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
@@ -8,6 +8,7 @@ import { createTurnDaemonMemoryReporter } from './turnDaemonMemoryReporter.js';
|
||||
import { createGatewayProfileGate } from './gatewayProfileGate.js';
|
||||
import { retryTurnDaemonLeaseStartup } from './leaseStartupRetry.js';
|
||||
import { TurnDaemonLeaseUnavailableError } from '../lifecycle/databaseTurnDaemonLease.js';
|
||||
import { loadHostBootMonitor } from './hostBootMonitor.js';
|
||||
|
||||
export interface TurnDaemonCliOptions {
|
||||
profile?: string;
|
||||
@@ -80,6 +81,7 @@ export const runTurnDaemonCli = async (options: TurnDaemonCliOptions = {}): Prom
|
||||
throw new Error(`GAME_CLOCK_MODE must be realtime or manual: ${rawGameClockMode}`);
|
||||
}
|
||||
const gameClockMode = rawGameClockMode as GameClockMode | undefined;
|
||||
const hostBootMonitor = await loadHostBootMonitor(env);
|
||||
|
||||
const runtime = await retryTurnDaemonLeaseStartup(() =>
|
||||
createTurnDaemonRuntime({
|
||||
@@ -95,6 +97,7 @@ export const runTurnDaemonCli = async (options: TurnDaemonCliOptions = {}): Prom
|
||||
pauseGateIntervalMs,
|
||||
adminActionIntervalMs,
|
||||
gameClockMode,
|
||||
hostBootMonitor,
|
||||
})
|
||||
).catch(async (error: unknown) => {
|
||||
// 중복 starter가 정상 owner를 멈추면 안 된다. 그 밖의 초기화 실패는
|
||||
|
||||
@@ -0,0 +1,142 @@
|
||||
import { createHmac, randomBytes, timingSafeEqual } from 'node:crypto';
|
||||
import { readFile } from 'node:fs/promises';
|
||||
import { performance } from 'node:perf_hooks';
|
||||
import { isRecord } from '@sammo-ts/common';
|
||||
|
||||
export interface HostBootMonitor {
|
||||
hasChanged(): Promise<boolean>;
|
||||
}
|
||||
|
||||
const mac = (key: Buffer, value: string): string => createHmac('sha256', key).update(value).digest('hex');
|
||||
|
||||
/** 부팅 세대는 첫 인증 응답에 고정한다. 통신 실패는 기존 lease 복구를 막지 않는다. */
|
||||
export const createHostBootMonitor = (options: {
|
||||
url: string;
|
||||
key: Buffer;
|
||||
pollMs?: number;
|
||||
timeoutMs?: number;
|
||||
now?: () => number;
|
||||
wallNow?: () => number;
|
||||
onUnavailable?: () => void;
|
||||
onVerified?: () => void;
|
||||
}): HostBootMonitor => {
|
||||
const url = new URL(options.url);
|
||||
if (
|
||||
!['http:', 'https:'].includes(url.protocol) ||
|
||||
url.username ||
|
||||
url.password ||
|
||||
url.search ||
|
||||
url.hash ||
|
||||
url.pathname !== '/v1/boot-id'
|
||||
) {
|
||||
throw new Error('Host boot URL must use HTTP(S) and /v1/boot-id without credentials or query.');
|
||||
}
|
||||
if (options.key.length !== 32) throw new Error('Host boot key must contain exactly 32 bytes.');
|
||||
const now = options.now ?? (() => performance.now());
|
||||
const wallNow = options.wallNow ?? (() => Date.now());
|
||||
let baseline: string | null = null;
|
||||
let changed = false;
|
||||
let lastAttempt = -Infinity;
|
||||
let lastWall = 0;
|
||||
let lastWarning = -Infinity;
|
||||
let inFlight: Promise<boolean> | null = null;
|
||||
const sample = async (): Promise<boolean> => {
|
||||
const nonce = randomBytes(32).toString('hex');
|
||||
const response = await fetch(url, {
|
||||
method: 'GET',
|
||||
redirect: 'error',
|
||||
signal: AbortSignal.timeout(options.timeoutMs ?? 1_000),
|
||||
headers: {
|
||||
'x-sammo-challenge': nonce,
