VM 복귀 후 예약턴 잠금 복구와 Windows 부팅 식별자 확인을 추가한다
This commit is contained in:
1 parent
fd8145d6eb
commit
f36bc2c8d7
24 files changed
+1048
-28
No files matched your search
@@ -0,0 +1,134 @@
|
||||
import { createHmac, randomBytes } from 'node:crypto';
|
||||
import { once } from 'node:events';
|
||||
import { createServer } from 'node:http';
|
||||
import { afterEach, describe, expect, it, vi } from 'vitest';
|
||||
import { createHostBootMonitor, loadHostBootMonitor } from '../src/turn/hostBootMonitor.js';
|
||||
|
||||
const cleanups: Array<() => Promise<void>> = [];
|
||||
afterEach(async () => {
|
||||
await Promise.all(cleanups.splice(0).map((cleanup) => cleanup()));
|
||||
});
|
||||
|
||||
const fixture = async () => {
|
||||
const key = randomBytes(32);
|
||||
let bootId = 'a'.repeat(64);
|
||||
let replay: string | null = null;
|
||||
let mode: 'valid' | 'replay' | 'forged' | 'oversized' | 'offline' | 'hang' = 'valid';
|
||||
const requests: string[] = [];
|
||||
const server = createServer((request, response) => {
|
||||
const nonce = String(request.headers['x-sammo-challenge']);
|
||||
requests.push(nonce);
|
||||
expect(request.headers['x-sammo-signature']).toBe(
|
||||
createHmac('sha256', key).update(`sammo-host-boot-request-v1\n${nonce}`).digest('hex')
|
||||
);
|
||||
if (mode === 'hang') return;
|
||||
if (mode === 'offline') {
|
||||
response.writeHead(503).end();
|
||||
return;
|
||||
}
|
||||
if (mode === 'oversized') {
|
||||
response.end('x'.repeat(4096));
|
||||
return;
|
||||
}
|
||||
if (mode === 'replay') {
|
||||
response.end(replay);
|
||||
return;
|
||||
}
|
||||
const body = JSON.stringify({
|
||||
version: 1,
|
||||
challenge: nonce,
|
||||
bootId,
|
||||
signature:
|
||||
mode === 'forged'
|
||||
? '0'.repeat(64)
|
||||
: createHmac('sha256', key)
|
||||
.update(`sammo-host-boot-response-v1\n${nonce}\n${bootId}`)
|
||||
.digest('hex'),
|
||||
});
|
||||
replay ??= body;
|
||||
response.end(body);
|
||||
});
|
||||
server.listen(0, '127.0.0.1');
|
||||
await once(server, 'listening');
|
||||
cleanups.push(
|
||||
() =>
|
||||
new Promise<void>((resolve) => {
|
||||
server.close(() => resolve());
|
||||
server.closeAllConnections();
|
||||
})
|
||||
);
|
||||
const address = server.address();
|
||||
if (!address || typeof address === 'string') throw new Error('Missing fixture listener');
|
||||
return {
|
||||
key,
|
||||
requests,
|
||||
url: `http://127.0.0.1:${address.port}/v1/boot-id`,
|
||||
setBoot: (value: string) => {
|
||||
bootId = value;
|
||||
},
|
||||
setMode: (value: typeof mode) => {
|
||||
mode = value;
|
||||
},
|
||||
};
|
||||
};
|
||||
|
||||
describe('Windows boot observation', () => {
|
||||
it('uses a stable baseline, survives outages, and latches a signed boot change', async () => {
|
||||
const f = await fixture();
|
||||
const monitor = createHostBootMonitor({ ...f, pollMs: 0 });
|
||||
expect(await monitor.hasChanged()).toBe(false);
|
||||
f.setMode('offline');
|
||||
expect(await monitor.hasChanged()).toBe(false);
|
||||
f.setBoot('b'.repeat(64));
|
||||
f.setMode('valid');
|
||||
expect(await monitor.hasChanged()).toBe(true);
