feat(server): impl doc gc (#15282)

#### PR Dependency Tree


* **PR #15282** 👈

This tree was auto-generated by
[Charcoal](https://github.com/danerwilliams/charcoal)

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit

* **New Features**
* Added automated document cleanup to reconcile missing workspace docs,
delete related stored data, and recover if the doc returns.
* Added effect-based follow-up reconciliation for search indexing,
Copilot embeddings, and comment attachment cleanup with explicit
acknowledgements.
* **Bug Fixes**
* Deleted-document references now persist as dangling references rather
than disappearing.
* Improved document deletion flow to enforce permissions and ensure
authorized deletions succeed.
* **Tests**
* Expanded coverage for cleanup recovery, indexing/embedding
reconciliation, permissions, and reference semantics.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
This commit is contained in:
DarkSky
2026-07-20 15:29:23 +08:00
committed by GitHub
parent 81df4751a3
commit bb55d6fd21
30 changed files with 2657 additions and 337 deletions
@@ -28,7 +28,7 @@ import {
WorkspaceModel,
WorkspaceRole,
} from '../../models';
import type { LlmToolCallbackRequest } from '../../native';
import { addDocToRootDoc, type LlmToolCallbackRequest } from '../../native';
import { CopilotModule } from '../../plugins/copilot';
import { CopilotContextService } from '../../plugins/copilot/context';
import { CopilotContextResolver } from '../../plugins/copilot/context/resolver';
@@ -260,6 +260,80 @@ test.after.always(async t => {
await t.context.module?.close();
});
test('document cleanup reconciles missing and restored copilot state before ack', async t => {
const { db, jobs, models, module, workspace } = t.context;
const queue = module.get(JobQueue);
const deleteEmbedding = Sinon.spy(
models.copilotContext,
'purgeWorkspaceEmbedding'
);
const scheduleEmbedding = Sinon.stub(
jobs,
'addDocEmbeddingQueueFromEvent'
).resolves();
for (const [docId, restored, cleanupVersion] of [
['missing-doc', false, 'missing-version'],
['restored-doc', true, 'restored-version'],
] as const) {
const ws = await workspace.create(userId);
const root = addDocToRootDoc(Buffer.from([0, 0]), docId, docId);
const missingRoot = addDocToRootDoc(
Buffer.from([0, 0]),
'live-doc',
'Live'
);
await db.snapshot.create({
data: {
workspaceId: ws.id,
id: ws.id,
blob: restored ? root : missingRoot,
state: Buffer.from([0, 0]),
updatedAt: new Date(),
createdAt: new Date(),
},
});
if (restored) {
await db.snapshot.create({
data: {
workspaceId: ws.id,
id: docId,
blob: addDocToRootDoc(Buffer.from([0, 0]), 'content', 'Content'),
state: Buffer.from([0, 0]),
updatedAt: new Date(),
createdAt: new Date(),
},
});
}
await jobs.reconcileDocumentCleanup({
workspaceId: ws.id,
docId,
cleanupVersion,
});
if (restored) {
t.true(scheduleEmbedding.calledOnceWith({ workspaceId: ws.id, docId }));
t.false(deleteEmbedding.called);
} else {
t.true(deleteEmbedding.calledOnceWith(ws.id, docId));
t.false(scheduleEmbedding.called);
}
const { payload } = await module.queue.waitFor(
'backendRuntime.ackDocumentCleanupEffect'
);
t.deepEqual(payload, {
workspaceId: ws.id,
docId,
cleanupVersion,
effect: 'copilot',
});
deleteEmbedding.resetHistory();
scheduleEmbedding.resetHistory();
(queue.add as Sinon.SinonStub).resetHistory();
}
});
test('MCP credentials stay bound to their endpoint, workspace, and profile', async t => {
const { db, mcpCredentials, mcpProvider, models, workspace } = t.context;
const ws = await workspace.create(userId);
@@ -671,7 +671,7 @@ test('active users metric should dedupe multiple sockets for one user', async t
test('workspace sync delete-doc should enforce doc permissions', async t => {
const db = app.get(PrismaClient);
const models = app.get(Models);
const { user: owner } = await login(app);
const { user: owner, cookieHeader: ownerCookieHeader } = await login(app);
const { user: collaborator, cookieHeader } = await login(app);
const workspace = await models.workspace.create(owner.id);
const docId = 'private-doc';
@@ -692,9 +692,10 @@ test('workspace sync delete-doc should enforce doc permissions', async t => {
});
const socket = createClient(url, cookieHeader);
const ownerSocket = createClient(url, ownerCookieHeader);
try {
await waitForConnect(socket);
await Promise.all([waitForConnect(socket), waitForConnect(ownerSocket)]);
const join = unwrapResponse(
t,
@@ -719,8 +720,37 @@ test('workspace sync delete-doc should enforce doc permissions', async t => {
})
);
t.true(error.message.includes('Doc.Delete'));
const ownerJoin = unwrapResponse(
t,
await emitWithAck<{ clientId: string; success: boolean }>(
