Files
PaperClipAI/server/src/services/delivery-work-coordinator.ts
T
DottaandPaperclip 006d5691a6 feat(server): support volume-free web server handoff (#18518)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work.
> - Its web server also owns controller admission and user commands.
> - A rolling deployment starts a replacement before the old server
exits.
> - Shared database state alone does not preserve local agent files,
browser chunks, or in-process commands.
> - This pull request adds an opt-in ephemeral server contract and a
local overlap test.
> - The benefit is a web server that can relinquish traffic and local
files without losing acknowledged work.

## Linked Issues or Issue Description

Refs #18403, #18327, #18335, #18341.

**Current behavior**

The control-plane web server still has local agent-file authority and
local command targets. A new container can miss saved bytes or send a
command to the wrong process. An existing browser tab also needs chunks
from its original build.

**Proposed behavior**

Eligible servers share only PostgreSQL, object storage and stable
secrets. A replacement waits for readiness. The old server drains
admitted requests, keeps owner commands available through handoff, and
releases controller leases. Explicit maintenance adopts existing files
before the volume is removed.

**Reason and benefit**

This is the application prerequisite for rolling app-only deployments.
Schema migrations and permanent multi-replica operation are outside this
change.

## What Changed

- Store agent-file manifests in PostgreSQL and immutable bytes in S3.
Treat local roots as disposable projections. Preserve existing
permission checks and edit conflicts.
- Add an explicit migration, eligibility and object-retention command to
the production runtime bundle.
- Add `/api/ready`, admitted-request drain, browser restart signaling
and retained immutable UI chunks.
- Carry transaction-aware work hints across controllers during handoff
and active execution. Keep listeners closed when idle.
- Deliver durable Stop and runtime responses to the execution owner.
Recheck request lifecycle and access before response dispatch.
- Reject local-only configuration in ephemeral mode. Keep the feature
disabled for existing installations by default.
- Add an isolated PostgreSQL/S3/proxy/browser deployment harness with
independent server homes and retained failure evidence.

## Verification

- All CI checks pass on `ce17b95019d3cf1cabd357d19b6d37f96782cc5c`:
workspace typecheck, build, Runner checks, server/general tests and
browser shards. Greptile reviewed the runtime changes at 5/5 with all
findings resolved; the final commit updates maps and the plan.
- Local targeted checks passed: coordinator 27; Stop/recovery 33;
PostgreSQL command delivery 6; route/workspace regressions 8; bootstrap
Stop 1; restart recovery 75. PostgreSQL command tests simulate the
provider edge.
- Local browser proof passed 91 reads and 19 saves; reconnect took
1,013ms. The tab opened Audit from S3 after deletion of A's whole
home/build. This uses compiled workspace code and local-trusted auth,
not released containers or live agents.
- Local workspace typecheck/build passed; current server and runtime
bundle passed. The earlier broad local suite was stopped after three
hours and is not a passing receipt. Its retained failures and isolated
reruns are documented in the plan.
- Stacked #18526 adds stricter graceful-exit proof, an unready
replacement control, live Runner qualification and a Stop handoff
correction. Hosted enablement remains gated on those results and Cloud
#1096. No hosted stack has been enabled.

## Risks

- Apply migration 0422 before enabling the feature. Existing data must
pass each owner's migration and eligibility checks. Do not discard a
volume based on a single migration page.
- Durable manifests remain authoritative after the opt-in is disabled.
Rolling back to a binary that ignores them is unsupported.
- Cold directory restoration and complete shutdown timing still need
qualification before enabling a hosted stack.
- Retained UI assets must cover the supported browser and rollback
window. Cleanup is explicit; no new idle maintenance polling is
introduced.
- This change does not claim production zero downtime or authorize a
fleet-wide enablement.

## Model Used

OpenAI Codex, GPT-6 family. The exact serving model ID and
context-window limit are not exposed in this session. Used reasoning,
code editing, shell execution and browser automation.