|
||||
'x-sammo-signature': mac(options.key, `sammo-host-boot-request-v1\n${nonce}`),
|
||||
},
|
||||
});
|
||||
if (!response.ok || !response.body) {
|
||||
await response.body?.cancel();
|
||||
throw new Error('Host boot response unavailable.');
|
||||
}
|
||||
const reader = response.body.getReader();
|
||||
const chunks: Uint8Array[] = [];
|
||||
let size = 0;
|
||||
try {
|
||||
for (;;) {
|
||||
const { done, value } = await reader.read();
|
||||
if (done) break;
|
||||
size += value.length;
|
||||
if (size > 2_048) throw new Error('Host boot response exceeds its limit.');
|
||||
chunks.push(value);
|
||||
}
|
||||
} finally {
|
||||
await reader.cancel();
|
||||
}
|
||||
const body: unknown = JSON.parse(Buffer.concat(chunks).toString('utf8'));
|
||||
if (
|
||||
!isRecord(body) ||
|
||||
body.version !== 1 ||
|
||||
body.challenge !== nonce ||
|
||||
typeof body.bootId !== 'string' ||
|
||||
!/^[a-f0-9]{64}$/.test(body.bootId) ||
|
||||
typeof body.signature !== 'string' ||
|
||||
!/^[a-f0-9]{64}$/.test(body.signature)
|
||||
) {
|
||||
throw new Error('Host boot response is invalid.');
|
||||
}
|
||||
const expected = mac(options.key, `sammo-host-boot-response-v1\n${nonce}\n${body.bootId}`);
|
||||
if (!timingSafeEqual(Buffer.from(expected, 'hex'), Buffer.from(body.signature, 'hex'))) {
|
||||
throw new Error('Host boot response signature is invalid.');
|
||||
}
|
||||
if (baseline === null) {
|
||||
baseline = body.bootId;
|
||||
options.onVerified?.();
|
||||
}
|
||||
changed ||= body.bootId !== baseline;
|
||||
return changed;
|
||||
};
|
||||
return {
|
||||
hasChanged(): Promise<boolean> {
|
||||
if (changed) return Promise.resolve(true);
|
||||
if (inFlight) return inFlight;
|
||||
const current = now();
|
||||
const wall = wallNow();
|
||||
const discontinuity = Math.abs(wall - lastWall - (current - lastAttempt)) > 1_000;
|
||||
if (current - lastAttempt < (options.pollMs ?? 2_000) && !discontinuity) return Promise.resolve(false);
|
||||
lastAttempt = current;
|
||||
lastWall = wall;
|
||||
inFlight = sample()
|
||||
.catch(() => {
|
||||
if (current - lastWarning >= 60_000) {
|
||||
options.onUnavailable?.();
|
||||
lastWarning = current;
|
||||
}
|
||||
return false;
|
||||
})
|
||||
.finally(() => {
|
||||
inFlight = null;
|
||||
});
|
||||
return inFlight;
|
||||
},
|
||||
};
|
||||
};
|
||||
|
||||
export const loadHostBootMonitor = async (env: NodeJS.ProcessEnv): Promise<HostBootMonitor | undefined> => {
|
||||
const url = env.TURN_HOST_BOOT_URL?.trim();
|
||||
const keyFile = env.TURN_HOST_BOOT_KEY_FILE?.trim();
|
||||
if (!url && !keyFile) return undefined;
|
||||
try {
|
||||
if (!url || !keyFile) throw new Error('Incomplete optional configuration.');
|
||||
const encoded = (await readFile(keyFile, 'utf8')).trim();
|
||||
if (!/^[a-f0-9]{64}$/.test(encoded)) throw new Error('Invalid optional key file.');
|
||||
return createHostBootMonitor({
|
||||
url,
|
||||
key: Buffer.from(encoded, 'hex'),
|
||||
onVerified: () => console.info('[turn-daemon] host boot identity verified.'),
|
||||
onUnavailable: () => {
|
||||
console.warn('[turn-daemon] host boot identity unavailable; ordinary lease recovery remains active.');
|
||||
},
|
||||
});
|
||||
} catch {
|
||||
// 보조 관측기의 설정/파일 부재도 게임 기동의 필수 의존성이 되지 않는다.
