fix: reclaim queued Pi parent capacity

This commit is contained in:
2026-08-23 13:53:28 +08:00
parent a10b98e484
commit 1727b75f30
3 changed files with 171 additions and 17 deletions

View File

@@ -8,7 +8,7 @@
- Worktree: D:\Datas\OthersProjects\makelore-pi-child-workers-5c8e2a71
- Base commit: b806c78139aa11e338c680d4c3fa92e076901aa2
- Owner: codex
- Status: Second planner corrections complete; re-review pending
- Status: Third planner correction complete; final re-review pending
## Scope
@@ -77,6 +77,16 @@
reservations race direct capacity against this cancellable event-driven
reclaim and always settle/release on reclaim rejection, parent abort, or any
other early exit. Parent `stop()` now releases its process lease in `finally`.
- Planner re-review of `a10b98e` confirmed both lease-lifecycle findings closed,
then reproduced the production-capacity shape `4 running + 4 queued`: queued
parent workers retained all eight process leases, leaving no ready/idle
worker for four children to reclaim. The pool now prefers ready/idle
eviction, then suspends the oldest queued parent process while retaining its
top-level FIFO entry, Conversation state, and session binding. When its turn
arrives, the same queue entry reopens that session as a new worker generation
before sending the original prompt. This closes the deadlock without
dropping queued work, reordering it, adding another queue, or exceeding the
shared eight-process cap.
- Added the managed ephemeral child opener. It resolves only enabled,
unarchived project Agents from `.niancode/project.json`, materializes their
exact model/prompt/skills, keeps credentials in the child environment and
@@ -113,7 +123,10 @@
unknown schema/version raw-payload suppression; first reclaim finding no
idle followed by delayed parent readiness and automatic child execution;
reclaim rejection/abort with zero queued lease; and stop rejection with zero
active/waiting lease.
active/waiting lease. A production-scale regression also covers four running
plus four queued parents dispatching four parallel children: all four queued
workers suspend, all children complete, and the queued parents then resume
FIFO with unchanged session bindings and generation 2.
- Locked real Pi 0.84.2 workspace smoke passed for both the parent worker and
an ephemeral read-only child launched through Electron Node with the child
extension role and `--no-session`. A probe extension reads Pi's actual
@@ -129,17 +142,18 @@
authenticated Main bridge; the active-tool process probes and bridge execute
test jointly cover visibility and tool invocation without an external
Provider.
- All cumulative Pi tests passed: 22 files, 110 passed and 1 staged-only
- All cumulative Pi tests passed: 22 files, 111 passed and 1 staged-only
skipped; the staged-only command passed separately as described above.
- `pnpm run typecheck`: passed.
- `pnpm run lint:check`: passed with 0 errors and 6 pre-existing frontend
warnings outside this task.
- `pnpm run build:vite`: passed for Renderer, Main, Preload, and utility
worker bundles; existing chunk-size/dynamic-import warnings remain.
- The final full-suite first pass hit the known Windows temporary JSON `rename`
`EPERM` in the unchanged conversation store. The failing runtime test passed
1/1 in isolation, and the single full-suite rerun passed: 202 files, 2213
passed and 1 staged-only skipped.
- An earlier full-suite pass hit the known Windows temporary JSON `rename`
`EPERM` in the unchanged conversation store; the failing runtime test passed
1/1 in isolation and the rerun passed. After the queued-parent correction,
the final full suite passed on its first run: 202 files, 2214 passed and 1
staged-only skipped.
- Real external Provider validation remains **Explicitly Waived / Accepted
Risk** with `realTurnVerified=false`. Provider concurrency, credential
isolation, protocol compatibility, and image-path risk are accepted rather
@@ -166,7 +180,10 @@
reservations ahead of normal parent-start waiters, and wire the scheduler's
cancellable capacity reclaimer to `PiWorkerPool.reclaimIdleWorker`. The pool
must notify an already waiting child when a spawning/running parent becomes
`ready`/`idle`, and stop/reclaim failure must never retain a lease. Semantic
conflicts: none with the accepted PI runtime specification; this makes its
parent/child cap executable. Human confirmation required: no, unless
integration changes the accepted process-cap policy.
`ready`/`idle`, and stop/reclaim failure must never retain a lease. If the
cap consists only of running and queued parents, it must suspend queued
parent processes in FIFO-safe/LRU order, retain their queue entries and
session bindings, and reopen them as a new generation when scheduled.
Semantic conflicts: none with the accepted PI runtime specification; this
makes its parent/child cap executable. Human confirmation required: no,
unless integration changes the accepted process-cap policy.