ownerSocket,
'space:join',
{
spaceType: 'workspace',
spaceId: workspace.id,
clientVersion: '0.26.0',
}
)
);
t.true(ownerJoin.success);
unwrapResponse(
t,
await emitWithAck(ownerSocket, 'space:delete-doc', {
spaceType: 'workspace',
spaceId: workspace.id,
docId,
})
);
t.is(
await db.snapshot.count({
where: { workspaceId: workspace.id, id: docId },
}),
1
);
} finally {
socket.disconnect();
ownerSocket.disconnect();
}
});
@@ -137,8 +137,8 @@ export class PgWorkspaceDocStorageAdapter extends DocStorageAdapter {
}));
}
async deleteDoc(workspaceId: string, docId: string) {
await this.models.doc.delete(workspaceId, docId);
async deleteDoc(_workspaceId: string, _docId: string) {
return;
}
async deleteSpace(workspaceId: string) {
@@ -258,6 +258,33 @@ export class StorageRuntimeProvider
);
}
async reconcileWorkspaceDocuments(workspaceId: string) {
return await this.measured('reconcileWorkspaceDocuments', rt =>
rt.reconcileWorkspaceDocuments(workspaceId)
);
}
async executeDocumentCleanupCandidates(
workspaceId: string | null | undefined,
gracePeriodDays: number,
limit: number
) {
return await this.measured('executeDocumentCleanupCandidates', rt =>
rt.executeDocumentCleanupCandidates(workspaceId, gracePeriodDays, limit)
);
}
async ackDocumentCleanupEffect(
workspaceId: string,
docId: string,
cleanupVersion: string,
effect: 'search' | 'copilot'
) {
return await this.measured('ackDocumentCleanupEffect', rt =>
rt.ackDocumentCleanupEffect(workspaceId, docId, cleanupVersion, effect)
);
}
async planUnreferencedWorkspaceBlobs(
workspaceId: string,
gracePeriodDays: number,
@@ -6,10 +6,13 @@ import { StorageBlobJob } from '../blob-job';
interface Context {
runtime: {
health: Sinon.SinonStub;
reconcileWorkspaceDocuments: Sinon.SinonStub;
backfillMissingBlobMetadata: Sinon.SinonStub;
rebuildWorkspaceDocBlobRefs: Sinon.SinonStub;
planUnreferencedWorkspaceBlobs: Sinon.SinonStub;
executeBlobCleanupCandidates: Sinon.SinonStub;
executeDocumentCleanupCandidates: Sinon.SinonStub;
ackDocumentCleanupEffect: Sinon.SinonStub;
};
event: {
emitAsync: Sinon.SinonStub;
@@ -35,10 +38,18 @@ test.beforeEach(t => {
providerConfigured: true,
provider: 'fs',
}),
reconcileWorkspaceDocuments: Sinon.stub().resolves({
scannedDocs: 1,
marked: 0,
reset: 0,
recovered: 0,
}),
backfillMissingBlobMetadata: Sinon.stub(),
rebuildWorkspaceDocBlobRefs: Sinon.stub(),
planUnreferencedWorkspaceBlobs: Sinon.stub(),
executeBlobCleanupCandidates: Sinon.stub(),
executeDocumentCleanupCandidates: Sinon.stub(),
ackDocumentCleanupEffect: Sinon.stub(),
};
t.context.event = {
emitAsync: Sinon.stub().resolves(undefined),
@@ -47,7 +58,18 @@ test.beforeEach(t => {
add: Sinon.stub().resolves(undefined),
};
t.context.db = {
$queryRaw: Sinon.stub(),
$queryRaw: Sinon.stub().resolves([
{
marked: 0n,
failed: 0n,
effectsPending: 0n,
failedWorkspaceCheckpoints: 0n,
rootFailureCheckpoints: 0n,
staleProjectionBlocks: 0n,
oldestFailedSeconds: null,
oldestEffectsPendingSeconds: null,
},
]),
workspace: {
findMany: Sinon.stub(),
},
@@ -129,46 +151,6 @@ for (const scenario of objectStorageRequiredCases) {
});
}
test('doc blob refs sweep continues after one workspace fails', async t => {
t.context.db.workspace.findMany.resolves([
{ id: 'workspace-1', sid: 1 },
{ id: 'workspace-2', sid: 2 },
]);
t.context.runtime.rebuildWorkspaceDocBlobRefs
.onFirstCall()
.rejects(new Error('bad root doc'))
.onSecondCall()
.resolves({
scannedDocs: 1,
parsedDocs: 1,
refsWritten: 0,
refsDeleted: 0,
failedDocs: 0,
nextCursor: null,
});
await t.context.job.rebuildWorkspaceDocBlobRefsBySid({
workspaceLimit: 2,
docLimit: 100,
});
t.is(t.context.runtime.rebuildWorkspaceDocBlobRefs.callCount, 2);
t.deepEqual(t.context.runtime.rebuildWorkspaceDocBlobRefs.firstCall.args, [
'workspace-1',
100,
]);
t.deepEqual(t.context.runtime.rebuildWorkspaceDocBlobRefs.secondCall.args, [
'workspace-2',
100,
]);
t.true(
t.context.queue.add.calledWith(
'backendRuntime.rebuildWorkspaceDocBlobRefsBySid',
{ lastSid: 2, workspaceLimit: 2, docLimit: 100 }
)
);
});
test('blob cleanup planning drains each workspace cursor before continuing', async t => {
t.context.db.workspace.findMany.resolves([
{ id: 'workspace-1', sid: 1 },
@@ -229,6 +211,13 @@ test('blob cleanup planning drains each workspace cursor before continuing', asy
test('daily blob cleanup execution uses a fixed job id', async t => {
await t.context.job.dailyBlobCleanupExecution();
t.true(
t.context.queue.add.calledWith(
'backendRuntime.executeDocumentCleanupCandidates',
{},
{ jobId: 'daily-backend-runtime-document-cleanup-execution' }
)
);
t.true(
t.context.queue.add.calledWith(
