fix lease fencing during silent database partitions
This commit is contained in:
@@ -1,4 +1,5 @@
|
||||
import { randomUUID } from 'node:crypto';
|
||||
import { performance } from 'node:perf_hooks';
|
||||
|
||||
import { createGamePostgresConnector, GamePrisma, type GamePrismaClient } from '@sammo-ts/infra';
|
||||
|
||||
@@ -46,6 +47,8 @@ export class DatabaseTurnDaemonLease {
|
||||
private readonly heartbeatEnabled: boolean;
|
||||
private token: TurnDaemonLeaseToken | null = null;
|
||||
private heartbeatTimer: NodeJS.Timeout | null = null;
|
||||
private expiryTimer: NodeJS.Timeout | null = null;
|
||||
private renewalInFlight = false;
|
||||
private lost = false;
|
||||
|
||||
private constructor(
|
||||
@@ -71,6 +74,7 @@ export class DatabaseTurnDaemonLease {
|
||||
}
|
||||
|
||||
async acquire(): Promise<TurnDaemonLeaseToken | null> {
|
||||
const requestStartedAt = performance.now();
|
||||
const rows = await this.db.$queryRaw<LeaseRow[]>(GamePrisma.sql`
|
||||
INSERT INTO "turn_daemon_lease" (
|
||||
"profile",
|
||||
@@ -111,6 +115,7 @@ export class DatabaseTurnDaemonLease {
|
||||
fencingEpoch: BigInt(row.fencing_epoch),
|
||||
};
|
||||
this.lost = false;
|
||||
this.scheduleExpiryWatchdog(requestStartedAt);
|
||||
if (this.heartbeatEnabled) {
|
||||
this.startHeartbeat();
|
||||
}
|
||||
@@ -127,26 +132,36 @@ export class DatabaseTurnDaemonLease {
|
||||
|
||||
async renew(): Promise<boolean> {
|
||||
const token = this.token;
|
||||
if (!token || this.lost) {
|
||||
if (!token || this.lost || this.renewalInFlight) {
|
||||
return false;
|
||||
}
|
||||
const rows = await this.db.$queryRaw<LeaseRow[]>(GamePrisma.sql`
|
||||
UPDATE "turn_daemon_lease"
|
||||
SET
|
||||
"lease_until" = CURRENT_TIMESTAMP + (${this.leaseDurationMs} * INTERVAL '1 millisecond'),
|
||||
"heartbeat_at" = CURRENT_TIMESTAMP
|
||||
WHERE
|
||||
"profile" = ${token.profile}
|
||||
AND "owner_id" = ${token.ownerId}
|
||||
AND "fencing_epoch" = ${token.fencingEpoch}
|
||||
AND "lease_until" > CURRENT_TIMESTAMP
|
||||
RETURNING "profile", "owner_id", "fencing_epoch"
|
||||
`);
|
||||
if (rows.length === 0) {
|
||||
this.markLost();
|
||||
return false;
|
||||
const requestStartedAt = performance.now();
|
||||
this.renewalInFlight = true;
|
||||
try {
|
||||
const rows = await this.db.$queryRaw<LeaseRow[]>(GamePrisma.sql`
|
||||
UPDATE "turn_daemon_lease"
|
||||
SET
|
||||
"lease_until" = CURRENT_TIMESTAMP + (${this.leaseDurationMs} * INTERVAL '1 millisecond'),
|
||||
"heartbeat_at" = CURRENT_TIMESTAMP
|
||||
WHERE
|
||||
"profile" = ${token.profile}
|
||||
AND "owner_id" = ${token.ownerId}
|
||||
AND "fencing_epoch" = ${token.fencingEpoch}
|
||||
AND "lease_until" > CURRENT_TIMESTAMP
|
||||
RETURNING "profile", "owner_id", "fencing_epoch"
|
||||
`);
|
||||
if (rows.length === 0) {
|
||||
this.markLost();
|
||||
return false;
|
||||
}
|
||||
if (this.lost) {
|
||||
return false;
|
||||
}
|
||||
this.scheduleExpiryWatchdog(requestStartedAt);
|
||||
return true;
|
||||
} finally {
|
||||
this.renewalInFlight = false;
|
||||
}
|
||||
return true;
|
||||
}
|
||||
|
||||
async assertActive(transaction?: GamePrisma.TransactionClient): Promise<void> {
|
||||
@@ -173,6 +188,7 @@ export class DatabaseTurnDaemonLease {
|
||||
|
||||
async release(): Promise<void> {
|
||||
this.stopHeartbeat();
|
||||
this.stopExpiryWatchdog();
|
||||
const token = this.token;
|
||||
this.token = null;
|
||||
if (!token || this.lost) {
|
||||
@@ -216,8 +232,27 @@ export class DatabaseTurnDaemonLease {
|
||||
}
|
||||
}
|
||||
|
||||
private scheduleExpiryWatchdog(requestStartedAt: number): void {
|
||||
// DB가 lease_until을 정하는 시점보다 앞선 요청 시작 시각을 기준으로
|
||||
// 잡아, heartbeat 응답이 멈춰도 DB lease 만료보다 늦게 pause하지 않는다.
