Make document lifecycle fan-out durable and atomic

Pre-v47 `createDocumentWithLifecycle` wrapped each hook in its own
try/catch and always returned success, so a failure in any of the
downstream hooks (thread event, checkpoint approval, deliverable
review, automation dispatch, reviewer wake) left a live document row
behind with a silently incomplete lifecycle.

Two invariants now hold:

- DB-writing hooks (thread integration, checkpoint approval,
  deliverable review) run inside the same `BEGIN IMMEDIATE`
  transaction as `saveDocument`. A throw rolls back the document row
  along with every partial write, so an operator never sees a
  persisted document with a missing thread or skipped approval.

- Cross-process side effects (automation dispatch, reviewer wake) are
  enqueued as `document_lifecycle_events` rows inside the same
  transaction and processed post-commit. Success stamps
  `status = 'succeeded'`; failure stamps `status = 'failed'` with
  `last_error`, so partial lifecycle state is observable via
  `listFailedForCompany` instead of disappearing into the log tail.

Migration v47 installs the ledger; `buildDocumentLifecycleEventMethods`
exposes the enqueue/running/succeeded/failed transitions; the lifecycle
service drives both.
This commit is contained in:
ZacharyZhang-NY committed 2026-04-19 15:53:55 -04:00
1 parent a36884360a
commit edbebe6109
5 files changed
+326 -117

No files matched your search

+3
View File
@@ -24,6 +24,7 @@ import { buildWakeupMethods } from "./database/wakeup";
import { buildBackupMethods } from "./database/backup";
import { buildSessionEntryMethods } from "./database/session-entries";
import { buildInboxProjectionMethods } from "./database/inbox-projections";
import { buildDocumentLifecycleEventMethods } from "./database/document-lifecycle-events";
import { buildCredentialMethods } from "./database/credentials";
import { buildModelCapabilityMethods } from "./database/model-capabilities-crud";
import { MODEL_CAPABILITY_SEED } from "./model-capability-seed";
@@ -71,6 +72,7 @@ export class AppDatabaseBase {
public readonly backupMod: ReturnType<typeof buildBackupMethods>;
public readonly sessionEntries: ReturnType<typeof buildSessionEntryMethods>;
public readonly inboxProjections: ReturnType<typeof buildInboxProjectionMethods>;
public readonly documentLifecycleEvents: ReturnType<typeof buildDocumentLifecycleEventMethods>;
public readonly credentialsTable: ReturnType<typeof buildCredentialMethods>;
public readonly modelCapabilities: ReturnType<typeof buildModelCapabilityMethods>;
@@ -146,6 +148,7 @@ export class AppDatabaseBase {
this.wakeup = buildWakeupMethods(getDb, ra, tx);
this.sessionEntries = buildSessionEntryMethods(getDb, ra, tx);
this.inboxProjections = buildInboxProjectionMethods(getDb);
this.documentLifecycleEvents = buildDocumentLifecycleEventMethods(getDb, ra);
this.credentialsTable = buildCredentialMethods(getDb, ra, tx);
this.modelCapabilities = buildModelCapabilityMethods(getDb, ra, tx);
this.backupMod = buildBackupMethods({
@@ -0,0 +1,131 @@
import { randomUUID } from "node:crypto";
import type { DatabaseSync, SQLInputValue } from "node:sqlite";
import type { ReadAll } from "./helpers";
import { nowIso } from "./helpers";
/**
* Document-lifecycle event row shape returned by listers.
* `status` moves `pending -> running -> (succeeded | failed)`. `attempts`
* tracks how many processor passes have tried this hook so a sweeper can
* bound retries without losing the ledger.