|
||||
f.setMode('offline');
|
||||
expect(await monitor.hasChanged()).toBe(true);
|
||||
});
|
||||
|
||||
it.each(['replay', 'forged', 'oversized', 'hang'] as const)(
|
||||
'rejects %s without changing the baseline',
|
||||
async (mode) => {
|
||||
const f = await fixture();
|
||||
const onUnavailable = vi.fn();
|
||||
const monitor = createHostBootMonitor({ ...f, pollMs: 0, timeoutMs: 100, onUnavailable });
|
||||
expect(await monitor.hasChanged()).toBe(false);
|
||||
f.setBoot('b'.repeat(64));
|
||||
f.setMode(mode);
|
||||
expect(await monitor.hasChanged()).toBe(false);
|
||||
expect(onUnavailable).toHaveBeenCalledOnce();
|
||||
f.setMode('valid');
|
||||
expect(await monitor.hasChanged()).toBe(true);
|
||||
}
|
||||
);
|
||||
|
||||
it('samples immediately after a wall-clock jump despite a frozen monotonic cache and coalesces requests', async () => {
|
||||
const f = await fixture();
|
||||
let wall = 10_000;
|
||||
const monitor = createHostBootMonitor({ ...f, now: () => 0, wallNow: () => wall });
|
||||
expect(await monitor.hasChanged()).toBe(false);
|
||||
expect(await monitor.hasChanged()).toBe(false);
|
||||
expect(f.requests).toHaveLength(1);
|
||||
wall += 300_000;
|
||||
f.setBoot('b'.repeat(64));
|
||||
expect(await Promise.all([monitor.hasChanged(), monitor.hasChanged()])).toEqual([true, true]);
|
||||
expect(f.requests).toHaveLength(2);
|
||||
expect(new Set(f.requests).size).toBe(2);
|
||||
});
|
||||
|
||||
it('does not block game startup when optional configuration or its key file is absent', async () => {
|
||||
const warning = vi.spyOn(console, 'warn').mockImplementation(() => {});
|
||||
expect(await loadHostBootMonitor({})).toBeUndefined();
|
||||
expect(await loadHostBootMonitor({ TURN_HOST_BOOT_URL: 'http://host/v1/boot-id' })).toBeUndefined();
|
||||
expect(
|
||||
await loadHostBootMonitor({
|
||||
TURN_HOST_BOOT_URL: 'http://host/v1/boot-id',
|
||||
TURN_HOST_BOOT_KEY_FILE: '/nonexistent-host-boot-key',
|
||||
})
|
||||
).toBeUndefined();
|
||||
expect(() =>
|
||||
createHostBootMonitor({ url: 'http://user:password@host/v1/boot-id', key: randomBytes(32) })
|
||||
).toThrow('without credentials');
|
||||
warning.mockRestore();
|
||||
});
|
||||
});
|
||||
@@ -4,7 +4,9 @@ import { createGamePostgresConnector } from '@sammo-ts/infra';
|
||||
import {
|
||||
InMemoryReservedTurnStore,
|
||||
ReservedTurnLeaseConflictError,
|
||||
createReservedTurnStore,
|
||||
} from '../src/turn/reservedTurnStore.js';
|
||||
import { createReservedTurnLeaseRetry } from '../src/turn/reservedTurnLeaseRetry.js';
|
||||
|
||||
const databaseUrl = process.env.RESERVED_TURN_DATABASE_URL;
|
||||
const describeIntegration = databaseUrl ? describe : describe.skip;
|
||||
@@ -132,4 +134,98 @@ describeIntegration('reserved turn daemon/API lease integration', () => {
|
||||
});
|
||||
expect(after.map((turn) => turn.actionCode)).toEqual(before.map((turn) => turn.actionCode));
|
||||
});
|
||||
|
||||
it('cleans persisted ownership after rollback loses the journal, preserves other writers, and reloads once', async () => {
|
||||
if (!connector || !databaseUrl) throw new Error('integration connector is unavailable');