|
||||
console.warn('[turn-daemon] host boot observation disabled; check optional configuration and key file.');
|
||||
return undefined;
|
||||
}
|
||||
};
|
||||
@@ -0,0 +1,38 @@
|
||||
import { performance } from 'node:perf_hooks';
|
||||
import { setTimeout } from 'node:timers/promises';
|
||||
import { ReservedTurnLeaseBusyError } from './reservedTurnStore.js';
|
||||
|
||||
/** rollback 완료 뒤에만 대기한다. 다음 loop에서 pause/lease gate와 최신 큐를 다시 읽는다. */
|
||||
export const createReservedTurnLeaseRetry = (
|
||||
options: {
|
||||
now?: () => number;
|
||||
sleep?: (ms: number) => Promise<void>;
|
||||
maxWaitMs?: number;
|
||||
onWait?: (queueKey: string) => void;
|
||||
} = {}
|
||||
) => {
|
||||
const now = options.now ?? (() => performance.now());
|
||||
const sleep = options.sleep ?? ((ms: number) => setTimeout(ms));
|
||||
const maxWaitMs = options.maxWaitMs ?? 5 * 60_000 + 1_000;
|
||||
let startedAt: number | null = null;
|
||||
let lastReportedAt = -Infinity;
|
||||
return {
|
||||
reset(): void {
|
||||
startedAt = null;
|
||||
lastReportedAt = -Infinity;
|
||||
},
|
||||
async wait(error: unknown): Promise<boolean> {
|
||||
if (!(error instanceof ReservedTurnLeaseBusyError)) return false;
|
||||
const current = now();
|
||||
startedAt ??= current;
|
||||
const remaining = maxWaitMs - (current - startedAt);
|
||||
if (remaining <= 0) return false;
|
||||
if (current - lastReportedAt >= 30_000) {
|
||||
options.onWait?.(error.queueKey);
|
||||
lastReportedAt = current;
|
||||
}
|
||||
await sleep(Math.min(1_000, remaining));
|
||||
return true;
|
||||
},
|
||||
};
|
||||
};
|
||||
@@ -129,6 +129,14 @@ export class ReservedTurnLeaseConflictError extends Error {
|
||||
}
|
||||
}
|
||||
|
||||
/** 실행 전 획득 충돌만 재시도한다. flush/finalize의 소유권 상실과 구분한다. */
|
||||
export class ReservedTurnLeaseBusyError extends ReservedTurnLeaseConflictError {
|
||||
constructor(queueKey: string) {
|
||||
super(queueKey);
|
||||
this.name = 'ReservedTurnLeaseBusyError';
|
||||
}
|
||||
}
|
||||
|
||||
export class InMemoryReservedTurnStore {
|
||||
private readonly generalTurns = new Map<number, ReservedTurnEntry[]>();
|
||||
private readonly nationTurns = new Map<string, ReservedTurnEntry[]>();
|
||||
@@ -249,11 +257,7 @@ export class InMemoryReservedTurnStore {
|
||||
}
|
||||
};
|
||||
pruneNationKeys(Array.from(this.nationTurns.keys()), (key) => this.nationTurns.delete(key));
|
||||
for (const keys of [
|
||||
this.dirtyNationKeys,
|
||||
this.pendingNationInitializationKeys,
|
||||
this.leasedNationKeys,
|
||||
]) {
|
||||
for (const keys of [this.dirtyNationKeys, this.pendingNationInitializationKeys, this.leasedNationKeys]) {
|
||||
for (const key of Array.from(keys)) {
|
||||
const nationId = Number(key.split(':', 1)[0]);
|
||||
if (deletedNations.has(nationId)) {
|
||||
@@ -376,7 +380,7 @@ export class InMemoryReservedTurnStore {
|
||||
}
|
||||
}
|
||||
if (claimed.count !== 1) {
|
||||
throw new ReservedTurnLeaseConflictError(`general:${generalId}`);
|
||||
throw new ReservedTurnLeaseBusyError(`general:${generalId}`);
|
||||
}
|
||||
this.leasedGeneralIds.add(generalId);
|
||||
return !retainedExistingLease;
|
||||
@@ -427,7 +431,7 @@ export class InMemoryReservedTurnStore {
|
||||
}
|
||||
}
|
||||
if (claimed.count !== 1) {
|
||||
throw new ReservedTurnLeaseConflictError(`nation:${nationId}:${officerLevel}`);
|
||||
throw new ReservedTurnLeaseBusyError(`nation:${nationId}:${officerLevel}`);
