options.ai.promptOutputKeyringRootDirectory,
});
const preparedCopilot =
options.ai.copilot === undefined
? undefined
: await prepareProductionClusterCopilotFailureDiagnosisProjection(
options.ai.copilot,
);
let aiDatabase:
| Awaited<ReturnType<ReturnType<typeof createPostgresDatabaseOpener>>>
| undefined;
@@ -230,6 +300,12 @@ export async function startProductionClusterAiControlApplication(
| BootstrapPostgresPluginPackagePromptApplicationResult
| undefined;
let controlApplication: ClusterControlApplicationResult | undefined;
let copilotApplication:
| Readonly<CopilotFailureDiagnosisApplicationService>
| undefined;
let copilotSuccessfulCompletion:
| CopilotFailureDiagnosisModelCompletionCoordinator
| undefined;
let stopPromise: Promise<ClusterControlStopResult> | undefined;
let promptOutputPolicy: ProjectPolicyEngine | undefined;
const promptOutputReadAuthorizer = Object.freeze({
@@ -271,11 +347,9 @@ export async function startProductionClusterAiControlApplication(
return stopPromise;
};
try {
const openDatabase = aiDatabaseOpener(
options.control.config,
options.ai,
onAiUnavailable,
);
const openDatabase =
options.openAiDatabase ??
aiDatabaseOpener(options.control.config, options.ai, onAiUnavailable);
promptApplication = await bootstrapPrompt({
enabled: true,
async openDatabase() {
@@ -311,10 +385,41 @@ export async function startProductionClusterAiControlApplication(
promptOutputKeys,
promptOutputRead: { authorizer: promptOutputReadAuthorizer },
}),
...(preparedCopilot === undefined
? {}
: {
createAdditionalSuccessfulCompletion(
coordinator: DurableModelInvocationCoordinator,
) {
if (copilotSuccessfulCompletion) {
throw new Error(
'Cluster Copilot completion was created more than once',
);
}
copilotSuccessfulCompletion =
new CopilotFailureDiagnosisModelCompletionCoordinator({
coordinator,
keys: preparedCopilot.outputKeys,
});
return copilotSuccessfulCompletion;
},
}),
});
if (promptApplication.status !== 'active') {
throw new Error('Cluster AI Prompt application did not activate');
}
if (preparedCopilot !== undefined) {
if (!copilotSuccessfulCompletion || !aiDatabase || !copilotArtifactStore) {
throw new Error('Cluster Copilot shared authorities did not activate');
}
copilotApplication = await createCopilot({
pool: aiDatabase.pool,
gateway: promptApplication.capability,
prepared: preparedCopilot,
successfulCompletion: copilotSuccessfulCompletion,
artifactStore: copilotArtifactStore,
});
}
controlApplication = await startControl({
...options.control,
promptCatalog: {
@@ -356,6 +461,9 @@ export async function startProductionClusterAiControlApplication(
? 'unavailable'
: activeControl.availabilityStatus();
},
...(copilotApplication === undefined
? {}
: { copilot: copilotApplication }),
stop,
});
} catch (error) {
@@ -0,0 +1,466 @@
import { Buffer } from 'node:buffer';
import { basename, dirname } from 'node:path';
import {
CopilotFailureDiagnosisApplicationService,
type CopilotFailureDiagnosisApplicationDependencies,
} from '@qinglong/ai/failure-diagnosis-application';
import type { PrepareCopilotFailureDiagnosisModelIntent } from '@qinglong/ai/failure-diagnosis-execution-admission';
import {
CopilotFailureDiagnosisModelCompletionCoordinator,
executeCopilotFailureDiagnosisModel,
type CopilotFailureDiagnosisToolResultReader,
} from '@qinglong/ai/failure-diagnosis-model-execution';
import {
executeCopilotFailureDiagnosisTool,