|
||||
await connector.prisma.generalTurnRevision.update({
|
||||
where: { generalId: GENERAL_ID },
|
||||
data: { leaseOwner: null, leaseExpiresAt: null },
|
||||
});
|
||||
const before = await connector.prisma.generalTurn.findMany({
|
||||
where: { generalId: GENERAL_ID },
|
||||
orderBy: { turnIdx: 'asc' },
|
||||
});
|
||||
const revision = await connector.prisma.generalTurnRevision.findUniqueOrThrow({
|
||||
where: { generalId: GENERAL_ID },
|
||||
});
|
||||
const old = await createReservedTurnStore({ databaseUrl, leaseOwner: 'interrupted-daemon' });
|
||||
const savepoint = old.store.captureTransactionState();
|
||||
await old.store.prepareTurnsForExecution(GENERAL_ID);
|
||||
old.store.shiftGeneralTurns(GENERAL_ID, -1);
|
||||
old.store.restoreState(savepoint);
|
||||
expect(old.store.getQueueCounts().leasedGeneralQueues).toBe(0);
|
||||
const next = await createReservedTurnStore({ databaseUrl, leaseOwner: 'new-daemon' });
|
||||
try {
|
||||
await expect(next.store.prepareTurnsForExecution(GENERAL_ID)).rejects.toBeInstanceOf(
|
||||
ReservedTurnLeaseConflictError
|
||||
);
|
||||
await old.close();
|
||||
const released = await connector.prisma.generalTurnRevision.findUniqueOrThrow({
|
||||
where: { generalId: GENERAL_ID },
|
||||
});
|
||||
expect(released).toMatchObject({ leaseOwner: null, leaseExpiresAt: null, revision: revision.revision });
|
||||
await next.store.prepareTurnsForExecution(GENERAL_ID);
|
||||
// A delayed close of an old owner cannot clear the new owner's lease.
|
||||
const delayedCleaner = new InMemoryReservedTurnStore(connector.prisma, {
|
||||
maxGeneralTurns: 2,
|
||||
maxNationTurns: 1,
|
||||
leaseOwner: 'interrupted-daemon',
|
||||
});
|
||||
await delayedCleaner.releaseOwnedLeases();
|
||||
expect(
|
||||
await connector.prisma.generalTurnRevision.findUniqueOrThrow({ where: { generalId: GENERAL_ID } })
|
||||
).toMatchObject({ leaseOwner: 'new-daemon', revision: revision.revision });
|
||||
await next.store.releaseOwnedLeases();
|
||||
const actual = await connector.prisma.generalTurn.findMany({
|
||||
where: { generalId: GENERAL_ID },
|
||||
orderBy: { turnIdx: 'asc' },
|
||||
});
|
||||
expect(actual.map((row) => row.actionCode)).toEqual(before.map((row) => row.actionCode));
|
||||
} finally {
|
||||
await next.close();
|
||||
}
|
||||
});
|
||||
|
||||
it('waits out an orphaned lease without deleting it and reloads an intervening API edit', async () => {
|
||||
if (!connector) throw new Error('integration connector is unavailable');
|
||||
await connector.prisma.generalTurnRevision.update({
|
||||
where: { generalId: GENERAL_ID },
|
||||
data: { leaseOwner: 'killed-daemon', leaseExpiresAt: new Date(Date.now() + 300_000) },
|
||||
});
|
||||
const store = new InMemoryReservedTurnStore(connector.prisma, {
|
||||
maxGeneralTurns: 2,
|
||||
maxNationTurns: 1,
|
||||
leaseOwner: 'recovery-daemon',
|
||||
});
|
||||
await store.loadAll();
|
||||
const retry = createReservedTurnLeaseRetry({
|
||||
sleep: async () => {
|
||||
// Simulate the recorded expiry, then an authorized API edit before the next claim.