'backendRuntime.executeBlobCleanupCandidatesByMarkedRuns',
@@ -238,6 +227,132 @@ test('daily blob cleanup execution uses a fixed job id', async t => {
);
});
test('daily storage reconciliation uses a fixed job id', async t => {
await t.context.job.dailyStorageReconciliation();
t.true(
t.context.queue.add.calledWith(
'backendRuntime.reconcileWorkspaceStorageBySid',
{},
{ jobId: 'daily-backend-runtime-storage-reconciliation' }
)
);
});
test('storage reconciliation orders document retention before blob cleanup', async t => {
t.context.db.workspace.findMany.resolves([{ id: 'workspace-1', sid: 1 }]);
t.context.runtime.rebuildWorkspaceDocBlobRefs.resolves({
scannedDocs: 1,
parsedDocs: 1,
refsWritten: 1,
refsDeleted: 0,
failedDocs: 0,
nextCursor: null,
});
t.context.runtime.planUnreferencedWorkspaceBlobs.resolves({
runId: 'run-1',
scannedBlobs: 1,
candidatesMarked: 0,
nextCursor: null,
});
await t.context.job.reconcileWorkspaceStorageBySid({ workspaceLimit: 10 });
Sinon.assert.callOrder(
t.context.runtime.reconcileWorkspaceDocuments,
t.context.runtime.rebuildWorkspaceDocBlobRefs,
t.context.runtime.planUnreferencedWorkspaceBlobs
);
t.pass();
});
test('storage reconciliation still refreshes document retention without object storage', async t => {
t.context.runtime.health.resolves({
databaseConnected: true,
providerConfigured: true,
provider: undefined,
});
t.context.db.workspace.findMany.resolves([{ id: 'workspace-1', sid: 1 }]);
t.context.runtime.rebuildWorkspaceDocBlobRefs.resolves({
scannedDocs: 1,
parsedDocs: 1,
refsWritten: 0,
refsDeleted: 0,
failedDocs: 0,
nextCursor: null,
});
await t.context.job.reconcileWorkspaceStorageBySid({});
t.true(t.context.runtime.reconcileWorkspaceDocuments.calledOnce);
t.true(t.context.runtime.rebuildWorkspaceDocBlobRefs.calledOnce);
t.false(t.context.runtime.planUnreferencedWorkspaceBlobs.called);
});
test('document cleanup dispatches independent stable search and copilot effects', async t => {
t.context.runtime.executeDocumentCleanupCandidates.resolves({
scannedCandidates: 1,
serializationRetries: 0,
executed: 1,
recovered: 0,
reset: 0,
failed: 0,
deletedRows: 3,
effects: [
{
workspaceId: 'workspace-1',
docId: 'doc-1',
cleanupVersion: 'version-1',
commentObjectsDone: true,
searchDone: false,
copilotDone: false,
},
],
});
await t.context.job.executeDocumentCleanupCandidates({});
t.true(
t.context.queue.add.calledWith(
'indexer.reconcileDocumentCleanup',
Sinon.match({ docId: 'doc-1' }),
{
jobId: 'document-cleanup:search:workspace-1:doc-1:version-1',
}
)
);
t.true(
t.context.queue.add.calledWith(
'copilot.embedding.reconcileDocumentCleanup',
Sinon.match({ docId: 'doc-1' }),
{
jobId: 'document-cleanup:copilot:workspace-1:doc-1:version-1',
}
)
);
t.true(
t.context.event.emitAsync.calledWith('workspace.blobs.updated', {
workspaceId: 'workspace-1',
})
);
});
test('document cleanup effect ack delegates to storage runtime', async t => {
await t.context.job.ackDocumentCleanupEffect({
workspaceId: 'workspace-1',
docId: 'doc-1',
cleanupVersion: 'version-1',
effect: 'search',
});
t.deepEqual(t.context.runtime.ackDocumentCleanupEffect.firstCall.args, [
'workspace-1',
'doc-1',
'version-1',
'search',
]);
});
test('blob cleanup execution sweep drains marked runs and continues by page', async t => {
t.context.db.$queryRaw
.onFirstCall()
@@ -2,7 +2,7 @@ import { Injectable, Logger } from '@nestjs/common';
import { Cron, CronExpression } from '@nestjs/schedule';
import { PrismaClient } from '@prisma/client';
import { EventBus, JobQueue, OnJob } from '../../base';
import { EventBus, JobQueue, metrics, OnJob } from '../../base';
import { StorageRuntimeProvider } from '../storage-runtime';
// Queue keys are persisted API; keep the legacy backendRuntime.* names while
@@ -18,15 +18,22 @@ declare global {
workspaceLimit?: number;
objectLimit?: number;
};
'backendRuntime.rebuildWorkspaceDocBlobRefs': {
workspaceId: string;
limit?: number;
};
'backendRuntime.rebuildWorkspaceDocBlobRefsBySid': {
'backendRuntime.reconcileWorkspaceStorageBySid': {
lastSid?: number;
workspaceLimit?: number;
docLimit?: number;
};
'backendRuntime.executeDocumentCleanupCandidates': {
workspaceId?: string;
gracePeriodDays?: number;
limit?: number;
};
'backendRuntime.ackDocumentCleanupEffect': {
workspaceId: string;
docId: string;
cleanupVersion: string;
effect: 'search' | 'copilot';
};
'backendRuntime.planUnreferencedWorkspaceBlobs': {
workspaceId: string;
gracePeriodDays?: number;
@@ -89,25 +96,6 @@ export class StorageBlobJob {
});
}
async enqueueRebuildWorkspaceDocBlobRefs(workspaceId: string, limit = 1000) {
await this.queue.add('backendRuntime.rebuildWorkspaceDocBlobRefs', {