restoreCopilotFailureDiagnosisTrustedToolAuthority,
} from '@qinglong/ai/failure-diagnosis-tool-execution';
import { PostgresCopilotFailureDiagnosisAdmissionRepository } from '@qinglong/ai/postgres-failure-diagnosis-admission-storage';
import { PostgresCopilotFailureDiagnosisModelRepository } from '@qinglong/ai/postgres-failure-diagnosis-model-execution-storage';
import { PostgresCopilotFailureDiagnosisToolUnlockRepository } from '@qinglong/ai/postgres-failure-diagnosis-tool-execution-storage';
import type { ActiveModelGatewayCapability } from '@qinglong/ai/profile';
import {
PostgresProjectPolicyRepository,
PostgresProjectToolDefinitionSnapshotRepository,
PostgresRunAttemptLogRetentionClaimRepository,
PostgresRunRepository,
PostgresStepRunRepository,
PostgresToolExecutionCompletionRepository,
PostgresToolExecutionFailureCompletionRepository,
PostgresToolExecutionStartBarrierRepository,
PostgresToolInvocationArtifactRepository,
PostgresToolResultKeyCatalogReader,
PostgresToolResultRekeyReader,
type QingLongPostgresPool,
} from '@qinglong/cluster-postgres/runtime';
import {
BuiltInRunLogExcerptToolAdapter,
} from '@qinglong/runtime-core/builtin-run-log-excerpt-tool';
import type { RunAttemptLogReadPort } from '@qinglong/runtime-core/builtin-run-log-excerpt-projection';
import { ProjectPolicyEngine } from '@qinglong/runtime-core/project-policy';
import { RunAttemptLogReadService } from '@qinglong/runtime-core/run-attempt-log-read';
import { openTrustedToolSuccessCompletion } from '@qinglong/runtime-core/trusted-tool-completion';
import { TrustedToolExecutionAdapterRegistry } from '@qinglong/runtime-core/trusted-tool-execution';
import type { ClusterRemoteWorkerArtifactStore } from '../../remote-execution/remoteWorkerCompletionService';
import { PrivateProjectedFileReader } from '../../security/privateProjectedFile';
import { createClusterToolInvocationProjectedKeyring } from '../../trusted-tool/key-management/toolInvocationProjectedKeyring';
import { createClusterToolResultProjectedKeyring } from '../../trusted-tool/key-management/toolResultProjectedKeyring';
import { createClusterCopilotFailureDiagnosisOutputProjectedKeyring } from '../../copilot/failure-diagnosis/outputProjectedKeyring';
export const CLUSTER_COPILOT_FAILURE_DIAGNOSIS_CONFIG_SCHEMA =
'qinglong/cluster-copilot-failure-diagnosis-config@v1' as const;
export const MAX_CLUSTER_COPILOT_FAILURE_DIAGNOSIS_CONFIG_BYTES = 16 * 1024;
const IDENTITY = /^[A-Za-z0-9][A-Za-z0-9._:-]{0,127}$/;
const MODEL_BOUNDARIES = ['on_device', 'external'] as const;
const RESPONSE_LANGUAGES = ['en', 'zh-CN'] as const;
const EGRESS_SCHEMA = 'qinglong/copilot-model-egress-policy@v1' as const;
export interface ClusterCopilotFailureDiagnosisConfig {
readonly schema: typeof CLUSTER_COPILOT_FAILURE_DIAGNOSIS_CONFIG_SCHEMA;
readonly provider: string;
readonly model: string;
readonly modelBoundary: 'on_device' | 'external';
readonly responseLanguage: 'en' | 'zh-CN';
readonly maxOutputTokens: number;
readonly executionTimeoutMs: number;
readonly egressPolicy: PrepareCopilotFailureDiagnosisModelIntent['egressPolicy'];
}
export interface ClusterCopilotFailureDiagnosisProjection {
readonly configFile: string;
readonly invocationKeyringRootDirectory: string;
readonly resultKeyringRootDirectory: string;
readonly outputKeyringRootDirectory: string;
}
export interface CreateProductionClusterCopilotFailureDiagnosisOptions {
readonly pool: QingLongPostgresPool;