|
||||
}
|
||||
this.leasedNationKeys.add(buildNationKey(nationId, officerLevel));
|
||||
return !retainedExistingLease;
|
||||
@@ -904,6 +908,24 @@ export class InMemoryReservedTurnStore {
|
||||
await this.persistChanges(this.prisma, changes);
|
||||
this.acknowledgeDirtyState(changes);
|
||||
}
|
||||
|
||||
/** 실행 루프 종료 뒤 호출한다. rollback으로 journal에서 빠진 DB 임대도 해제한다. */
|
||||
async releaseOwnedLeases(): Promise<void> {
|
||||
const results = await Promise.allSettled([
|
||||
this.prisma.generalTurnRevision?.updateMany({
|
||||
where: { leaseOwner: this.leaseOwner },
|
||||
data: { leaseOwner: null, leaseExpiresAt: null },
|
||||
}),
|
||||
this.prisma.nationTurnRevision?.updateMany({
|
||||
where: { leaseOwner: this.leaseOwner },
|
||||
data: { leaseOwner: null, leaseExpiresAt: null },
|
||||
}),
|
||||
]);
|
||||
const errors = results.flatMap((result) => (result.status === 'rejected' ? [result.reason] : []));
|
||||
if (errors.length) throw new AggregateError(errors, 'Reserved turn lease cleanup failed.');
|
||||
this.leasedGeneralIds.clear();
|
||||
this.leasedNationKeys.clear();
|
||||
}
|
||||
}
|
||||
|
||||
export const createReservedTurnStore = async (options: ReservedTurnStoreOptions): Promise<ReservedTurnStoreHandle> => {
|
||||
@@ -918,6 +940,12 @@ export const createReservedTurnStore = async (options: ReservedTurnStoreOptions)
|
||||
await store.loadAll();
|
||||
return {
|
||||
store,
|
||||
close: () => connector.disconnect(),
|
||||
close: async () => {
|
||||
try {
|
||||
await store.releaseOwnedLeases();
|
||||
} finally {
|
||||
await connector.disconnect();
|
||||
}
|
||||
},
|
||||
};
|
||||
};
|
||||
@@ -2,7 +2,7 @@ import type { GameClockPhase } from '@sammo-ts/common';
|
||||
|
||||
/** Gateway의 실행 gate와 durable 시계를 명령 claim 전에 맞춘다. */
|
||||
export const createRuntimePauseGate = (options: {
|
||||
assertLease(): void;
|
||||
assertLease(): void | Promise<void>;
|
||||
shouldPause(): Promise<boolean>;
|
||||
getPhase(): GameClockPhase;
|
||||
isExplicitlyPaused(): boolean;
|
||||
@@ -11,7 +11,7 @@ export const createRuntimePauseGate = (options: {
|
||||
}): (() => Promise<boolean>) => {
|
||||
let lastPaused: boolean | null = null;
|
||||
return async () => {
|
||||
options.assertLease();
|
||||
await options.assertLease();
|
||||
const paused = await options.shouldPause();
|
||||
const phase = options.getPhase();
|
||||
// 오류 정지는 Gateway 상태만 PAUSED로 바꿀 수 있다. 그대로 두면 가입은
|
||||
|
||||
@@ -108,6 +108,8 @@ import { DatabaseTurnDaemonLease, TurnDaemonLeaseUnavailableError } from '../lif
|
||||
import { EngineStateManager } from './engineStateManager.js';
|
||||
import { applyRuntimeClockShift } from './runtimeClockShift.js';
|
||||
import { applyRuntimeGameSettings } from './runtimeGameSettings.js';
|
||||
import { createReservedTurnLeaseRetry } from './reservedTurnLeaseRetry.js';
|
||||
import type { HostBootMonitor } from './hostBootMonitor.js';
|
||||
|
||||
export interface TurnDaemonRuntimeOptions {
|
||||
profile: string;
|
||||
@@ -134,6 +136,7 @@ export interface TurnDaemonRuntimeOptions {
|
||||
leaseDurationMs?: number;
|
||||
leaseOwnerId?: string;
|
||||
enableLeaseHeartbeat?: boolean;
|
||||
hostBootMonitor?: HostBootMonitor;
|
||||
/**
|
||||
* Isolated, single-process fixture acceleration only. Reserved turns are
|
||||
* loaded once and no concurrent API writer may touch this database.
