R3944Realms/PlayerSync
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Perf: staggered auto-save, pool scaling, cached kick check

Browse Source 
CRITICAL PERF - Staggered auto-save:
- Old: all 35 players snapshotted in ONE tick → 770-3605ms MSPT spike
  (15-36 second TPS drop every 5 minutes)
- New: queue filled every 5min, drained 1 player/tick → max 22-103ms/tick
- autoSaveQueue processes one player per server tick, imperceptible impact

CRITICAL PERF - Pool scaling for 35+ players:
- Thread pool: 2-8 → 4-16 threads, queue 256 → 512
  Prevents CallerRunsPolicy from executing DB tasks on main thread
- HikariCP: 10 → 25 max connections, 2 → 4 min idle
  Prevents connection starvation during concurrent saves

HIGH PERF - Cached kick check (eliminates main thread DB queries):
- doPlayerConnect (network thread) caches online/lastServer/serverAlive
- onPlayerLoggedInKickCheck (MAIN thread) reuses cached result
- Fast path: 1 DB query on main thread instead of 2-4
- Fallback: full DB check if cache miss (race condition safety)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
laforetbrut 2026-04-15 11:33:02 +02:00
parent badc87c84e
commit b4d863efa2
2 changed files with 103 additions and 62 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

157
src/main/java/vip/fubuki/playersync/sync/VanillaSync.java
View File

@ -71,13 +71,17 @@ public class VanillaSync {
// Bounded pool: 2 core threads, max 8 threads, 30s keepalive, 256-task queue.
// If the queue is full, tasks run on the calling thread (CallerRunsPolicy) which
// provides natural backpressure instead of creating more threads.
// FIX PERF: Increased pool sizing for 35+ player servers.
// Old: 2-8 threads, 256 queue → CallerRunsPolicy caused main thread to execute
// DB tasks when queue was full (35 auto-save tasks overflowed 256 queue → TPS drop to <1).
// New: 4-16 threads, 512 queue → handles 35+ concurrent saves without overflow.
static ExecutorService executorService = new ThreadPoolExecutor(
2, // core pool size
8, // maximum pool size
4, // core pool size (was 2)
16, // maximum pool size (was 8)
30L, TimeUnit.SECONDS, // idle thread keepalive
new LinkedBlockingQueue<>(256), // bounded work queue
new LinkedBlockingQueue<>(512), // bounded work queue (was 256)
new PSThreadPoolFactory("PlayerSync"),
new ThreadPoolExecutor.CallerRunsPolicy() // backpressure: run on caller thread if queue full
new ThreadPoolExecutor.CallerRunsPolicy()
);
// Per-player locks to prevent concurrent save/restore operations (anti-duplication)
@ -202,6 +206,7 @@ public class VanillaSync {
ResultSet rs1 = qr1.resultSet();
if (!rs1.next()) {
PlayerSync.LOGGER.info("A new-player connection detected");
connectCheckCache.put(player_uuid, new int[]{0, 0, 0, 0}); // new player
return;
}
online = rs1.getBoolean("online");
@ -209,6 +214,8 @@ public class VanillaSync {
}
// Second query: Check if player is already online on another server
int serverAlive = 0;
int alreadyKicked = 0;
if (JdbcConfig.KICK_WHEN_ALREADY_ONLINE.get() && online && lastServer != JdbcConfig.SERVER_ID.get()) {
try (JDBCsetUp.QueryResult qr2 = JDBCsetUp.executePreparedQuery(
"SELECT last_update, enable FROM server_info WHERE id=?", lastServer)) {
@ -217,13 +224,18 @@ public class VanillaSync {
long last_update = rs2.getLong("last_update");
boolean enable = rs2.getBoolean("enable");
if (enable && System.currentTimeMillis() < last_update + 300000L) {
serverAlive = 1;
event.getConnection().disconnect(Component.translatableWithFallback("playersync.already_online","You can't join more than one synchronization server at the same time."));