readonly gateway: ActiveModelGatewayCapability;
readonly prepared: PreparedClusterCopilotFailureDiagnosisProjection;
readonly successfulCompletion: CopilotFailureDiagnosisModelCompletionCoordinator;
readonly artifactStore: ClusterRemoteWorkerArtifactStore;
}
export interface PreparedClusterCopilotFailureDiagnosisProjection {
readonly config: Readonly<ClusterCopilotFailureDiagnosisConfig>;
readonly invocationKeys: Awaited<
ReturnType<typeof createClusterToolInvocationProjectedKeyring>
>;
readonly resultKeys: Awaited<
ReturnType<typeof createClusterToolResultProjectedKeyring>
>;
readonly outputKeys: Awaited<
ReturnType<typeof createClusterCopilotFailureDiagnosisOutputProjectedKeyring>
>;
}
export class ClusterCopilotFailureDiagnosisCompositionError extends Error {
readonly code = 'QL3_CLUSTER_COPILOT_FAILURE_DIAGNOSIS_COMPOSITION_INVALID';
constructor(message: string, options?: ErrorOptions) {
super(`Cluster Copilot failure diagnosis composition is invalid: ${message}`, options);
this.name = 'ClusterCopilotFailureDiagnosisCompositionError';
}
}
function invalid(message: string, cause?: unknown): never {
throw new ClusterCopilotFailureDiagnosisCompositionError(message, {
cause: cause instanceof Error ? cause : undefined,
});
}
function record(value: unknown, label: string): Record<string, unknown> {
if (!value || typeof value !== 'object' || Array.isArray(value)) {
return invalid(`${label} must be an object`);
}
return value as Record<string, unknown>;
}
function exactKeys(
value: Readonly<Record<string, unknown>>,
expected: readonly string[],
label: string,
): void {
const actual = Object.keys(value).sort();
const canonical = [...expected].sort();
if (
actual.length !== canonical.length ||
actual.some((key, index) => key !== canonical[index])
) {
return invalid(`${label} shape is invalid`);
}
}
function identity(value: unknown, label: string): string {
if (typeof value !== 'string' || !IDENTITY.test(value)) {
return invalid(`${label} is invalid`);
}
return value;
}
function integer(
value: unknown,
minimum: number,
maximum: number,
label: string,
): number {
if (
typeof value !== 'number' ||
!Number.isSafeInteger(value) ||
value < minimum ||
value > maximum
) {
return invalid(`${label} is invalid`);
}
return value;
}
export function normalizeClusterCopilotFailureDiagnosisConfig(
value: unknown,
): Readonly<ClusterCopilotFailureDiagnosisConfig> {
const candidate = record(value, 'configuration');
exactKeys(
candidate,
[
'egressPolicy',
'executionTimeoutMs',
'maxOutputTokens',
'model',
'modelBoundary',
'provider',
'responseLanguage',
'schema',
],
'configuration',
);
if (candidate.schema !== CLUSTER_COPILOT_FAILURE_DIAGNOSIS_CONFIG_SCHEMA) {
return invalid('schema is invalid');
}
if (!MODEL_BOUNDARIES.includes(candidate.modelBoundary as never)) {
return invalid('model boundary is invalid');
}
if (!RESPONSE_LANGUAGES.includes(candidate.responseLanguage as never)) {
return invalid('response language is invalid');
}
const egress = record(candidate.egressPolicy, 'egress policy');
exactKeys(
egress,
[
'maxInputBytes',
'maxOutputTokens',
'potentiallySensitiveDataBoundaries',
'revision',
'schema',
],
'egress policy',
);
if (egress.schema !== EGRESS_SCHEMA) return invalid('egress schema is invalid');
const selected = egress.potentiallySensitiveDataBoundaries;
if (
!Array.isArray(selected) ||
selected.length < 1 ||
selected.length > MODEL_BOUNDARIES.length ||