|
||||
@@ -985,6 +988,7 @@ const createTurnDaemonRuntimeWithLease = async (
|
||||
prepareClockRecovery = dbHooks.prepareRealtimeRecovery;
|
||||
prunePreviousAudit = dbHooks.prunePreviousAudit;
|
||||
close = async () => {
|
||||
await adminActionConsumer?.stop();
|
||||
if (auctionBidder) {
|
||||
await auctionBidder.close();
|
||||
}
|
||||
@@ -995,12 +999,15 @@ const createTurnDaemonRuntimeWithLease = async (
|
||||
await tournamentRewardFinalizer.close();
|
||||
}
|
||||
await dbHooks.close();
|
||||
if (reservedTurnStoreHandle) {
|
||||
await reservedTurnStoreHandle.close();
|
||||
try {
|
||||
await reservedTurnStoreHandle?.close();
|
||||
} finally {
|
||||
try {
|
||||
await gatewayGate?.close();
|
||||
} finally {
|
||||
await turnDaemonLease?.close();
|
||||
}
|
||||
}
|
||||
await gatewayGate?.close();
|
||||
await adminActionConsumer?.stop();
|
||||
await turnDaemonLease?.close();
|
||||
};
|
||||
} else if (reservedTurnStoreHandle) {
|
||||
hooks = {
|
||||
@@ -1009,9 +1016,12 @@ const createTurnDaemonRuntimeWithLease = async (
|
||||
},
|
||||
};
|
||||
close = async () => {
|
||||
await reservedTurnStoreHandle.close();
|
||||
await gatewayGate?.close();
|
||||
await adminActionConsumer?.stop();
|
||||
try {
|
||||
await reservedTurnStoreHandle.close();
|
||||
} finally {
|
||||
await gatewayGate?.close();
|
||||
}
|
||||
};
|
||||
} else if (gatewayGate) {
|
||||
hooks = {
|
||||
@@ -1070,12 +1080,17 @@ const createTurnDaemonRuntimeWithLease = async (
|
||||
const baseClose = close;
|
||||
close = async () => {
|
||||
stopClockProjectionWorker();
|
||||
await baseClose();
|
||||
await neutralAuctionRegistrar.close();
|
||||
if (redisConnector) {
|
||||
await redisConnector.disconnect();
|
||||
}
|
||||
await commandConnector?.disconnect();
|
||||
const baseResults = await Promise.allSettled([baseClose()]);
|
||||
// queue 임대 정리 실패도 다른 연결/worker의 종료를 막지 않게 한다.
|
||||
const results = await Promise.allSettled([
|
||||
neutralAuctionRegistrar.close(),
|
||||
redisConnector?.disconnect(),
|
||||
commandConnector?.disconnect(),
|
||||
]);
|
||||
const failures = [...baseResults, ...results].flatMap((result) =>
|
||||
result.status === 'rejected' ? [result.reason] : []
|
||||
);
|
||||
if (failures.length) throw new AggregateError(failures, 'Turn daemon connection cleanup failed.');
|
||||
};
|
||||
|
||||
const resolvedControlQueue = options.controlQueue ?? databaseCommandQueue ?? controlQueue;
|
||||
@@ -1109,7 +1124,8 @@ const createTurnDaemonRuntimeWithLease = async (
|
||||
processor,
|
||||
hooks,
|
||||
pauseGate: createRuntimePauseGate({
|
||||
assertLease: () => {
|
||||
assertLease: async () => {
|
||||
await turnDaemonLease?.checkHostBoot();
|
||||
if (turnDaemonLease?.isLost()) throw turnDaemonLease.getLossError();
|
||||
},
|
||||
shouldPause: async () => (await pauseGate?.()) ?? false,
|
||||
@@ -1127,6 +1143,9 @@ const createTurnDaemonRuntimeWithLease = async (
|
||||
// accumulated world before every due tick in that isolated mode;
|
||||
// production and gateway-managed runtimes keep rollback savepoints.
|
||||
stateManager: options.exclusiveFastForward ? undefined : stateManager,
|
||||
runErrorRetry: createReservedTurnLeaseRetry({
|
||||
onWait: (queueKey) => console.info(`[turn-daemon] waiting for reserved turn lease: ${queueKey}`),
|
||||
}),
|
||||
},
|
||||
{ profile: options.profile, defaultBudget }
|
||||
);
|
||||
@@ -1179,6 +1198,7 @@ export const createTurnDaemonRuntime = async (options: TurnDaemonRuntimeOptions)
|
||||
ownerId: options.leaseOwnerId,
|
||||
leaseDurationMs: options.leaseDurationMs,
|
||||
heartbeat: options.enableLeaseHeartbeat,
|
||||
hostBootMonitor: options.hostBootMonitor,
|
||||
})
|
||||
: null;
|
||||
if (turnDaemonLease && !(await turnDaemonLease.acquire())) {
|
||||
|
||||
Reference in new issue
Block a user