*/
export interface DocumentLifecycleEventRow {
id: string;
companyId: string;
documentId: string;
hookType: string;
status: "pending" | "running" | "succeeded" | "failed";
attempts: number;
lastError: string | null;
payloadJson: string;
createdAt: string;
updatedAt: string;
completedAt: string | null;
}
export interface EnqueueInput {
companyId: string;
documentId: string;
hookType: string;
payloadJson?: string;
}
export function buildDocumentLifecycleEventMethods(
getDb: () => DatabaseSync,
readAll: ReadAll,
) {
return {
enqueue(input: EnqueueInput): string {
const id = randomUUID();
const ts = nowIso();
getDb().prepare(
`INSERT INTO document_lifecycle_events
(id, company_id, document_id, hook_type, status, attempts, last_error, payload_json, created_at, updated_at, completed_at)
VALUES (?, ?, ?, ?, 'pending', 0, NULL, ?, ?, ?, NULL)`,
).run(
id,
input.companyId,
input.documentId,
input.hookType,
input.payloadJson ?? "{}",
ts,
ts,
);
return id;
},
markRunning(id: string): void {
const ts = nowIso();
getDb().prepare(
"UPDATE document_lifecycle_events SET status = 'running', attempts = attempts + 1, updated_at = ? WHERE id = ?",
).run(ts, id);
},
markSucceeded(id: string): void {
const ts = nowIso();
getDb().prepare(
"UPDATE document_lifecycle_events SET status = 'succeeded', last_error = NULL, updated_at = ?, completed_at = ? WHERE id = ?",
).run(ts, ts, id);
},
markFailed(id: string, error: string): void {
const ts = nowIso();
getDb().prepare(
"UPDATE document_lifecycle_events SET status = 'failed', last_error = ?, updated_at = ? WHERE id = ?",
).run(error.slice(0, 2000), ts, id);
},
/**
* Claim the batch of pending events for a document so the caller can
* drain them post-commit. Ordered by `created_at` so the fan-out
* reproduces the enqueue order.
*/
listPendingForDocument(documentId: string): DocumentLifecycleEventRow[] {
return readRows(readAll,
"SELECT id, company_id, document_id, hook_type, status, attempts, last_error, payload_json, created_at, updated_at, completed_at FROM document_lifecycle_events WHERE document_id = ? AND status = 'pending' ORDER BY created_at ASC",
documentId,
);
},
listFailedForCompany(companyId: string, limit = 200): DocumentLifecycleEventRow[] {
return readRows(readAll,
"SELECT id, company_id, document_id, hook_type, status, attempts, last_error, payload_json, created_at, updated_at, completed_at FROM document_lifecycle_events WHERE company_id = ? AND status = 'failed' ORDER BY updated_at DESC LIMIT ?",
companyId,
limit,
);
},
};
}
function readRows(
readAll: ReadAll,
sql: string,
...params: SQLInputValue[]
): DocumentLifecycleEventRow[] {
const raw = readAll<{
id: string;
company_id: string;
document_id: string;
hook_type: string;
status: "pending" | "running" | "succeeded" | "failed";
attempts: number;
last_error: string | null;
payload_json: string;
created_at: string;
updated_at: string;
completed_at: string | null;
}>(sql, ...params);
return raw.map((row) => ({
id: row.id,
companyId: row.company_id,
documentId: row.document_id,
hookType: row.hook_type,
status: row.status,
attempts: row.attempts,
lastError: row.last_error,
payloadJson: row.payload_json,
createdAt: row.created_at,
updatedAt: row.updated_at,
completedAt: row.completed_at,
}));
}
+53
View File
@@ -0,0 +1,53 @@
import log from "electron-log/main.js";
import type { Migration, MigrationDeps } from "./types";
export function createMigrationsV47V47(deps: MigrationDeps): Migration[] {
const { db } = deps;
return [
{
// Durable fan-out ledger for document-creation side effects.
//
// Before v47 `createDocumentWithLifecycle` wrapped every downstream
// hook (automation dispatch, thread integration, checkpoint engine,
// reviewer wake) in an individual `try/catch` and always returned
// success. A failure in any single hook left the document row live
// with a silently incomplete lifecycle — thread views, approvals,
// or reviewer wakes disappeared without trace.
//
// The ledger is the tracked-state surface the operator reads to
// reason about lifecycle coverage: one row per hook per document
// is inserted inside the same transaction as the document write.
// The processor advances `status` from `pending` to `succeeded` or
// `failed`, stamps `last_error` on failure, and `attempts` lets the
// sweeper retry within a bounded policy.