selected.some((entry) => !MODEL_BOUNDARIES.includes(entry as never)) ||
new Set(selected).size !== selected.length ||
MODEL_BOUNDARIES.filter((entry) => selected.includes(entry)).some(
(entry, index) => entry !== selected[index],
) ||
!selected.includes(candidate.modelBoundary)
) {
return invalid('egress model boundaries are invalid');
}
const egressMaxOutputTokens = integer(
egress.maxOutputTokens,
1,
4_096,
'egress max output tokens',
);
const maxOutputTokens = integer(
candidate.maxOutputTokens,
1,
egressMaxOutputTokens,
'max output tokens',
);
return Object.freeze({
schema: CLUSTER_COPILOT_FAILURE_DIAGNOSIS_CONFIG_SCHEMA,
provider: identity(candidate.provider, 'provider'),
model: identity(candidate.model, 'model'),
modelBoundary: candidate.modelBoundary as 'on_device' | 'external',
responseLanguage: candidate.responseLanguage as 'en' | 'zh-CN',
maxOutputTokens,
executionTimeoutMs: integer(
candidate.executionTimeoutMs,
1,
5 * 60_000,
'execution timeout',
),
egressPolicy: Object.freeze({
schema: EGRESS_SCHEMA,
revision: identity(egress.revision, 'egress revision'),
potentiallySensitiveDataBoundaries: Object.freeze([...selected]) as (
| 'on_device'
| 'external'
)[],
maxInputBytes: integer(
egress.maxInputBytes,
1,
64 * 1024,
'egress max input bytes',
),
maxOutputTokens: egressMaxOutputTokens,
}),
});
}
export function canonicalClusterCopilotFailureDiagnosisConfig(
value: unknown,
): Buffer {
return Buffer.from(
`${JSON.stringify(normalizeClusterCopilotFailureDiagnosisConfig(value))}\n`,
'utf8',
);
}
export async function loadClusterCopilotFailureDiagnosisConfig(
configFile: string,
): Promise<Readonly<ClusterCopilotFailureDiagnosisConfig>> {
let bytes: Buffer | undefined;
let canonical: Buffer | undefined;
try {
const reader = new PrivateProjectedFileReader({
rootDirectory: dirname(configFile),
minimumBytes: 1,
maximumBytes: MAX_CLUSTER_COPILOT_FAILURE_DIAGNOSIS_CONFIG_BYTES,
access: 'read_only_keyring',
});
bytes = await reader.read(basename(configFile));
const parsed = JSON.parse(bytes.toString('utf8')) as unknown;
const config = normalizeClusterCopilotFailureDiagnosisConfig(parsed);
canonical = canonicalClusterCopilotFailureDiagnosisConfig(config);
if (!canonical.equals(bytes)) return invalid('file is not canonical');
return config;
} catch (cause) {
return cause instanceof ClusterCopilotFailureDiagnosisCompositionError
? invalid(cause.message, cause)
: invalid('projected configuration is unavailable', cause);
} finally {
bytes?.fill(0);
canonical?.fill(0);
}
}
function modelIntent(
config: Readonly<ClusterCopilotFailureDiagnosisConfig>,
): Readonly<PrepareCopilotFailureDiagnosisModelIntent> {
return Object.freeze({
provider: config.provider,
model: config.model,
modelBoundary: config.modelBoundary,
responseLanguage: config.responseLanguage,
maxOutputTokens: config.maxOutputTokens,
egressPolicy: config.egressPolicy,
});
}
export async function prepareProductionClusterCopilotFailureDiagnosisProjection(
projection: ClusterCopilotFailureDiagnosisProjection,
): Promise<Readonly<PreparedClusterCopilotFailureDiagnosisProjection>> {
if (!projection || typeof projection !== 'object' || Array.isArray(projection)) {
return invalid('projection is invalid');
}
const [config, invocationKeys, resultKeys, outputKeys] = await Promise.all([
loadClusterCopilotFailureDiagnosisConfig(projection.configFile),
createClusterToolInvocationProjectedKeyring({