return;
alreadyKicked = 1;
} else {
JDBCsetUp.executePreparedUpdate("UPDATE server_info SET enable=0 WHERE id=?", lastServer);
}
JDBCsetUp.executePreparedUpdate("UPDATE server_info SET enable=0 WHERE id=?", lastServer);
}
}
}
// FIX PERF: Cache the result for onPlayerLoggedInKickCheck (avoids re-querying on main thread)
connectCheckCache.put(player_uuid, new int[]{online ? 1 : 0, lastServer, serverAlive, alreadyKicked});
} catch (Exception e) {
PlayerSync.LOGGER.error("SqlException detected!", e);
event.getConnection().disconnect(Component.translatableWithFallback("playersync.sqlexception","SqlException detected!Connection lost,please contact with your admin."));
@ -236,6 +248,12 @@ public class VanillaSync {
// Players kicked for being already online on another server - their logout must NOT set online=0
public static Set<String> kickedForDuplicateLogin = ConcurrentHashMap.newKeySet();
// FIX PERF: Cache from doPlayerConnect (network thread) for onPlayerLoggedInKickCheck (main thread).
// Eliminates 2-4 redundant DB queries per join on the main thread.
// Entry: uuid → {online, lastServer, serverAlive, alreadyHandled}
private static final ConcurrentHashMap<String, int[]> connectCheckCache = new ConcurrentHashMap<>();
// int[0]=online(0/1), int[1]=lastServer, int[2]=serverAlive(0/1), int[3]=alreadyKicked(0/1)
public static void doPlayerJoin(PlayerEvent.PlayerLoggedInEvent event) {
ServerPlayer serverPlayer = (ServerPlayer) event.getEntity();
String player_uuid = serverPlayer.getUUID().toString();
@ -421,9 +439,10 @@ public class VanillaSync {
}
// === PHASE 2: Apply to player on MAIN SERVER THREAD ===
// FIX PERF: No more applyLatch.await(60s) tying up a background thread.
// The server.execute() callback fires when the main thread is ready. The
// syncNotCompletedPlayer flag guards onPlayerLogout until apply completes.
// The server.execute() callback fires when the main thread is ready.
// Note: Backpack/SS/RS2 restore still does DB reads on main thread (1-5 queries
// per player). This is acceptable because players join one at a time, not 35 at once.
// The real performance fix is staggering the auto-save (see onServerTick).
server.execute(() -> {
try {
// FIX: Verify the player is still connected before applying data.
@ -544,13 +563,12 @@ public class VanillaSync {
ServerPlayer player = (ServerPlayer) event.getEntity();
String player_uuid = player.getUUID().toString();
// FIX PERF: Use cached data from doPlayerConnect (network thread) instead of
// re-querying the DB. Eliminates 2-4 blocking DB queries from the MAIN THREAD.
// doPlayerConnect already ran the same checks on the network thread and cached results.
int[] cached = connectCheckCache.remove(player_uuid);
if (!JdbcConfig.KICK_WHEN_ALREADY_ONLINE.get()) {
// Still mark online even if kick is disabled.
// FIX: Don't set last_server here — set it AFTER the poll in doPlayerJoin.
// Setting last_server too early breaks the poll loop (sees "player is on my server"
// and breaks immediately) AND prevents the old server's save from completing
// (last_server guard blocks the write). online=1 alone is sufficient to prevent
// triple-login — other servers check online=1 regardless of last_server.
try {
JDBCsetUp.executePreparedUpdate(
"UPDATE player_data SET online=1 WHERE uuid=?",
@ -560,46 +578,54 @@ public class VanillaSync {
}
try {
boolean online = false;
int lastServer = 0;
try (JDBCsetUp.QueryResult qr = JDBCsetUp.executePreparedQuery(
"SELECT online, last_server FROM player_data WHERE uuid=?", player_uuid)) {