|
||||
this.stopExpiryWatchdog();
|
||||
const remainingMs = Math.max(0, this.leaseDurationMs - (performance.now() - requestStartedAt));
|
||||
this.expiryTimer = setTimeout(() => {
|
||||
this.markLost();
|
||||
}, remainingMs);
|
||||
this.expiryTimer.unref();
|
||||
}
|
||||
|
||||
private stopExpiryWatchdog(): void {
|
||||
if (this.expiryTimer) {
|
||||
clearTimeout(this.expiryTimer);
|
||||
this.expiryTimer = null;
|
||||
}
|
||||
}
|
||||
|
||||
private markLost(): void {
|
||||
this.lost = true;
|
||||
this.stopHeartbeat();
|
||||
this.stopExpiryWatchdog();
|
||||
}
|
||||
}
|
||||
|
||||
@@ -57,31 +57,43 @@ const createPartitionableTcpProxy = async (
|
||||
restore: () => Promise<void>;
|
||||
close: () => Promise<void>;
|
||||
}> => {
|
||||
const sockets = new Set<Socket>();
|
||||
type ProxyPair = {
|
||||
client: Socket;
|
||||
upstream: Socket;
|
||||
};
|
||||
const pairs = new Set<ProxyPair>();
|
||||
let server: Server | null = null;
|
||||
let port = 0;
|
||||
let partitioned = false;
|
||||
|
||||
const resumePair = (pair: ProxyPair): void => {
|
||||
if (partitioned || pair.client.destroyed || pair.upstream.destroyed || pair.upstream.connecting) {
|
||||
return;
|
||||
}
|
||||
pair.client.pipe(pair.upstream);
|
||||
pair.upstream.pipe(pair.client);
|
||||
};
|
||||
|
||||
const listen = async (requestedPort: number): Promise<void> => {
|
||||
const nextServer = createServer((client) => {
|
||||
const upstream = createConnection({ host: upstreamHost, port: upstreamPort });
|
||||
sockets.add(client);
|
||||
sockets.add(upstream);
|
||||
const pair = { client, upstream };
|
||||
pairs.add(pair);
|
||||
client.on('error', () => undefined);
|
||||
upstream.on('error', () => undefined);
|
||||
client.once('close', () => {
|
||||
sockets.delete(client);
|
||||
pairs.delete(pair);
|
||||
if (!upstream.destroyed) {
|
||||
upstream.destroy();
|
||||
}
|
||||
});
|
||||
upstream.once('close', () => {
|
||||
sockets.delete(upstream);
|
||||
pairs.delete(pair);
|
||||
if (!client.destroyed) {
|
||||
client.destroy();
|
||||
}
|
||||
});
|
||||
client.pipe(upstream);
|
||||
upstream.pipe(client);
|
||||
upstream.once('connect', () => resumePair(pair));
|
||||
});
|
||||
await new Promise<void>((resolve, reject) => {
|
||||
const onError = (error: Error): void => reject(error);
|
||||
@@ -99,7 +111,16 @@ const createPartitionableTcpProxy = async (
|
||||
port = address.port;
|
||||
};
|
||||
|
||||
const partition = async (): Promise<void> => {
|
||||
const partition = (): Promise<void> => {
|
||||
partitioned = true;
|
||||
for (const pair of pairs) {
|
||||
pair.client.unpipe(pair.upstream);
|
||||
pair.upstream.unpipe(pair.client);
|
||||
}
|
||||
return Promise.resolve();
|
||||
};
|
||||
|
||||
const close = async (): Promise<void> => {
|
||||
const activeServer = server;
|
||||
server = null;
|
||||
if (!activeServer) {
|
||||
@@ -114,10 +135,11 @@ const createPartitionableTcpProxy = async (
|
||||
}
|
||||
});
|
||||
});
|
||||
for (const socket of sockets) {
|
||||
socket.destroy();
|
||||
for (const pair of pairs) {
|
||||
pair.client.destroy();
|
||||
pair.upstream.destroy();
|
||||
}
|
||||
sockets.clear();
|
||||
pairs.clear();
|
||||
await closed;
|
||||
};
|
||||
|
||||
@@ -127,12 +149,14 @@ const createPartitionableTcpProxy = async (
|
||||
return port;
|
||||
},
|
||||
partition,
|
||||
restore: async () => {
|
||||
if (!server) {
|
||||
await listen(port);
|
||||
restore: () => {
|
||||
partitioned = false;
|
||||
for (const pair of pairs) {
|
||||
resumePair(pair);
|
||||
}
|
||||
return Promise.resolve();
|
||||
},
|
||||
close: partition,
|
||||
close,
|
||||
};
|
||||
};
|
||||
|
||||
@@ -321,15 +345,15 @@ integration('database turn daemon lease and fencing', () => {
|
||||
expect((await partitionedOwner.acquire())?.fencingEpoch).toBe(1n);
|
||||
|
||||
await proxy.partition();
|
||||
successor = await createLease(profile, 'partition-successor');
|
||||
expect(await successor.acquire()).toBeNull();
|
||||
|
||||
const lostDeadline = Date.now() + leaseDurationMs * 2;
|
||||
while (!partitionedOwner.isLost() && Date.now() < lostDeadline) {
|
||||
await delay(50);
|
||||
}
|
||||
expect(partitionedOwner.isLost()).toBe(true);
|
||||
|
||||
successor = await createLease(profile, 'partition-successor');
|
||||
expect(await successor.acquire()).toBeNull();
|
||||
|
||||
const takeoverDeadline = Date.now() + leaseDurationMs * 4;
|
||||
let successorToken = null;
|
||||
while (Date.now() < takeoverDeadline) {
|
||||
|
||||
Reference in New Issue
Block a user