workspaceId,
limit,
});
}
async enqueueRebuildWorkspaceDocBlobRefsBySid(
lastSid = 0,
workspaceLimit = 100,
docLimit = 1000
) {
await this.queue.add('backendRuntime.rebuildWorkspaceDocBlobRefsBySid', {
lastSid,
workspaceLimit,
docLimit,
});
}
@OnJob('backendRuntime.backfillMissingBlobMetadataBySid')
async backfillMissingBlobMetadataBySid({
lastSid = 0,
@@ -211,25 +199,21 @@ export class StorageBlobJob {
}
@Cron(CronExpression.EVERY_DAY_AT_2AM)
async dailyDocBlobRefsRebuild() {
async dailyStorageReconciliation() {
await this.queue.add(
'backendRuntime.rebuildWorkspaceDocBlobRefsBySid',
'backendRuntime.reconcileWorkspaceStorageBySid',
{},
{ jobId: 'daily-backend-runtime-doc-blob-refs-rebuild' }
);
}
@Cron(CronExpression.EVERY_DAY_AT_3AM)
async dailyBlobCleanupPlanning() {
await this.queue.add(
'backendRuntime.planUnreferencedWorkspaceBlobsBySid',
{},
{ jobId: 'daily-backend-runtime-blob-cleanup-planning' }
{ jobId: 'daily-backend-runtime-storage-reconciliation' }
);
}
@Cron(CronExpression.EVERY_DAY_AT_4AM)
async dailyBlobCleanupExecution() {
await this.queue.add(
'backendRuntime.executeDocumentCleanupCandidates',
{},
{ jobId: 'daily-backend-runtime-document-cleanup-execution' }
);
await this.queue.add(
'backendRuntime.executeBlobCleanupCandidatesByMarkedRuns',
{},
@@ -249,44 +233,39 @@ export class StorageBlobJob {
await this.drainBlobMetadataBackfill(workspaceId, limit);
}
@OnJob('backendRuntime.rebuildWorkspaceDocBlobRefs')
async rebuildWorkspaceDocBlobRefs({
workspaceId,
limit = 1000,
}: Jobs['backendRuntime.rebuildWorkspaceDocBlobRefs']) {
await this.drainWorkspaceDocBlobRefs(workspaceId, limit);
}
@OnJob('backendRuntime.rebuildWorkspaceDocBlobRefsBySid')
async rebuildWorkspaceDocBlobRefsBySid({
@OnJob('backendRuntime.reconcileWorkspaceStorageBySid')
async reconcileWorkspaceStorageBySid({
lastSid = 0,
workspaceLimit = 100,
docLimit = 1000,
}: Jobs['backendRuntime.rebuildWorkspaceDocBlobRefsBySid']) {
}: Jobs['backendRuntime.reconcileWorkspaceStorageBySid']) {
const workspaces = await this.db.workspace.findMany({
where: {
sid: {
gt: lastSid,
},
},
orderBy: {
sid: 'asc',
},
select: {
id: true,
sid: true,
},
where: { sid: { gt: lastSid } },
orderBy: { sid: 'asc' },
select: { id: true, sid: true },
take: workspaceLimit,
});
const objectStorageConfigured = await this.hasObjectStorage(
'storage reconciliation blob cleanup planning'
);
for (const workspace of workspaces) {
try {
await this.rt.reconcileWorkspaceDocuments(workspace.id);
await this.drainWorkspaceDocBlobRefs(workspace.id, docLimit, {
sid: workspace.sid,
});
if (objectStorageConfigured) {
await this.drainBlobCleanupPlanning(workspace.id, 30, 1000, {
sid: workspace.sid,
});
}
} catch (err) {
metrics.storage
.counter('document_cleanup_workspace_failure_total')
.add(1);
this.logger.error(
`doc blob refs rebuild failed workspace=${workspace.id} sid=${workspace.sid}`,
`storage reconciliation failed workspace=${workspace.id} sid=${workspace.sid}`,
err
);
}
@@ -294,14 +273,71 @@ export class StorageBlobJob {
const nextSid = workspaces.at(-1)?.sid;
if (nextSid !== undefined && workspaces.length === workspaceLimit) {
await this.enqueueRebuildWorkspaceDocBlobRefsBySid(
nextSid,
await this.queue.add('backendRuntime.reconcileWorkspaceStorageBySid', {
lastSid: nextSid,
workspaceLimit,
docLimit
);
docLimit,
});
}
}
@OnJob('backendRuntime.executeDocumentCleanupCandidates')
async executeDocumentCleanupCandidates({
workspaceId,
gracePeriodDays = 30,
limit = 100,
}: Jobs['backendRuntime.executeDocumentCleanupCandidates']) {
const result = await this.rt.executeDocumentCleanupCandidates(
workspaceId,
gracePeriodDays,
limit
);
metrics.storage
.counter('document_cleanup_serialization_retry_total')
.add(result.serializationRetries);
metrics.storage
.counter('document_cleanup_execute_failure_total')
.add(result.failed);
for (const effect of result.effects) {
if (!effect.searchDone) {
await this.queue.add('indexer.reconcileDocumentCleanup', effect, {
jobId: `document-cleanup:search:${effect.workspaceId}:${effect.docId}:${effect.cleanupVersion}`,
});
}
if (!effect.copilotDone) {
await this.queue.add(
'copilot.embedding.reconcileDocumentCleanup',
effect,
{
jobId: `document-cleanup:copilot:${effect.workspaceId}:${effect.docId}:${effect.cleanupVersion}`,
}
);
}
if (effect.commentObjectsDone) {
await this.event.emitAsync('workspace.blobs.updated', {
workspaceId: effect.workspaceId,
});
}
}
await this.recordDocumentCleanupHealth();
return result;
}
@OnJob('backendRuntime.ackDocumentCleanupEffect')
async ackDocumentCleanupEffect({
workspaceId,
docId,
cleanupVersion,
effect,
}: Jobs['backendRuntime.ackDocumentCleanupEffect']) {