|
||||
expect(
|
||||
await connector.prisma.generalTurnRevision.findUniqueOrThrow({ where: { generalId: GENERAL_ID } })
|
||||
).toMatchObject({ leaseOwner: 'killed-daemon' });
|
||||
await connector.prisma.generalTurnRevision.update({
|
||||
where: { generalId: GENERAL_ID },
|
||||
data: { leaseExpiresAt: new Date(0) },
|
||||
});
|
||||
await connector.prisma.generalTurn.updateMany({
|
||||
where: { generalId: GENERAL_ID, turnIdx: 0 },
|
||||
data: { actionCode: 'che_수련' },
|
||||
});
|
||||
},
|
||||
});
|
||||
try {
|
||||
let failure: unknown;
|
||||
try {
|
||||
await store.prepareTurnsForExecution(GENERAL_ID);
|
||||
} catch (error) {
|
||||
failure = error;
|
||||
}
|
||||
expect(await retry.wait(failure)).toBe(true);
|
||||
await store.prepareTurnsForExecution(GENERAL_ID);
|
||||
expect(store.getGeneralTurns(GENERAL_ID)[0]?.action).toBe('che_수련');
|
||||
} finally {
|
||||
await store.releaseOwnedLeases();
|
||||
}
|
||||
});
|
||||
});
|
||||
@@ -169,6 +169,18 @@ const buildHarness = (initialRevision: RevisionRow | null = null) => {
|
||||
};
|
||||
|
||||
describe('reserved turn daemon lease', () => {
|
||||
it('cleans both persisted lease kinds even after their journal is rolled back', async () => {
|
||||
const harness = buildHarness();
|
||||
const snapshot = harness.store.captureTransactionState();
|
||||
await harness.store.prepareTurnsForExecution(7, { nationId: 3, officerLevel: 12 });
|
||||
harness.store.restoreState(snapshot);
|
||||
expect(harness.store.getQueueCounts()).toMatchObject({ leasedGeneralQueues: 0, leasedNationQueues: 0 });
|
||||
await harness.store.releaseOwnedLeases();
|
||||
expect(harness.getRevision()).toMatchObject({ leaseOwner: null, leaseExpiresAt: null, revision: 0 });
|
||||
expect(harness.getNationRevision()).toMatchObject({ leaseOwner: null, leaseExpiresAt: null, revision: 0 });
|
||||
expect(harness.generalDeleteMany).not.toHaveBeenCalled();
|
||||
});
|
||||
|
||||
it('prunes deleted general and nation queues together with their journal state', () => {
|
||||
const harness = buildHarness();
|
||||
harness.store.ensureGeneralTurns(7);
|
||||
|
||||
@@ -0,0 +1,38 @@
|
||||
import { describe, expect, it, vi } from 'vitest';
|
||||
import { createReservedTurnLeaseRetry } from '../src/turn/reservedTurnLeaseRetry.js';
|
||||
import { ReservedTurnLeaseBusyError, ReservedTurnLeaseConflictError } from '../src/turn/reservedTurnStore.js';
|
||||
|
||||
describe('reserved turn acquisition retry', () => {
|
||||
it('bounds the entire contention episode across different queues and resets after progress', async () => {
|
||||
let now = 0;
|
||||
const onWait = vi.fn();
|
||||
const retry = createReservedTurnLeaseRetry({
|
||||
now: () => now,
|
||||
sleep: async (ms) => {
|
||||
now += ms;
|
||||
},
|
||||
maxWaitMs: 2_500,
|
||||
onWait,
|
||||
});
|
||||
expect(await retry.wait(new ReservedTurnLeaseBusyError('general:1'))).toBe(true);
|
||||
expect(await retry.wait(new ReservedTurnLeaseBusyError('nation:2:12'))).toBe(true);
|
||||
expect(await retry.wait(new ReservedTurnLeaseBusyError('general:3'))).toBe(true);
|
||||
expect(now).toBe(2_500);
|
||||
expect(await retry.wait(new ReservedTurnLeaseBusyError('general:1'))).toBe(false);
|
||||
expect(onWait).toHaveBeenCalledTimes(1);
|
||||
retry.reset();
|
||||