version: 47,
name: "document_lifecycle_events_ledger",
fn: () => {
db.exec(`
CREATE TABLE IF NOT EXISTS document_lifecycle_events (
id TEXT PRIMARY KEY,
company_id TEXT NOT NULL,
document_id TEXT NOT NULL,
hook_type TEXT NOT NULL,
status TEXT NOT NULL,
attempts INTEGER NOT NULL DEFAULT 0,
last_error TEXT,
payload_json TEXT NOT NULL DEFAULT '{}',
created_at TEXT NOT NULL,
updated_at TEXT NOT NULL,
completed_at TEXT
)
`);
db.exec(`
CREATE INDEX IF NOT EXISTS idx_document_lifecycle_events_company
ON document_lifecycle_events(company_id, status, updated_at)
`);
db.exec(`
CREATE INDEX IF NOT EXISTS idx_document_lifecycle_events_document
ON document_lifecycle_events(document_id, hook_type)
`);
log.info("[migration v47] document_lifecycle_events ledger installed");
},
},
];
}
+2
View File
@@ -11,6 +11,7 @@ import { createMigrationsV43V43 } from "./migrations/v43-v43";
import { createMigrationsV44V44 } from "./migrations/v44-v44";
import { createMigrationsV45V45 } from "./migrations/v45-v45";
import { createMigrationsV46V46 } from "./migrations/v46-v46";
import { createMigrationsV47V47 } from "./migrations/v47-v47";
import {
columnExists,
nowIso,
@@ -197,6 +198,7 @@ export function runMigrations(
...createMigrationsV44V44(deps),
...createMigrationsV45V45(deps),
...createMigrationsV46V46(deps),
...createMigrationsV47V47(deps),
];
for (const migration of migrations) {
+137 -117
View File
@@ -1,20 +1,23 @@
/**
* Unified document-creation lifecycle.
*
* Previous layout had two divergent entry points: the HTTP API
* (`routes-files.ts`) fired thread, checkpoint, automation, reviewer-wake,
* and deliverable-review side effects, while the desktop IPC save path
* (`ipc-content-operations-handlers.ts`) only fired automation and the
* reviewer wake. A document created from the API and one created from the
* desktop reviewer ended up in inconsistent thread-event and checkpoint
* states, so the UI showed different lifecycle progress depending on
* which surface produced the row.
* Two invariants:
*
* This module is the single owner of "what happens when a new document
* lands in the DB". Both entry points call `createDocumentWithLifecycle`
* and rely on its deterministic fan-out. Update-only edits (when `id` is
* present) skip the creation-time side effects but still publish the
* domain-change event via `publishDomainChanged`.
* 1. DB-writing hooks (thread events, checkpoint approval,
* deliverable-review approval) commit atomically with the
* document row. A failure in any of them rolls back the save so
* the operator never sees a live document with a missing thread
* or a skipped approval.
*
* 2. Cross-process side effects (automation dispatch, reviewer wake)
* are durable via the `document_lifecycle_events` ledger. Every
* hook attempt lands a row and the processor stamps its outcome;
* a crash between "document saved" and "reviewer woken" is
* observable as a `failed` ledger row instead of a silent gap.
*
* The HTTP API route and the desktop IPC handler both call
* `createDocumentWithLifecycle`, so both surfaces end up in the same
* state after a save — no more divergence between the two entry points.
*/
import log from "electron-log/main.js";
@@ -55,99 +58,132 @@ export interface DocumentLifecycleInput {
version: number;
status: string;
tagsJson: string;
/** Run that produced the save call, or null for operator-authored saves. */
sourceRunId: string | null;
}
export interface DocumentLifecycleOutcome {
id: string;
isNew: boolean;
/** Task derived from the originating run, if any — used for thread/approval binding. */
taskId: string | null;
/** Approval raised by the deliverable-review gate, when applicable. */
deliverableReviewApprovalId: string | null;
}
/**
* Save a document and dispatch every downstream side effect associated
* with its creation. Side effects are fired inside individual `try`s
* because this is the authoritative fanout — an early throw should not
* prevent later hooks from running (e.g. an automation failure must not
* suppress the thread event).
*/
export type SideEffectHook = "automation_dispatch" | "reviewer_wake";
interface PendingSideEffect {
eventId: string;
hook: SideEffectHook;
run: () => void;
}
export function createDocumentWithLifecycle(
deps: DocumentLifecycleDeps,
input: DocumentLifecycleInput,
): DocumentLifecycleOutcome {
const isNew = !input.id;
const id = deps.db.saveDocument(input);
const taskId = deriveTaskIdFromRun(deps.db, input.sourceRunId, input.companyId, deps.logger);
const pendingSideEffects: PendingSideEffect[] = [];
if (isNew) {
fireCreationHooks(deps, input, id, taskId);
}
const outcome = deps.db.withTransaction(() => {
const isNew = !input.id;
const id = deps.db.saveDocument(input);
const taskId = deriveTaskIdFromRun(deps.db, input.sourceRunId, input.companyId, deps.logger);
let deliverableReviewApprovalId: string | null = null;
if (isNew && IMPORTANT_DOC_TYPE_SET.has(input.type)) {
deliverableReviewApprovalId = maybeRaiseDeliverableReviewApproval(deps, input, id, taskId);
let deliverableReviewApprovalId: string | null = null;
if (isNew) {
// DB-writing hooks run inside the transaction. Any throw rolls
// back `saveDocument` along with the thread event / approvals.