View File

@@ -351,11 +351,20 @@ export class PiWorkerPool {
while (true) {
if (signal?.aborted) throw new Error('Pi idle worker reclaim cancelled');
if (this.shuttingDown) throw new Error('Pi worker pool is shutting down');
const record = [...this.workers.values()]
const idleRecord = [...this.workers.values()]
.filter((candidate) => candidate.state === 'ready' || candidate.state === 'idle')
.sort((left, right) => left.lastUsed - right.lastUsed)[0];
if (record) {
if (await this.evict(record)) return true;
if (idleRecord) {
if (await this.evict(idleRecord)) return true;
continue;
}
const queuedRecord = [...this.workers.values()]
.filter((candidate) => candidate.state === 'queued'
&& candidate.processLease !== null
&& !candidate.processStopFlight)
.sort((left, right) => left.lastUsed - right.lastUsed)[0];
if (queuedRecord) {
if (await this.suspendQueuedWorker(queuedRecord)) return true;
continue;
}
await this.waitForReclaimableWorker(signal);
@@ -501,6 +510,7 @@ export class PiWorkerPool {
this.waitingRuns.push(pending);
pending.queuedAt = this.now();
record.state = 'queued';
this.notifyReclaimableWorker();
return { queuePosition: this.waitingRuns.length, accepted };
}
@@ -550,7 +560,7 @@ export class PiWorkerPool {
this.cancelGenerationResources(record);
this.revisions.removeWorker(record.revisionWorkerId);
if (this.workers.get(conversationId) === record) this.workers.delete(conversationId);
await this.stopAndRelease(record);
await this.ensureStoppedAndReleased(record);
}
private async performShutdown(): Promise<void> {
@@ -569,7 +579,7 @@ export class PiWorkerPool {
record.unsubscribeInvalidation();
this.cancelGenerationResources(record);
this.revisions.removeWorker(record.revisionWorkerId);
await this.stopAndRelease(record);
await this.ensureStoppedAndReleased(record);
}));
}
@@ -651,6 +661,10 @@ export class PiWorkerPool {
if (!current) throw new Error('Conversation worker is no longer available');
if (current !== record) return await this.ensureFresh(current);
if (record.rebuildFlight) return await record.rebuildFlight;
if (!record.processLease) {
record.rebuildFlight = this.beginRebuild(record, this.revisions.current);
return await record.rebuildFlight;
}
const action = this.revisions.beforePrompt(record.revisionWorkerId);
if (action.action !== 'rebuild-before-prompt') return record;
record.rebuildFlight = this.beginRebuild(record, action.revision);
@@ -817,7 +831,23 @@ export class PiWorkerPool {
this.cancelGenerationResources(record);
this.revisions.removeWorker(record.revisionWorkerId);
this.workers.delete(conversationId);
await this.stopAndRelease(record);
await this.ensureStoppedAndReleased(record);
return true;
}
private async suspendQueuedWorker(record: WorkerRecord): Promise<boolean> {
const conversationId = record.conversation.conversationId;
if (this.workers.get(conversationId) !== record
|| record.state !== 'queued'
|| !record.processLease
|| record.processStopFlight) {
return false;
}
record.unsubscribeEvent();
record.unsubscribeInvalidation();
this.cancelGenerationResources(record);
record.processStopFlight = this.stopAndRelease(record);
await record.processStopFlight;
return true;
}
@@ -924,6 +954,13 @@ export class PiWorkerPool {
}
}
private async ensureStoppedAndReleased(record: WorkerRecord): Promise<void> {
if (!record.processStopFlight) {
record.processStopFlight = this.stopAndRelease(record);
}
await record.processStopFlight;
}
private waitForReclaimableWorker(signal?: AbortSignal): Promise<void> {
return new Promise((resolve, reject) => {
const cleanup = () => {