await this.rt.ackDocumentCleanupEffect(
workspaceId,
docId,
cleanupVersion,
effect
);
}
@OnJob('backendRuntime.planUnreferencedWorkspaceBlobs')
async planUnreferencedWorkspaceBlobs({
workspaceId,
@@ -545,6 +581,82 @@ export class StorageBlobJob {
return rows.map(row => row.runId);
}
private async recordDocumentCleanupHealth() {
const [health] = await this.db.$queryRaw<
{
marked: bigint;
failed: bigint;
effectsPending: bigint;
failedWorkspaceCheckpoints: bigint;
rootFailureCheckpoints: bigint;
staleProjectionBlocks: bigint;
oldestFailedSeconds: number | null;
oldestEffectsPendingSeconds: number | null;
}[]
>`
SELECT
COUNT(*) FILTER (WHERE status = 'marked') AS marked,
COUNT(*) FILTER (WHERE status = 'failed') AS failed,
COUNT(*) FILTER (WHERE status = 'effects_pending') AS "effectsPending",
EXTRACT(EPOCH FROM CURRENT_TIMESTAMP -
MIN(updated_at) FILTER (WHERE status = 'failed'))::double precision AS "oldestFailedSeconds",
EXTRACT(EPOCH FROM CURRENT_TIMESTAMP -
MIN(updated_at) FILTER (WHERE status = 'effects_pending'))::double precision AS "oldestEffectsPendingSeconds",
(SELECT COUNT(*) FROM storage_reconciliation_checkpoints
WHERE kind = 'document_cleanup' AND status = 'failed') AS "failedWorkspaceCheckpoints",
(SELECT COUNT(*) FROM storage_reconciliation_checkpoints
WHERE kind = 'document_cleanup' AND status = 'failed'
AND metadata->>'failureKind' = 'root') AS "rootFailureCheckpoints",
(SELECT COUNT(*) FROM storage_reconciliation_runs
WHERE kind = 'blob_cleanup_plan'
AND started_at >= CURRENT_TIMESTAMP - INTERVAL '1 day'
AND jsonb_array_length(COALESCE(metadata->'staleOrFailedProjectionWorkspaces', '[]')) > 0
) AS "staleProjectionBlocks"
FROM document_cleanup_candidates
`;
if (!health) {
return;
}
for (const [name, value] of [
['document_cleanup_marked', health.marked],
['document_cleanup_failed', health.failed],
['document_cleanup_effects_pending', health.effectsPending],
[
'document_cleanup_failed_workspace_checkpoints',
health.failedWorkspaceCheckpoints,
],
[
'document_cleanup_root_failure_checkpoints',
health.rootFailureCheckpoints,
],
[
'document_cleanup_stale_projection_blocks',
health.staleProjectionBlocks,
],
] as const) {
metrics.storage.gauge(name).record(Number(value));
}
metrics.storage
.gauge('document_cleanup_oldest_failed_seconds')
.record(health.oldestFailedSeconds ?? 0);
metrics.storage
.gauge('document_cleanup_oldest_effects_pending_seconds')
.record(health.oldestEffectsPendingSeconds ?? 0);
if (
health.failedWorkspaceCheckpoints > 0n ||
health.rootFailureCheckpoints > 0n ||
health.staleProjectionBlocks > 0n ||
(health.oldestFailedSeconds ?? 0) >= 86_400 ||
(health.oldestEffectsPendingSeconds ?? 0) >= 86_400 ||
health.marked >= 10_000n
) {
this.logger.warn(
`document cleanup health marked=${health.marked} failed=${health.failed} effectsPending=${health.effectsPending} failedWorkspaceCheckpoints=${health.failedWorkspaceCheckpoints} rootFailureCheckpoints=${health.rootFailureCheckpoints} staleProjectionBlocks=${health.staleProjectionBlocks} oldestFailedSeconds=${health.oldestFailedSeconds ?? 0} oldestEffectsPendingSeconds=${health.oldestEffectsPendingSeconds ?? 0}`
);
}
}
private async hasMarkedBlobCleanupCandidates() {
const rows = await this.db.$queryRaw<{ exists: boolean }[]>`
SELECT EXISTS(
@@ -1,5 +1,6 @@
import test from 'ava';
import { omit } from 'lodash-es';
import * as Y from 'yjs';
import { createModule } from '../../../__tests__/create-module';
import { Mockers } from '../../../__tests__/mocks';
@@ -38,6 +39,86 @@ test('can read all doc ids from workspace snapshot', async t => {
t.snapshot(docIds);
});
test('merged root updates retain trash and exclude permanently removed docs', async t => {
const rootDoc = await models.doc.get(workspace.id, workspace.id);
t.truthy(rootDoc);
const root = new Y.Doc();
Y.applyUpdate(root, rootDoc!.blob);
const pending = new Y.Doc();
Y.applyUpdate(pending, Y.encodeStateAsUpdate(root));
const trashMeta = new Y.Map<unknown>();
trashMeta.set('id', 'trash-doc');
trashMeta.set('title', 'Trash');
trashMeta.set('trash', true);
(pending.getMap('meta').get('pages') as Y.Array<Y.Map<unknown>>).push([
trashMeta,
]);
const pendingUpdate = Y.encodeStateAsUpdate(
pending,
Y.encodeStateVector(root)
);
const merged = Y.mergeUpdates([rootDoc!.blob, pendingUpdate]);
t.false(readAllDocIdsFromWorkspaceSnapshot(merged).includes('trash-doc'));
t.true(
readAllDocIdsFromWorkspaceSnapshot(merged, true).includes('trash-doc')
);
const removed = new Y.Doc();
Y.applyUpdate(removed, merged);
const pages = removed.getMap('meta').get('pages') as Y.Array<Y.Map<unknown>>;
const target = pages
.toArray()