deps.threadIntegration.onDocumentCreated(
input.companyId,
input.authorAgentId,
taskId,
input.title,
);
deps.checkpointEngine.onDocumentDrafted(
input.companyId,
id,
input.title,
input.type,
input.authorAgentId,
taskId,
);
if (IMPORTANT_DOC_TYPE_SET.has(input.type)) {
deliverableReviewApprovalId = maybeRaiseDeliverableReviewApproval(deps, input, id, taskId);
}
// Side effects (cross-process / advisory) land in the ledger so
// a post-commit failure is observable instead of a silent gap.
const automationEventId = deps.db.documentLifecycleEvents.enqueue({
companyId: input.companyId,
documentId: id,
hookType: "automation_dispatch",
payloadJson: JSON.stringify({
trigger: "document_created",
agentId: input.authorAgentId,
projectId: input.projectId,
}),
});
pendingSideEffects.push({
eventId: automationEventId,
hook: "automation_dispatch",
run: () => deps.dispatchAutomation("document_created", {
companyId: input.companyId,
agentId: input.authorAgentId ?? undefined,
projectId: input.projectId ?? undefined,
}),
});
if (input.reviewerAgentId && input.reviewerAgentId !== input.authorAgentId) {
const wakeEventId = deps.db.documentLifecycleEvents.enqueue({
companyId: input.companyId,
documentId: id,
hookType: "reviewer_wake",
payloadJson: JSON.stringify({ reviewerAgentId: input.reviewerAgentId }),
});
const reviewerAgentId = input.reviewerAgentId;
const companyId = input.companyId;
pendingSideEffects.push({
eventId: wakeEventId,
hook: "reviewer_wake",
run: () => {
deps.wakeAgentIfPossible(reviewerAgentId, companyId, "assignment");
},
});
}
}
return { id, isNew, taskId, deliverableReviewApprovalId };
});
for (const pending of pendingSideEffects) {
runSideEffect(deps, pending);
}
deps.publishDomainChanged();
return { id, isNew, taskId, deliverableReviewApprovalId };
return outcome;
}
function fireCreationHooks(
deps: DocumentLifecycleDeps,
input: DocumentLifecycleInput,
documentId: string,
taskId: string | null,
): void {
function runSideEffect(deps: DocumentLifecycleDeps, pending: PendingSideEffect): void {
try {
deps.dispatchAutomation("document_created", {
companyId: input.companyId,
agentId: input.authorAgentId ?? undefined,
projectId: input.projectId ?? undefined,
});
deps.db.documentLifecycleEvents.markRunning(pending.eventId);
} catch (error) {
deps.logger.warn(
`[document-lifecycle] dispatchAutomation failed for document ${documentId}: ${error instanceof Error ? error.message : String(error)}`,
`[document-lifecycle] markRunning failed for ${pending.eventId}: ${error instanceof Error ? error.message : String(error)}`,
);
return;
}
try {
deps.threadIntegration.onDocumentCreated(
input.companyId,
input.authorAgentId,
taskId,
input.title,
);
pending.run();
deps.db.documentLifecycleEvents.markSucceeded(pending.eventId);
} catch (error) {
deps.logger.warn(
`[document-lifecycle] threadIntegration.onDocumentCreated failed for ${documentId}: ${error instanceof Error ? error.message : String(error)}`,
const message = error instanceof Error ? error.message : String(error);
deps.logger.error(
`[document-lifecycle] hook ${pending.hook} failed for event ${pending.eventId}: ${message}`,
);
}
try {
deps.checkpointEngine.onDocumentDrafted(
input.companyId,
documentId,
input.title,
input.type,
input.authorAgentId,
taskId,
);
} catch (error) {
deps.logger.warn(
`[document-lifecycle] checkpointEngine.onDocumentDrafted failed for ${documentId}: ${error instanceof Error ? error.message : String(error)}`,
);
}
if (input.reviewerAgentId && input.reviewerAgentId !== input.authorAgentId) {
try {
deps.wakeAgentIfPossible(input.reviewerAgentId, input.companyId, "assignment");
} catch (error) {
deps.db.documentLifecycleEvents.markFailed(pending.eventId, message);
} catch (markError) {
deps.logger.warn(
`[document-lifecycle] reviewer wake failed for ${documentId}: ${error instanceof Error ? error.message : String(error)}`,
`[document-lifecycle] markFailed failed for ${pending.eventId}: ${markError instanceof Error ? markError.message : String(markError)}`,
);
}
}
@@ -172,60 +208,44 @@ function deriveTaskIdFromRun(
}
}
/**
* Raises a `deliverable_review` approval when the company has opted in,
* the document type is important, execution mode is `ai_ceo` (cofounder
* mode handles reviews via the checkpoint engine), and no pending
* deliverable review already exists. Mirrors the gate that used to live
* inline in `routes-files.ts`; consolidating it here ensures operator-
* created documents get the same reviewer treatment.