rootDirectory: projection.invocationKeyringRootDirectory,
}),
createClusterToolResultProjectedKeyring({
rootDirectory: projection.resultKeyringRootDirectory,
}),
createClusterCopilotFailureDiagnosisOutputProjectedKeyring({
rootDirectory: projection.outputKeyringRootDirectory,
}),
]);
return Object.freeze({ config, invocationKeys, resultKeys, outputKeys });
}
export async function createProductionClusterCopilotFailureDiagnosis(
options: CreateProductionClusterCopilotFailureDiagnosisOptions,
): Promise<Readonly<CopilotFailureDiagnosisApplicationService>> {
if (
!options ||
typeof options !== 'object' ||
Array.isArray(options) ||
typeof options.pool?.query !== 'function' ||
typeof options.pool?.connect !== 'function' ||
typeof options.gateway?.generate !== 'function' ||
typeof options.gateway?.supportsSuccessfulCompletionSink !== 'function' ||
typeof options.successfulCompletion?.begin !== 'function' ||
typeof options.successfulCompletion?.record !== 'function' ||
!options.prepared ||
typeof options.artifactStore?.readLogRange !== 'function'
) {
return invalid('dependencies are unavailable');
}
if (
!options.gateway.supportsSuccessfulCompletionSink(
options.successfulCompletion,
)
) {
return invalid('shared Model completion authority is unavailable');
}
const { config, invocationKeys, resultKeys } = options.prepared;
const admissions = new PostgresCopilotFailureDiagnosisAdmissionRepository(
options.pool,
);
const snapshots = new PostgresProjectToolDefinitionSnapshotRepository(
options.pool,
);
const runs = new PostgresRunRepository(options.pool);
const artifacts = new PostgresToolInvocationArtifactRepository(options.pool);
const stepRuns = new PostgresStepRunRepository(options.pool);
const barriers = new PostgresToolExecutionStartBarrierRepository(options.pool);
const completions = new PostgresToolExecutionCompletionRepository(options.pool);
const failureCompletions =
new PostgresToolExecutionFailureCompletionRepository(options.pool);
const resultKeyCatalog = new PostgresToolResultKeyCatalogReader(options.pool);
const resultRekeys = new PostgresToolResultRekeyReader(options.pool);
const unlocks = new PostgresCopilotFailureDiagnosisToolUnlockRepository(
options.pool,
);
const models = new PostgresCopilotFailureDiagnosisModelRepository(options.pool);
const logReader = new RunAttemptLogReadService(
runs,
Object.freeze({
read: options.artifactStore.readLogRange.bind(options.artifactStore),
}),
{
executorType: 'remote_worker',
artifactIdPattern: /^wlog-[a-f0-9]{30}$/,
maximumReadBytes: 256 * 1024,
activeMissingIsPending: true,
},
new PostgresRunAttemptLogRetentionClaimRepository(options.pool),
);
const logs: RunAttemptLogReadPort = Object.freeze({
read: logReader.read.bind(logReader),
});
const successfulCompletion = options.successfulCompletion;
const toolResults: CopilotFailureDiagnosisToolResultReader = Object.freeze({
async open(requestId: string, startId: string) {
const plan = await admissions.findPlanByRequestId(requestId);
if (!plan) return invalid('diagnosis plan is unavailable');
const snapshot = await snapshots.findCurrent(plan.projectId);
if (!snapshot) return invalid('Tool snapshot is unavailable');
const authority = restoreCopilotFailureDiagnosisTrustedToolAuthority(
plan,
snapshot.snapshot,
);
const definitions = authority.bindings.definitionRegistry();
const adapters = new TrustedToolExecutionAdapterRegistry(
authority.bindings,
[