expect(await retry.wait(new ReservedTurnLeaseBusyError('general:1'))).toBe(true);
|
||||
});
|
||||
|
||||
it('never retries flush ownership loss, calculation errors, or failed savepoint restoration', async () => {
|
||||
const sleep = vi.fn();
|
||||
const retry = createReservedTurnLeaseRetry({ sleep });
|
||||
for (const error of [
|
||||
new ReservedTurnLeaseConflictError('general:1'),
|
||||
new Error('calculation failed'),
|
||||
new AggregateError([new ReservedTurnLeaseBusyError('general:1')], 'restore failed'),
|
||||
])
|
||||
expect(await retry.wait(error)).toBe(false);
|
||||
expect(sleep).not.toHaveBeenCalled();
|
||||
});
|
||||
});
|
||||
@@ -3,12 +3,36 @@ import type { GameClockPhase } from '@sammo-ts/common';
|
||||
import { createRuntimePauseGate } from '../src/turn/runtimePauseGate.js';
|
||||
|
||||
describe('runtime pause clock boundary', () => {
|
||||
it('awaits a host authority check before reading profile state or changing the clock', async () => {
|
||||
const calls: string[] = [];
|
||||
const gate = createRuntimePauseGate({
|
||||
assertLease: async () => {
|
||||
await Promise.resolve();
|
||||
calls.push('host');
|
||||
throw new Error('host changed');
|
||||
},
|
||||
shouldPause: async () => {
|
||||
calls.push('profile');
|
||||
return false;
|
||||
},
|
||||
isExplicitlyPaused: () => false,
|
||||
getPhase: () => 'RUNNING',
|
||||
prepareRecovery: async () => {
|
||||
calls.push('recovery');
|
||||
},
|
||||
synchronize: async () => {},
|
||||
});
|
||||
await expect(gate()).rejects.toThrow('host changed');
|
||||
expect(calls).toEqual(['host']);
|
||||
});
|
||||
it('freezes a live error pause before commands, resumes without restart, and does not freeze twice', async () => {
|
||||
let phase: GameClockPhase = 'RUNNING';
|
||||
let paused = true;
|
||||
const calls: string[] = [];
|
||||
const gate = createRuntimePauseGate({
|
||||
assertLease: () => calls.push('lease'),
|
||||
assertLease: () => {
|
||||
calls.push('lease');
|
||||
},
|
||||
shouldPause: async () => paused,
|
||||
isExplicitlyPaused: () => paused,
|
||||
getPhase: () => phase,
|
||||
|
||||
@@ -207,6 +207,39 @@ integration('database turn daemon lease and fencing', () => {
|
||||
expect(row.ownerId).toBe(firstToken ? 'owner-a' : 'owner-b');
|
||||
});
|
||||
|
||||
it('fences a host boot change inside a transaction before mutation, without expiring the DB lease', async () => {
|
||||
const profile = `${profilePrefix}host-boot`;
|
||||
let changed = false;
|
||||
const lease = await DatabaseTurnDaemonLease.connect(databaseUrl!, {
|
||||
profile,
|
||||
ownerId: 'saved-vm-owner',
|
||||
heartbeat: false,
|
||||
leaseDurationMs: 60_000,
|
||||
hostBootMonitor: { hasChanged: async () => changed },
|
||||
});
|
||||
leases.push(lease);
|
||||
await lease.acquire();
|
||||
await lease.markClockReady();
|
||||
changed = true;
|
||||
let committed = false;
|
||||
await expect(
|
||||
db.$transaction(async (transaction) => {
|
||||
await lease.assertActive(transaction);
|
||||
committed = true;
|
||||
})
|
||||
).rejects.toBeInstanceOf(TurnDaemonLeaseLostError);
|
||||
expect(committed).toBe(false);
|
||||
expect(lease.isLost()).toBe(true);
|
||||
expect(await lease.renew()).toBe(false);
|
||||
// The monitor does not force-clear ownership or bypass the successor's fencing.