*/
function maybeRaiseDeliverableReviewApproval(
deps: DocumentLifecycleDeps,
input: DocumentLifecycleInput,
documentId: string,
taskId: string | null,
): string | null {
try {
const companySettings = deps.db.getCompanyApprovalSettings(input.companyId);
if (!companySettings?.reviewDeliverables) return null;
if (deps.db.getCompanyExecutionMode(input.companyId) === "ai_cofounder") return null;
if (deps.db.hasPendingDeliverableReviews(input.companyId)) return null;
const companySettings = deps.db.getCompanyApprovalSettings(input.companyId);
if (!companySettings?.reviewDeliverables) return null;
if (deps.db.getCompanyExecutionMode(input.companyId) === "ai_cofounder") return null;
if (deps.db.hasPendingDeliverableReviews(input.companyId)) return null;
const approvalId = deps.db.requestApproval({
companyId: input.companyId,
relatedTaskId: taskId,
requestedByAgentId: input.authorAgentId ?? "",
type: "deliverable_review",
payloadSummary: `Review deliverable: ${input.title}`.slice(0, 280),
impactSummary: `New ${input.type.replaceAll("_", " ")} created. Awaiting your review before workflow continues.`.slice(0, 400),
payloadJson: JSON.stringify({
documentIds: [documentId],
documents: [{ id: documentId, type: input.type, title: input.title }],
}),
});
const approvalId = deps.db.requestApproval({
companyId: input.companyId,
relatedTaskId: taskId,
requestedByAgentId: input.authorAgentId ?? "",
type: "deliverable_review",
payloadSummary: `Review deliverable: ${input.title}`.slice(0, 280),
impactSummary: `New ${input.type.replaceAll("_", " ")} created. Awaiting your review before workflow continues.`.slice(0, 400),
payloadJson: JSON.stringify({
documentIds: [documentId],
documents: [{ id: documentId, type: input.type, title: input.title }],
}),
});
if (input.authorAgentId) {
try {
const agent = deps.db.getAgent(input.authorAgentId);
if (agent.heartbeatEnabled) {
deps.db.setHeartbeat(input.authorAgentId, false, agent.heartbeatIntervalSec || 120);
}
} catch (error) {
deps.logger.warn(
`[document-lifecycle] heartbeat pause failed for agent ${input.authorAgentId}: ${error instanceof Error ? error.message : String(error)}`,
);
if (input.authorAgentId) {
try {
const agent = deps.db.getAgent(input.authorAgentId);
if (agent.heartbeatEnabled) {
deps.db.setHeartbeat(input.authorAgentId, false, agent.heartbeatIntervalSec || 120);
}
} catch (error) {
deps.logger.warn(
`[document-lifecycle] heartbeat pause failed for agent ${input.authorAgentId}: ${error instanceof Error ? error.message : String(error)}`,
);
}
return approvalId;
} catch (error) {
deps.logger.warn(
`[document-lifecycle] deliverable-review gate failed for document ${documentId}: ${error instanceof Error ? error.message : String(error)}`,
);
return null;
}
return approvalId;
}
/** Re-export for backwards-compatible logging in callers that do not have electron-log wired. */
export const defaultLogger: DocumentLifecycleDeps["logger"] = log;