new BuiltInRunLogExcerptToolAdapter(
authority.binding,
'cluster-control',
definitions,
logs,
),
],
);
return openTrustedToolSuccessCompletion(startId, {
completions,
barriers,
resultKeyCatalog,
resultRekeys,
resultKeys,
adapters,
});
},
});
const policy = new ProjectPolicyEngine(
new PostgresProjectPolicyRepository(options.pool),
);
const tool = Object.freeze({
admissions,
snapshots,
artifacts,
invocationKeys,
resultKeys,
stepRuns,
runs,
barriers,
completions,
failureCompletions,
resultKeyCatalog,
resultRekeys,
logs,
unlocks,
});
const model = Object.freeze({
admissions,
unlocks,
toolResults,
modelInvocations: models,
outputs: models,
gateway: options.gateway,
successfulCompletion,
finalizations: models,
});
const dependencies: CopilotFailureDiagnosisApplicationDependencies = {
admissions,
snapshots,
runs,
artifacts,
invocationKeys,
authorizer: policy,
tool,
model,
executeTool: executeCopilotFailureDiagnosisTool,
executeModel: executeCopilotFailureDiagnosisModel,
modelIntent: modelIntent(config),
executionTimeoutMs: config.executionTimeoutMs,
};
return new CopilotFailureDiagnosisApplicationService(dependencies);
}
@@ -13,6 +13,18 @@ const {
canonicalPluginPackagePromptOutputKeyringManifest,
PLUGIN_PACKAGE_PROMPT_OUTPUT_KEYRING_MANIFEST_SCHEMA,
} = require('@qinglong/ai/plugin-package-prompt-output-keyring-manifest');
const {
canonicalClusterToolInvocationKeyringManifest,
CLUSTER_TOOL_INVOCATION_KEYRING_MANIFEST_SCHEMA,
} = require('@qinglong/cluster-control/trusted-tool-invocation-keyring');
const {
canonicalClusterToolResultKeyringManifest,
CLUSTER_TOOL_RESULT_KEYRING_MANIFEST_SCHEMA,
} = require('@qinglong/cluster-control/trusted-tool-result-keyring');
const {
canonicalClusterCopilotFailureDiagnosisOutputKeyringManifest,
CLUSTER_COPILOT_FAILURE_DIAGNOSIS_OUTPUT_KEYRING_MANIFEST_SCHEMA,
} = require('@qinglong/cluster-control/failure-diagnosis-output-keyring');
function enabledEnvironment(overrides = {}) {
return {
@@ -74,6 +86,161 @@ test('AI config is fail-closed and bounded behind the explicit process flag', ()
),
/QL3_CLUSTER_AI_PROMPT_OUTPUT_KEYRING_ROOT is invalid/,
);
assert.throws(
() =>
loadProductionClusterAiConfig(
enabledEnvironment({ QL3_CLUSTER_AI_COPILOT_ENABLED: 'true' }),
),
/QL3_CLUSTER_AI_COPILOT_CONFIG_FILE is invalid/,
);
assert.deepEqual(
loadProductionClusterAiConfig(
enabledEnvironment({
QL3_CLUSTER_AI_COPILOT_ENABLED: 'true',
QL3_CLUSTER_AI_COPILOT_CONFIG_FILE: '/run/ql3/copilot/config.json',
QL3_CLUSTER_AI_COPILOT_INVOCATION_KEYRING_ROOT: '/run/ql3/invocation',
QL3_CLUSTER_AI_COPILOT_RESULT_KEYRING_ROOT: '/run/ql3/result',
QL3_CLUSTER_AI_COPILOT_OUTPUT_KEYRING_ROOT: '/run/ql3/output',
}),
).copilot,
{
configFile: '/run/ql3/copilot/config.json',
invocationKeyringRootDirectory: '/run/ql3/invocation',
resultKeyringRootDirectory: '/run/ql3/result',
outputKeyringRootDirectory: '/run/ql3/output',
},
);
});
async function projectedFile(root, name, bytes) {
await writeFile(join(root, name), bytes, { mode: 0o440 });
await chmod(join(root, name), 0o440);
}
test('Copilot composition is explicit, shares the Prompt gateway and exposes no route', async () => {
const secretRoot = await mkdtemp(join(tmpdir(), 'ql3-cluster-ai-secret-'));
const configRoot = await mkdtemp(join(tmpdir(), 'ql3-copilot-config-'));
const invocationRoot = await mkdtemp(join(tmpdir(), 'ql3-copilot-invocation-'));
const resultRoot = await mkdtemp(join(tmpdir(), 'ql3-copilot-result-'));