.findIndex(meta => meta instanceof Y.Map && meta.get('id') === 'trash-doc');
pages.delete(target, 1);
t.false(
readAllDocIdsFromWorkspaceSnapshot(
Y.encodeStateAsUpdate(removed),
true
).includes('trash-doc')
);
});
test('nested concurrent meta edits do not restore a deleted entry', t => {
const initial = new Y.Doc();
const pages = new Y.Array<Y.Map<unknown>>();
const meta = new Y.Map<unknown>();
meta.set('id', 'doc-1');
meta.set('title', 'Initial');
pages.push([meta]);
initial.getMap('meta').set('pages', pages);
const state = Y.encodeStateAsUpdate(initial);
const deletingClient = new Y.Doc();
const editingClient = new Y.Doc();
Y.applyUpdate(deletingClient, state);
Y.applyUpdate(editingClient, state);
(
deletingClient.getMap('meta').get('pages') as Y.Array<Y.Map<unknown>>
).delete(0, 1);
(editingClient.getMap('meta').get('pages') as Y.Array<Y.Map<unknown>>)
.get(0)
.set('title', 'Offline edit');
const merged = new Y.Doc();
Y.applyUpdate(merged, Y.encodeStateAsUpdate(deletingClient));
Y.applyUpdate(merged, Y.encodeStateAsUpdate(editingClient));
t.is(
(merged.getMap('meta').get('pages') as Y.Array<Y.Map<unknown>>).length,
0
);
(editingClient.getMap('meta').get('pages') as Y.Array<Y.Map<unknown>>).push([
meta.clone(),
]);
Y.applyUpdate(merged, Y.encodeStateAsUpdate(editingClient));
t.is(
(merged.getMap('meta').get('pages') as Y.Array<Y.Map<unknown>>).length,
1
);
});
test('can read all blocks from doc snapshot', async t => {
const rootDoc = await models.doc.get(workspace.id, workspace.id);
t.truthy(rootDoc);
@@ -45,8 +45,11 @@ export function parsePageDoc(
);
}
export function readAllDocIdsFromWorkspaceSnapshot(snapshot: Uint8Array) {
return readAllDocIdsFromRootDoc(Buffer.from(snapshot), false);
export function readAllDocIdsFromWorkspaceSnapshot(
snapshot: Uint8Array,
includeTrash = false
) {
return readAllDocIdsFromRootDoc(Buffer.from(snapshot), includeTrash);
}
function safeParseJson<T>(str: string): T | undefined {
@@ -337,10 +337,14 @@ export class CopilotContextModel extends BaseModel {
}
async deleteWorkspaceEmbedding(workspaceId: string, docId: string) {
await this.purgeWorkspaceEmbedding(workspaceId, docId);
await this.fulfillEmptyEmbedding(workspaceId, docId);
}
async purgeWorkspaceEmbedding(workspaceId: string, docId: string) {
await this.db.aiWorkspaceEmbedding.deleteMany({
where: { workspaceId, docId },
});
await this.fulfillEmptyEmbedding(workspaceId, docId);
}
async matchWorkspaceEmbedding(
@@ -205,20 +205,49 @@ export class CopilotEmbeddingJob {
);
}
@OnJob('copilot.embedding.deleteDoc')
async deleteDocEmbeddingQueueFromEvent(
doc: Jobs['copilot.embedding.deleteDoc']
) {
private async deleteDocEmbedding(doc: {
workspaceId: string;
docId: string;
}) {
await this.queue.remove(
`workspace:embedding:${doc.workspaceId}:${doc.docId}`,
'copilot.embedding.docs'
);
await this.models.copilotContext.deleteWorkspaceEmbedding(
await this.models.copilotContext.purgeWorkspaceEmbedding(
doc.workspaceId,
doc.docId
);
}
@OnJob('copilot.embedding.reconcileDocumentCleanup')
async reconcileDocumentCleanup({
workspaceId,
docId,
cleanupVersion,
}: Jobs['copilot.embedding.reconcileDocumentCleanup']) {
const root = await this.doc.getDoc(workspaceId, workspaceId);
if (!root) {
throw new Error(`workspace root ${workspaceId} not found`);
}
const live = readAllDocIdsFromWorkspaceSnapshot(root.bin, true).includes(
docId
);
if (live) {
if (!(await this.doc.getDoc(workspaceId, docId))) {
throw new Error(`restored document ${workspaceId}/${docId} not found`);
}
await this.addDocEmbeddingQueueFromEvent({ workspaceId, docId });
} else {
await this.deleteDocEmbedding({ workspaceId, docId });
}
await this.queue.add('backendRuntime.ackDocumentCleanupEffect', {
workspaceId,
docId,
cleanupVersion,
effect: 'copilot',
});
}
private async readCopilotBlob(
userId: string,
workspaceId: string,
@@ -68,9 +68,10 @@ declare global {
docId: string;
};
'copilot.embedding.deleteDoc': {
'copilot.embedding.reconcileDocumentCleanup': {
workspaceId: string;
docId: string;
cleanupVersion: string;
};
'copilot.embedding.files': {
@@ -39,10 +39,6 @@ declare global {
'copilot.session.generateTitle': {
sessionId: string;
};
'copilot.session.deleteDoc': {
workspaceId: string;
docId: string;
};
}
}
@@ -512,23 +508,6 @@ export class ChatSessionService {
return null;
}
@OnJob('copilot.session.deleteDoc')
async deleteDocSessions(doc: Jobs['copilot.session.deleteDoc']) {
const sessionIds = await this.models.copilotSession
.list({
userId: undefined,
workspaceId: doc.workspaceId,
docId: doc.docId,
})
.then(s => s.map(s => [s.userId, s.id]));
for (const [userId, sessionId] of sessionIds) {