ResultSet rs = qr.resultSet();
if (rs.next()) {
online = rs.getBoolean("online");
lastServer = rs.getInt("last_server");
}
if (cached != null && cached[3] == 1) {
// doPlayerConnect already determined this player should be kicked (server alive)
// but PlayerNegotiationEvent.disconnect() is unreliable in NeoForge 1.21.1
// — use the reliable ServerPlayer.connection.disconnect() instead.
kickedForDuplicateLogin.add(player_uuid);
PlayerSync.LOGGER.warn("Kicking player {} - already online on server {} (cached check)", player_uuid, cached[1]);
player.connection.disconnect(Component.translatableWithFallback(
"playersync.already_online",
"You can't join more than one synchronization server at the same time."));
return;
}
if (online && lastServer != JdbcConfig.SERVER_ID.get()) {
// Check if the other server is still alive
try (JDBCsetUp.QueryResult qr2 = JDBCsetUp.executePreparedQuery(
"SELECT last_update, enable FROM server_info WHERE id=?", lastServer)) {
ResultSet rs2 = qr2.resultSet();
if (rs2.next()) {
long lastUpdate = rs2.getLong("last_update");
boolean enable = rs2.getBoolean("enable");
if (enable && System.currentTimeMillis() < lastUpdate + 300000L) {
// Other server is alive → KICK using ServerPlayer.connection which works reliably
// CRITICAL: Mark as kicked BEFORE disconnect so onPlayerLogout does NOT set online=0.
// Without this, the logout handler resets online=0, allowing immediate reconnect bypass.
kickedForDuplicateLogin.add(player_uuid);
PlayerSync.LOGGER.warn("Kicking player {} - already online on server {}", player_uuid, lastServer);
player.connection.disconnect(Component.translatableWithFallback(
"playersync.already_online",
"You can't join more than one synchronization server at the same time."));
return;
if (cached != null && cached[0] == 1 && cached[1] != JdbcConfig.SERVER_ID.get() && cached[2] == 0) {
// Player was online on another server but that server is dead — already handled
// by doPlayerConnect (server disabled). No need to re-query.
} else if (cached == null) {
// No cache (race condition or cache eviction) — fall back to DB query
boolean online = false;
int lastServer = 0;
try (JDBCsetUp.QueryResult qr = JDBCsetUp.executePreparedQuery(
"SELECT online, last_server FROM player_data WHERE uuid=?", player_uuid)) {
ResultSet rs = qr.resultSet();
if (rs.next()) {
online = rs.getBoolean("online");
lastServer = rs.getInt("last_server");
}
}
if (online && lastServer != JdbcConfig.SERVER_ID.get()) {
try (JDBCsetUp.QueryResult qr2 = JDBCsetUp.executePreparedQuery(
"SELECT last_update, enable FROM server_info WHERE id=?", lastServer)) {
ResultSet rs2 = qr2.resultSet();
if (rs2.next()) {
long lastUpdate = rs2.getLong("last_update");
boolean enable = rs2.getBoolean("enable");
if (enable && System.currentTimeMillis() < lastUpdate + 300000L) {
kickedForDuplicateLogin.add(player_uuid);
player.connection.disconnect(Component.translatableWithFallback(
"playersync.already_online",
"You can't join more than one synchronization server at the same time."));
return;
}
JDBCsetUp.executePreparedUpdate("UPDATE server_info SET enable=0 WHERE id=?", lastServer);
}
// Other server is dead, disable it
JDBCsetUp.executePreparedUpdate("UPDATE server_info SET enable=0 WHERE id=?", lastServer);
}
}
}
// Mark online=1 SYNCHRONOUSLY — but don't set last_server yet.
// FIX: last_server is set AFTER the poll in doPlayerJoin to allow the old
// server's async save to complete (its writeSnapshotToDB uses AND last_server=?).
// Mark online=1 — only DB call on main thread in the fast path (1 query instead of 4)