const outputRoot = await mkdtemp(join(tmpdir(), 'ql3-copilot-output-'));
const key = Buffer.alloc(32, 0x55).toString('base64url');
const config = Buffer.from(`${JSON.stringify({
schema: 'qinglong/cluster-copilot-failure-diagnosis-config@v1',
provider: 'provider-primary',
model: 'model-diagnosis',
modelBoundary: 'external',
responseLanguage: 'zh-CN',
maxOutputTokens: 512,
executionTimeoutMs: 60_000,
egressPolicy: {
schema: 'qinglong/copilot-model-egress-policy@v1',
revision: 'cluster-copilot-v1',
potentiallySensitiveDataBoundaries: ['external'],
maxInputBytes: 64 * 1024,
maxOutputTokens: 1024,
},
})}\n`);
const invocation = canonicalClusterToolInvocationKeyringManifest({
schema: CLUSTER_TOOL_INVOCATION_KEYRING_MANIFEST_SCHEMA,
activeKeyId: 'invocation-key-1',
keys: { 'invocation-key-1': key },
});
const result = canonicalClusterToolResultKeyringManifest({
schema: CLUSTER_TOOL_RESULT_KEYRING_MANIFEST_SCHEMA,
keys: { 'result-key-1': key },
});
const output = canonicalClusterCopilotFailureDiagnosisOutputKeyringManifest({
schema: CLUSTER_COPILOT_FAILURE_DIAGNOSIS_OUTPUT_KEYRING_MANIFEST_SCHEMA,
activeKeyId: 'output-key-1',
keys: { 'output-key-1': key },
});
const gateway = {
generate() {},
supportsSuccessfulCompletionSink(sink) {
return sink === registeredSink;
},
};
const fakePool = { query() {}, connect() {} };
const artifactStore = { put() {}, inspect() {}, readLogRange() {} };
const copilot = Object.freeze({ execute() {} });
let registeredSink;
let created;
let controlOptions;
try {
await Promise.all([
projectedFile(configRoot, 'config.json', config),
projectedFile(invocationRoot, 'keyring.json', invocation),
projectedFile(resultRoot, 'keyring.json', result),
projectedFile(outputRoot, 'keyring.json', output),
]);
const application = await startProductionClusterAiControlApplication({
control: {
config: controlConfig(),
workerIngress: {
config: { enabled: true },
artifactStore,
},
},
ai: {
enabled: true,
providerAuthorityFile: '/unused/providers.json',
secretRootDirectory: secretRoot,
copilot: {
configFile: join(configRoot, 'config.json'),
invocationKeyringRootDirectory: invocationRoot,
resultKeyringRootDirectory: resultRoot,
outputKeyringRootDirectory: outputRoot,
},
maxConcurrent: 1,
recoveryLimit: 1,
databaseMaxConnections: 1,
},
audit() {},
async openAiDatabase() {
return { pool: fakePool, async close() {} };
},
async bootstrapPrompt(options) {
await options.openDatabase();
registeredSink = options.createAdditionalSuccessfulCompletion({
async recordWithAtomicSuccess() {},
});
return {
status: 'active', profile: 'cluster', readiness: {}, capability: gateway,
prompts: {}, promptCatalog: {}, promptExecutions: {},
promptExecutionInspections: {}, async stop() { return 'stopped'; },
};
},
async createCopilot(options) {
created = options;
return copilot;
},
async startControl(options) {
controlOptions = options;
return {
status: 'active', address: { host: '127.0.0.1', port: 5800 },
evidence: {}, recovery: { safe: true, remaining: 0, failed: 0 },
unavailable: new Promise(() => {}), availabilityStatus() { return 'ready'; },
async stop() { return 'stopped'; },
};
},
});
assert.equal(application.copilot, copilot);
assert.equal(created.pool, fakePool);
assert.equal(created.gateway, gateway);
assert.equal(created.successfulCompletion, registeredSink);
assert.equal(created.artifactStore, artifactStore);
assert.equal('copilot' in controlOptions, false);