View File

@@ -12,6 +12,7 @@ import type { PiProcessError } from '../../electron/coding-runtime/pi/process-er
import { PiProcessError as PiProcessFailure } from '../../electron/coding-runtime/pi/process-errors';
import type { PiRpcCommand, PiRpcEvent } from '../../electron/coding-runtime/pi/rpc-client';
import type { PiRuntimeTelemetryEvent } from '../../electron/coding-runtime/pi/telemetry';
import { PiSubagentScheduler } from '../../electron/coding-runtime/pi/subagent';
const MODEL = {
model: {
@@ -254,6 +255,105 @@ describe('Pi worker pool', () => {
expect(processBudget.waitingCount).toBe(0);
});
it('suspends and later resumes the oldest queued parent to make child capacity', async () => {
const processBudget = new PiProcessBudget(8);
const workers = new Map<string, FakeWorker[]>();
const pool = new PiWorkerPool({
processBudget,
maxRunning: 4,
maxIdle: 8,
openWorker: async ({ conversation: input, generation, existingSession }) => {
const worker = new FakeWorker(`worker-${input.conversationId}-${generation}`);
workers.set(input.conversationId, [...(workers.get(input.conversationId) ?? []), worker]);
return {
worker,
session: existingSession ?? {
piSessionId: `session-${input.conversationId}`,
sessionKey: `key-${input.conversationId}`,
},
};
},
});
const ids = Array.from({ length: 8 }, (_, index) => `conversation-${index + 1}`);
for (const id of ids) await pool.prepare(conversation(id));
const tickets = ids.map((conversationId, index) => pool.startTopLevel({
conversationId,
runId: `run-${index + 1}`,
command: { type: 'prompt', message: conversationId },
}));
for (const ticket of tickets) void ticket.accepted.catch(() => undefined);
await expect.poll(() => ids.map((id) => workers.get(id)?.[0]?.requests.length ?? 0))
.toEqual([1, 1, 1, 1, 0, 0, 0, 0]);
expect(ids.slice(4).map((id) => pool.getState(id)?.state))
.toEqual(['queued', 'queued', 'queued', 'queued']);
expect(processBudget.activeCount).toBe(8);
const scheduler = new PiSubagentScheduler({
processBudget,
reclaimProcessCapacity: (signal) => pool.reclaimIdleWorker(signal),
openChild: async (input) => ({
id: input.taskId,
async run() { return { summary: 'child complete' }; },
async stop() {},
}),
});
await expect(scheduler.dispatch({
conversationId: ids[0]!,
workerGeneration: 1,
runId: 'run-child',
projectId: 'project-a',
request: {
mode: 'parallel',
tasks: Array.from({ length: 4 }, (_, index) => ({
agentId: `agent-${index + 1}`,
task: `inspect ${index + 1}`,
toolProfile: 'read-only',
})),
},
})).resolves.toMatchObject({
details: {
tasks: Array.from({ length: 4 }, () => ({ status: 'complete' })),
},
});
const firstQueuedId = ids[4]!;
expect(ids.slice(4).map((id) => workers.get(id)?.[0]?.stopped))
.toEqual([true, true, true, true]);
expect(pool.getState(firstQueuedId)).toMatchObject({
state: 'queued',
generation: 1,
session: {
piSessionId: `session-${firstQueuedId}`,
sessionKey: `key-${firstQueuedId}`,
},
});
expect(processBudget.activeCount).toBe(4);
expect(processBudget.waitingCount).toBe(0);
for (let index = 0; index < 4; index += 1) {
const queuedId = ids[index + 4]!;
workers.get(ids[index]!)?.[0]?.emit({ type: 'agent_settled' });
await expect.poll(() => workers.get(queuedId)?.length).toBe(2);
await expect.poll(() => workers.get(queuedId)?.[1]?.requests.length).toBe(1);
await expect(tickets[index + 4]!.accepted).resolves.toMatchObject({ success: true });
}
expect(pool.getState(firstQueuedId)).toMatchObject({
state: 'running',
generation: 2,
session: {
piSessionId: `session-${firstQueuedId}`,
sessionKey: `key-${firstQueuedId}`,
},
});
expect(ids.slice(4).map((id) => workers.get(id)?.length)).toEqual([2, 2, 2, 2]);
expect(processBudget.activeCount).toBe(8);
await scheduler.close();
await pool.shutdown();
expect(processBudget.activeCount).toBe(0);
expect(processBudget.waitingCount).toBe(0);
});
it('never evicts a running worker when the warm-idle LRU exceeds its cap', async () => {
const workers = new Map<string, FakeWorker>();
const pool = new PiWorkerPool({