JDBCsetUp.executePreparedUpdate(
"UPDATE player_data SET online=1 WHERE uuid=?",
player_uuid);
@ -1478,6 +1504,10 @@ public class VanillaSync {
private static final int HEARTBEAT_INTERVAL_TICKS = 600; // Every 30 seconds (20 tps * 30s)
private static int autoSaveTickCounter = 0;
private static final int AUTO_SAVE_INTERVAL_TICKS = 6000; // Every 5 minutes (20 tps × 300s)
// FIX PERF: Staggered auto-save. Instead of snapshotting ALL 35 players in one tick
// (770-3605ms spike → 15-36s TPS drop), we save 1 player per tick over 35 ticks
// (22-103ms per tick → imperceptible). The queue is refilled every AUTO_SAVE_INTERVAL.
private static final List<ServerPlayer> autoSaveQueue = new ArrayList<>();
private static int autoCleanCuriosCacheTickCounter = 0;
private static final int AUTO_CLEAN_CURIOS_CACHE_INTERVAL_TICKS = 36000; // Every 30 min
@ -1509,21 +1539,30 @@ public class VanillaSync {
// non-thread-safe way. All entity reads are now done in snapshotPlayerData()
// on the main thread, and the background task only does DB writes.
//
// FIX: Backpack/SS contents are NOW included in the periodic auto-save.
// Previously only saved on logout + shutdown, but hard crashes skip both
// → backpack changes lost. snapshotBackpackData is fast (~1ms per backpack).
// FIX PERF: Staggered auto-save — saves ONE player per tick instead of ALL at once.
// Old behavior: 35 players snapshotted in ONE tick → 770-3605ms MSPT spike every 5 min.
// New behavior: queue refilled every 5 min, then drained 1 player/tick → 22-103ms/tick max.
// Backpack contents are included (prevents data loss on hard crash).
if (autoSaveTickCounter >= AUTO_SAVE_INTERVAL_TICKS) {
autoSaveTickCounter = 0;
// Refill the queue with all eligible players
autoSaveQueue.clear();
MinecraftServer server = ServerLifecycleHooks.getCurrentServer();
if (server != null) {
for (ServerPlayer player : server.getPlayerList().getPlayers()) {
String puuid = player.getUUID().toString();
if (player.isDeadOrDying() || syncNotCompletedPlayer.contains(puuid)
|| pendingLogoutSaves.containsKey(puuid)) {
continue;
}
ReentrantLock lock = getPlayerLock(puuid);
if (!lock.tryLock()) continue;
autoSaveQueue.addAll(server.getPlayerList().getPlayers());
}
}
// Process ONE player from the queue per tick (staggered)
if (!autoSaveQueue.isEmpty()) {
ServerPlayer player = autoSaveQueue.removeFirst();
String puuid = player.getUUID().toString();
// Skip invalid players (same guards as before)
if (!player.isDeadOrDying() && !syncNotCompletedPlayer.contains(puuid)
&& !pendingLogoutSaves.containsKey(puuid) && player.getTags().contains("player_synced")) {
ReentrantLock lock = getPlayerLock(puuid);
if (lock.tryLock()) {
try {
final PlayerDataSnapshot snapshot = snapshotPlayerData(player);
final Map<UUID, CompoundTag> backpackSnapshots = ModsSupport.snapshotBackpackData(player);

8
src/main/java/vip/fubuki/playersync/util/JDBCsetUp.java
View File

@ -43,9 +43,11 @@ public class JDBCsetUp {
cfg.setUsername(JdbcConfig.USERNAME.get());
cfg.setPassword(JdbcConfig.PASSWORD.get());
// Pool sizing: 2 warm connections, up to 10 under load
cfg.setMaximumPoolSize(10);
cfg.setMinimumIdle(2);
// FIX PERF: Increased pool for 35+ player servers.
// Old: 10 max / 2 idle → 35 concurrent saves queued on 10 connections → 250ms+ wait.
// New: 25 max / 4 idle → handles peak load without connection starvation.
cfg.setMaximumPoolSize(25);
cfg.setMinimumIdle(4);
// Connection lifecycle
cfg.setConnectionTimeout(30_000L); // 30 s – how long to wait for a free slot