assert.equal(await application.stop(), 'stopped');
} finally {
config.fill(0); invocation.fill(0); result.fill(0); output.fill(0);
await Promise.all([
rm(secretRoot, { recursive: true, force: true }),
rm(configRoot, { recursive: true, force: true }),
rm(invocationRoot, { recursive: true, force: true }),
rm(resultRoot, { recursive: true, force: true }),
rm(outputRoot, { recursive: true, force: true }),
]);
}
});
test('explicit AI composition injects one reviewed Prompt capability and drains it after HTTP control', async () => {
@@ -0,0 +1,92 @@
const assert = require('node:assert/strict');
const { chmod, mkdtemp, rm, writeFile } = require('node:fs/promises');
const { tmpdir } = require('node:os');
const { join } = require('node:path');
const { test } = require('node:test');
const {
CLUSTER_COPILOT_FAILURE_DIAGNOSIS_CONFIG_SCHEMA,
ClusterCopilotFailureDiagnosisCompositionError,
canonicalClusterCopilotFailureDiagnosisConfig,
loadClusterCopilotFailureDiagnosisConfig,
normalizeClusterCopilotFailureDiagnosisConfig,
} = require('@qinglong/cluster-control/copilot-production');
function config(overrides = {}) {
return {
schema: CLUSTER_COPILOT_FAILURE_DIAGNOSIS_CONFIG_SCHEMA,
provider: 'provider-primary',
model: 'model-diagnosis',
modelBoundary: 'external',
responseLanguage: 'zh-CN',
maxOutputTokens: 512,
executionTimeoutMs: 60_000,
egressPolicy: {
schema: 'qinglong/copilot-model-egress-policy@v1',
revision: 'cluster-copilot-v1',
potentiallySensitiveDataBoundaries: ['external'],
maxInputBytes: 64 * 1024,
maxOutputTokens: 1024,
},
...overrides,
};
}
test('normalizes one bounded deployment-owned Copilot model intent', () => {
const normalized = normalizeClusterCopilotFailureDiagnosisConfig(config());
assert.equal(normalized.provider, 'provider-primary');
assert.equal(normalized.executionTimeoutMs, 60_000);
assert.equal(Object.isFrozen(normalized), true);
assert.equal(Object.isFrozen(normalized.egressPolicy), true);
assert.throws(
() => normalizeClusterCopilotFailureDiagnosisConfig(config({ extra: true })),
ClusterCopilotFailureDiagnosisCompositionError,
);
assert.throws(
() => normalizeClusterCopilotFailureDiagnosisConfig(config({
modelBoundary: 'on_device',
})),
/egress model boundaries/,
);
assert.throws(
() => normalizeClusterCopilotFailureDiagnosisConfig(config({
executionTimeoutMs: 300_001,
})),
/execution timeout/,
);
});
test('loads only canonical read-only projected Copilot configuration', async () => {
const root = await mkdtemp(join(tmpdir(), 'ql3-copilot-config-test-'));
const file = join(root, 'config.json');
const canonical = canonicalClusterCopilotFailureDiagnosisConfig(config());
try {
await writeFile(file, canonical, { mode: 0o440 });
await chmod(file, 0o440);
assert.deepEqual(
await loadClusterCopilotFailureDiagnosisConfig(file),
normalizeClusterCopilotFailureDiagnosisConfig(config()),
);
await chmod(file, 0o640);
await writeFile(file, Buffer.from(`${JSON.stringify(config(), null, 2)}\n`), {
mode: 0o440,
});
await chmod(file, 0o440);
await assert.rejects(
loadClusterCopilotFailureDiagnosisConfig(file),
/not canonical/,
);
await chmod(file, 0o640);
await writeFile(file, canonical, { mode: 0o640 });
await chmod(file, 0o640);
await assert.rejects(
loadClusterCopilotFailureDiagnosisConfig(file),
ClusterCopilotFailureDiagnosisCompositionError,
);
} finally {
canonical.fill(0);
await rm(root, { recursive: true, force: true });
}
});