|
||||
expect((await db.turnDaemonLease.findUniqueOrThrow({ where: { profile } })).ownerId).toBe('saved-vm-owner');
|
||||
const next = await createLease(profile, 'fresh-vm-owner');
|
||||
expect(await next.acquire()).toBeNull();
|
||||
await lease.release();
|
||||
expect((await db.turnDaemonLease.findUniqueOrThrow({ where: { profile } })).clockReady).toBe(false);
|
||||
expect((await next.acquire())?.fencingEpoch).toBe(2n);
|
||||
});
|
||||
|
||||
it('persists and evaluates lease wall time as UTC under a non-UTC database session', async () => {
|
||||
const profile = `${profilePrefix}utc-wall`;
|
||||
const zonedUrl = new URL(databaseUrl!);
|
||||
|
||||
@@ -1,6 +1,8 @@
|
||||
import { describe, expect, it, vi } from 'vitest';
|
||||
import { ManualClock } from '@sammo-ts/common';
|
||||
import { TurnDaemonLeaseLostError } from '../src/lifecycle/databaseTurnDaemonLease.js';
|
||||
import { ReservedTurnLeaseBusyError } from '../src/turn/reservedTurnStore.js';
|
||||
import { createReservedTurnLeaseRetry } from '../src/turn/reservedTurnLeaseRetry.js';
|
||||
|
||||
import {
|
||||
InMemoryControlQueue,
|
||||
@@ -15,6 +17,68 @@ import {
|
||||
const addMinutes = (time: Date, minutes: number): Date => new Date(time.getTime() + minutes * 60_000);
|
||||
|
||||
describe('TurnDaemonLifecycle', () => {
|
||||
it('rolls back a busy queue, checks the operator gate, and retries an explicit run exactly once', async () => {
|
||||
const now = new Date('2026-11-12T00:00:00Z');
|
||||
const queue = new InMemoryControlQueue();
|
||||
queue.enqueue({ type: 'run', reason: 'manual', targetTime: now.toISOString() });
|
||||
let resource = 10;
|
||||
let runs = 0;
|
||||
const stateManager = new EngineStateManager();
|
||||
stateManager.register('resource', {
|
||||
capture: () => resource,
|
||||
restore: (value) => {
|
||||
resource = value;
|
||||
},
|
||||
});
|
||||
const onRunError = vi.fn();
|
||||
const flushChanges = vi.fn();
|
||||
const gate = vi.fn(async () => false);
|
||||
const retry = createReservedTurnLeaseRetry({
|
||||
sleep: async () => {
|
||||
expect(resource).toBe(10);
|
||||
},
|
||||
});
|
||||
const lifecycle = new TurnDaemonLifecycle(
|
||||
{
|
||||
clock: new ManualClock(now.getTime()),
|
||||
controlQueue: queue,
|
||||
stateManager,
|
||||
runErrorRetry: retry,
|
||||
getNextTickTime: (value) => addMinutes(value, 5),
|
||||
pauseGate: gate,
|
||||
stateStore: {
|
||||
loadLastTurnTime: async () => now,
|
||||
loadNextGeneralTurnTime: async () => null,
|
||||
saveLastTurnTime: async () => {},
|
||||
loadCheckpoint: async () => undefined,
|
||||
saveCheckpoint: async () => {},
|
||||
},
|
||||
processor: {
|
||||
run: async () => {
|
||||
resource -= 2;
|
||||
if (++runs === 1) throw new ReservedTurnLeaseBusyError('general:7');
|
||||
queue.enqueue({ type: 'shutdown' });
|
||||
return {
|
||||
lastTurnTime: now.toISOString(),
|
||||
processedGenerals: 1,
|
||||
processedTurns: 0,
|
||||
durationMs: 0,
|
||||
partial: false,
|
||||
};
|
||||
},
|
||||
},
|
||||
hooks: { onRunError, flushChanges },
|
||||
},
|
||||
{ profile: 'che', defaultBudget: { budgetMs: 100, maxGenerals: 1, catchUpCap: 1 } }
|
||||
);
|
||||
await lifecycle.start();
|
||||
expect(runs).toBe(2);
|
||||
expect(resource).toBe(8);
|
||||
expect(gate.mock.calls.length).toBeGreaterThanOrEqual(2);
|
||||
expect(flushChanges).toHaveBeenCalledOnce();
|
||||
expect(onRunError).not.toHaveBeenCalled();
|
||||
});
|
||||
|
||||
it.each([false, true])('reports a fatal lease gate and exits even if reporting fails (%s)', async (reportFails) => {
|
||||
const now = new Date('2026-09-09T17:30:00Z');
|
||||
const error = new TurnDaemonLeaseLostError('che:default');
|
||||
|
||||
Reference in new issue
Block a user