await this.models.copilotSession.update(
{ userId, sessionId, docId: null },
true
);
}
}
@OnJob('copilot.session.generateTitle')
async generateSessionTitle(job: Jobs['copilot.session.generateTitle']) {
const { sessionId } = job;
@@ -6,14 +6,17 @@ import Sinon from 'sinon';
import { createModule } from '../../../__tests__/create-module';
import { Mockers } from '../../../__tests__/mocks';
import { JOB_SIGNAL } from '../../../base';
import { Config, JOB_SIGNAL } from '../../../base';
import { ConfigModule } from '../../../base/config';
import { ServerConfigModule } from '../../../core/config';
import { DocReader } from '../../../core/doc';
import { Models } from '../../../models';
import { addDocToRootDoc } from '../../../native';
import { SearchProviderFactory } from '../factory';
import { IndexerModule, IndexerService } from '../index';
import { IndexerJob } from '../job';
import { ManticoresearchProvider } from '../providers';
import { blockSQL, docSQL, SearchTable } from '../tables';
const module = await createModule({
imports: [
@@ -32,6 +35,8 @@ const indexerJob = module.get(IndexerJob);
const searchProviderFactory = module.get(SearchProviderFactory);
const manticoresearch = module.get(ManticoresearchProvider);
const models = module.get(Models);
const docReader = module.get(DocReader);
const config = module.get(Config);
const user = await module.create(Mockers.User);
const workspace = await module.create(Mockers.Workspace, {
@@ -39,6 +44,11 @@ const workspace = await module.create(Mockers.Workspace, {
owner: user,
});
test.before(async () => {
await manticoresearch.recreateTable(SearchTable.block, blockSQL);
await manticoresearch.recreateTable(SearchTable.doc, docSQL);
});
test.after.always(async () => {
await module.close();
});
@@ -105,7 +115,7 @@ test('should not sync existing doc', async t => {
t.is(module.queue.count('indexer.indexDoc'), count);
});
test('should delete doc from indexer when docId is not in workspace', async t => {
test('should delete dangling indexed docs absent from the root live set', async t => {
const count = module.queue.count('indexer.deleteDoc');
mock.method(indexerService, 'listDocIds', async () => {
return ['mock-doc-id1', 'mock-doc-id2'];
@@ -121,6 +131,104 @@ test('should delete doc from indexer when docId is not in workspace', async t =>
t.is(module.queue.count('indexer.deleteDoc'), count + 2);
});
test('document cleanup reconcile deletes missing search state before ack', async t => {
const deleteSpy = Sinon.spy(indexerService, 'deleteDoc');
const indexSpy = Sinon.spy(indexerService, 'indexDoc');
const cleanupWorkspace = await module.create(Mockers.Workspace, {
owner: user,
});
await module.create(Mockers.DocSnapshot, {
workspaceId: cleanupWorkspace.id,
docId: cleanupWorkspace.id,
user,
blob: addDocToRootDoc(Buffer.from([0, 0]), 'live-doc', 'Live'),
});
await indexerJob.reconcileDocumentCleanup({
workspaceId: cleanupWorkspace.id,
docId: 'missing-doc',
cleanupVersion: 'version-1',
});
t.true(deleteSpy.calledOnceWith(cleanupWorkspace.id, 'missing-doc'));
t.false(indexSpy.called);
const { payload } = await module.queue.waitFor(
'backendRuntime.ackDocumentCleanupEffect'
);
t.deepEqual(payload, {
workspaceId: cleanupWorkspace.id,
docId: 'missing-doc',
cleanupVersion: 'version-1',
effect: 'search',
});
});
test('document cleanup reconcile reindexes restored doc before ack', async t => {
const deleteSpy = Sinon.spy(indexerService, 'deleteDoc');
const indexSpy = Sinon.spy(indexerService, 'indexDoc');
const cleanupWorkspace = await module.create(Mockers.Workspace, {
owner: user,
});
await module.create(Mockers.DocSnapshot, {
workspaceId: cleanupWorkspace.id,
docId: cleanupWorkspace.id,
user,
blob: addDocToRootDoc(Buffer.from([0, 0]), 'restored-doc', 'Restored'),
});
await module.create(Mockers.DocSnapshot, {
workspaceId: cleanupWorkspace.id,
docId: 'restored-doc',
user,
});
const getDocSpy = Sinon.spy(docReader, 'getDoc');
await indexerJob.reconcileDocumentCleanup({
workspaceId: cleanupWorkspace.id,
docId: 'restored-doc',
cleanupVersion: 'version-2',
});
t.true(indexSpy.calledOnceWith(cleanupWorkspace.id, 'restored-doc'));
t.false(deleteSpy.called);
t.true(getDocSpy.calledWith(cleanupWorkspace.id, cleanupWorkspace.id));
t.true(getDocSpy.calledWith(cleanupWorkspace.id, 'restored-doc'));
const { payload } = await module.queue.waitFor(
'backendRuntime.ackDocumentCleanupEffect'
);
t.deepEqual(payload, {
workspaceId: cleanupWorkspace.id,
docId: 'restored-doc',
cleanupVersion: 'version-2',
effect: 'search',
});
});
test('document cleanup reconcile only acknowledges when indexer is disabled', async t => {