## Checklist

- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge

---------

Co-authored-by: Paperclip <noreply@paperclip.ing>
2026-11-01 01:58:54 -06:00

144 lines
6.8 KiB
TypeScript

import { subscribeDatabaseWork, databaseWorkPending, reconcileDatabaseWork } from "@paperclipai/db/work-signals";
import { createWorkScheduler } from "./work-scheduler.js";
import { type DeliveryQueue } from "./delivery-work-notifications.js";
import { beginIdleTrackedWork } from "./task-admission.js";
export function createDeliveryWorkCoordinator(input: {
owner: object;
canRun: () => boolean;
// Idle drain may reconcile admitted writes without dispatching new work.
// Warm standby can disable both operations.
canReconcile?: () => boolean;
onError: (error: unknown, queue: DeliveryQueue) => void;
}) {
const scheduler = createWorkScheduler();
const workers = new Map<DeliveryQueue, { wake: () => void; stop: () => Promise<void> }>();
let stopped = false;
return {
// Only these registered tasks are represented; not yet the whole instance.
nextWakeAt: scheduler.nextWakeAt,
register(queue: DeliveryQueue, task: {
retryMs: number;
run: (signal: AbortSignal) => Promise<unknown>;
hasPending: () => Promise<boolean>;
// Scoped hints belong to the outer transaction, including uncertainty recovery.
onSettled?: (hints?: ReadonlySet<string>) => void;
// Optional maintenance/retry deadline restored from durable state.
nextRunAt?: () => number | null;
// Opt in only when run has recorded per-item failure backoffs. Other
// eligible items may already be due, unlike an unchanged failed deadline.
nextRunAtOnError?: () => number | null;
// An empty queue's maintenance may wait until dispatch admission reopens.
// Notifications and unresolved producer writes still reconcile in drain.
pausePassiveMaintenance?: boolean;
// Accounting and stop delivery finish already-admitted obligations;
// they must continue during admission drain, never warm standby.
runDuringDrain?: boolean;
}) {
if (stopped) throw new Error("Delivery coordinator is stopped");
if (workers.has(queue)) throw new Error(`Delivery worker already registered: ${queue}`);
let dirty = false;
let scheduledAt: number | null = null;
let running: Promise<void> | null = null;
let workerStopped = false;
let controller: AbortController | null = null;
let finishIntent: (() => void) | undefined;
let pendingWork = true;
function schedule(delay: number, deadline?: number) {
if (workerStopped) return;
const at = deadline ?? Date.now() + delay;
if (scheduledAt !== null && scheduledAt <= at) return;
scheduledAt = at;
scheduler.schedule(queue, at, start);
}
function start() {
if (workerStopped || running) return;
scheduler.cancel(queue);
scheduledAt = null;
if (!input.canRun() && !(input.canReconcile?.() ?? false)) { schedule(task.retryMs); return; }
if (!input.canRun() && task.pausePassiveMaintenance && !pendingWork && !dirty &&
!finishIntent && !databaseWorkPending(input.owner, queue)) {
// A timer alone must not change the idle generation during a slow
// backup. Retry without querying or admitting work; a notification
// still preempts this deadline and retains its normal intent hold.
schedule(task.retryMs);
return;
}
dirty = false;
const finish = beginIdleTrackedWork();
const attempt = new AbortController();
let runFailed = false;
controller = attempt;
running = Promise.resolve().then(() => reconcileDatabaseWork(input.owner, queue))
.then(() => workerStopped || !(input.canRun() || (task.runDuringDrain && input.canReconcile?.())) ? undefined : task.run(attempt.signal).catch(error => { runFailed = true; throw error; })).then(() => workerStopped ? false : task.hasPending()).then(pending => {
pendingWork = pending || databaseWorkPending(input.owner, queue);
if (pendingWork) schedule(task.retryMs);
else { finishIntent?.(); finishIntent = undefined; }
const next = task.nextRunAt?.();
if (next != null) {
const delay = Math.max(0, next - Date.now());
schedule(delay === 0 && !input.canRun() ? task.retryMs : delay, delay > 0 ? next : undefined);
}
}).catch(error => {
schedule(task.retryMs);
let failure = error;
try {
const scopedRetry = runFailed && task.nextRunAtOnError;
const next = input.canRun() ? (scopedRetry ? scopedRetry() : task.nextRunAt?.()) : null;
if (next != null && (scopedRetry || next > Date.now())) schedule(0, Math.max(Date.now(), next));
} catch (deadlineError) {
failure = new AggregateError([error, deadlineError], "Delivery failure deadline could not be read");
}
// Logging must never turn a recoverable sweep failure into an
// unhandled rejection or prevent other queues from running.
try { if (!workerStopped) input.onError(failure, queue); } catch { /* Recovery remains scheduled. */ }
}).finally(() => {
running = null;
controller = null;
finish();
if (dirty) schedule(0);
});
}
function wake() {
if (workerStopped) return;
dirty = true;
if (!running) schedule(0);
}
const unsubscribe = subscribeDatabaseWork(input.owner, queue, (event, hints) => {
if (workerStopped) return;
if (databaseWorkPending(input.owner, queue)) {
finishIntent ??= beginIdleTrackedWork();
// Give the transaction time to settle. Its settlement notification
// replaces this deadline with an immediate wake on the normal path.
if (!running) schedule(task.retryMs);
}
if (event === "settled") { task.onSettled?.(hints); wake(); }
});
const worker = { wake, async stop() {
workerStopped = true;
controller?.abort(new Error("Delivery worker stopped"));
unsubscribe();
scheduler.cancel(queue);
await running;
finishIntent?.(); finishIntent = undefined;
} };
workers.set(queue, worker);
const ready = Promise.resolve().then(() => { start(); return running; });
return { ready, ...worker };
},
// Stop dispatch queues first, keeping only already-admitted controller
// commands alive until their execution owners finish the handoff. A later
// stop() closes the retained workers and the shared scheduler as well.
async stop(keepQueues?: ReadonlySet<DeliveryQueue>) {
stopped = true;
if (!keepQueues?.size) scheduler.stop();
await Promise.all(Array.from(workers, async ([queue, worker]) => {
if (keepQueues?.has(queue)) return;
await worker.stop();
workers.delete(queue);
}));
},
};
}