Sinon.stub(config.indexer, 'enabled').value(false);
const deleteSpy = Sinon.spy(indexerService, 'deleteDoc');
const indexSpy = Sinon.spy(indexerService, 'indexDoc');
const getDocSpy = Sinon.spy(docReader, 'getDoc');
await indexerJob.reconcileDocumentCleanup({
workspaceId: workspace.id,
docId: 'disabled-doc',
cleanupVersion: 'version-disabled',
});
t.false(deleteSpy.called);
t.false(indexSpy.called);
t.false(getDocSpy.called);
const { payload } = await module.queue.waitFor(
'backendRuntime.ackDocumentCleanupEffect'
);
t.deepEqual(payload, {
workspaceId: workspace.id,
docId: 'disabled-doc',
cleanupVersion: 'version-disabled',
effect: 'search',
});
});
test('should handle indexer.deleteWorkspace job', async t => {
const spy = Sinon.spy(indexerService, 'deleteWorkspace');
@@ -1887,12 +1887,9 @@ test('should delete doc work', async t => {
t.is(result4.nodes.length, 1);
t.deepEqual(result4.nodes[0].fields.docId, [docId2]);
const count = module.queue.count('copilot.embedding.deleteDoc');
await indexerService.deleteDoc(workspaceId, docId1, {
refresh: true,
});
t.is(module.queue.count('copilot.embedding.deleteDoc'), count + 1);
// make sure the docId1 is deleted
result1 = await indexerService.search({
@@ -3,6 +3,7 @@ import './config';
import { Module } from '@nestjs/common';
import { ServerConfigModule } from '../../core/config';
import { DocStorageModule } from '../../core/doc';
import { PermissionModule } from '../../core/permission';
import { QuotaServiceModule } from '../../core/quota';
import { IndexerEvent } from './event';
@@ -13,7 +14,12 @@ import { IndexerResolver } from './resolver';
import { IndexerService } from './service';
@Module({
imports: [ServerConfigModule, PermissionModule, QuotaServiceModule],
imports: [
ServerConfigModule,
DocStorageModule,
PermissionModule,
QuotaServiceModule,
],
providers: [
IndexerResolver,
IndexerService,
@@ -1,6 +1,7 @@
import { Injectable, Logger } from '@nestjs/common';
import { Config, JOB_SIGNAL, JobQueue, OnJob } from '../../base';
import { DocReader } from '../../core/doc';
import { readAllDocIdsFromWorkspaceSnapshot } from '../../core/utils/blocksuite';
import { Models } from '../../models';
import { IndexerService } from './service';
@@ -24,6 +25,11 @@ declare global {
'indexer.autoIndexWorkspaces': {
lastIndexedWorkspaceSid?: number;
};
'indexer.reconcileDocumentCleanup': {
workspaceId: string;
docId: string;
cleanupVersion: string;
};
}
}
@@ -35,7 +41,8 @@ export class IndexerJob {
private readonly models: Models,
private readonly service: IndexerService,
private readonly queue: JobQueue,
private readonly config: Config
private readonly config: Config,
private readonly doc: DocReader
) {}
@OnJob('indexer.indexDoc')
@@ -66,6 +73,39 @@ export class IndexerJob {
await this.service.deleteDoc(workspaceId, docId);
}
@OnJob('indexer.reconcileDocumentCleanup')
async reconcileDocumentCleanup({
workspaceId,
docId,
cleanupVersion,
}: Jobs['indexer.reconcileDocumentCleanup']) {
if (this.config.indexer.enabled) {
const root = await this.doc.getDoc(workspaceId, workspaceId);
if (!root) {
throw new Error(`workspace root ${workspaceId} not found`);
}
const live = readAllDocIdsFromWorkspaceSnapshot(root.bin, true).includes(
docId
);
if (live) {
if (!(await this.doc.getDoc(workspaceId, docId))) {
throw new Error(
`restored document ${workspaceId}/${docId} not found`
);
}
await this.service.indexDoc(workspaceId, docId);
} else {
await this.service.deleteDoc(workspaceId, docId);
}
}
await this.queue.add('backendRuntime.ackDocumentCleanupEffect', {
workspaceId,
docId,
cleanupVersion,
effect: 'search',
});
}
@OnJob('indexer.indexWorkspace')
async indexWorkspace({ workspaceId }: Jobs['indexer.indexWorkspace']) {
if (!this.config.indexer.enabled) {
@@ -79,16 +119,13 @@ export class IndexerJob {
return;
}
const snapshot = await this.models.doc.getSnapshot(
workspaceId,
workspaceId
);
if (!snapshot) {
const root = await this.doc.getDoc(workspaceId, workspaceId);
if (!root) {
this.logger.warn(`workspace snapshot ${workspaceId} not found`);
return;
}
const docIdsInWorkspace = readAllDocIdsFromWorkspaceSnapshot(snapshot.blob);
const docIdsInWorkspace = readAllDocIdsFromWorkspaceSnapshot(root.bin);
const docIdsInIndexer = await this.service.listDocIds(workspaceId);
const docIdsInWorkspaceSet = new Set(docIdsInWorkspace);
@@ -374,14 +374,6 @@ export class IndexerService {
);
await this.deleteBlocksByDocId(workspaceId, docId, options);
await this.queue.add('copilot.session.deleteDoc', {
workspaceId,
docId,
});
await this.queue.add('copilot.embedding.deleteDoc', {
workspaceId,
docId,
});
this.logger.log(`deleted doc ${workspaceId}/${docId}`);
}