[refactor/#582] FCM과 채팅 스레드풀 분리

result-bulkhead-before.json

@@ -0,0 +1,197 @@
+{
+    "root_group": {
+        "name": "",
+        "path": "",
+        "id": "d41d8cd98f00b204e9800998ecf8427e",
+        "groups": {},
+        "checks": {
+                "[notif] status 200": {
+                    "passes": 1567,
+                    "fails": 0,
+                    "name": "[notif] status 200",
+                    "path": "::[notif] status 200",
+                    "id": "9fc740afe9cf21a0327678441a348978"
+                },
+                "[ws] status 101": {
+                    "name": "[ws] status 101",
+                    "path": "::[ws] status 101",
+                    "id": "3cc6802624f263d3928ed6de98f3c352",
+                    "passes": 2,
+                    "fails": 0
+                }
+            }
+    },
+    "metrics": {
+        "ws_connecting": {
+            "max": 1155.4575,
+            "p(90)": 1155.34769,
+            "p(95)": 1155.402595,
+            "avg": 1154.90845,
+            "min": 1154.3594,
+            "med": 1154.90845
+        },
+        "ws_msgs_sent": {
+            "count": 1651,
+            "rate": 4.985553421339417
+        },
+        "ws_msgs_received": {
+            "count": 3304,
+            "rate": 9.977146277471492
+        },
+        "http_reqs": {
+            "count": 1567,
+            "rate": 4.731897160047769
+        },
+        "http_req_receiving": {
+            "p(90)": 0.7683200000000002,
+            "p(95)": 0.9485099999999989,
+            "avg": 0.21188544990427544,
+            "min": 0,
+            "med": 0,
+            "max": 6.9107
+        },
+        "checks": {
+            "passes": 1569,
+            "fails": 0,
+            "value": 1
+        },
+        "vus_max": {
+            "value": 12,
+            "min": 12,
+            "max": 12
+        },
+        "chat_messages_received": {
+            "count": 1648,
+            "rate": 4.97649426915043
+        },
+        "iterations": {
+            "count": 1569,
+            "rate": 4.737936594840427
+        },
+        "ws_sessions": {
+            "count": 2,
+            "rate": 0.006039434792658289
+        },
+        "http_req_duration{expected_response:true}": {
+            "p(90)": 2076.8168400000004,
+            "p(95)": 2721.69289,
+            "avg": 1151.817254243779,
+            "min": 431.0899,
+            "med": 977.7864,
+            "max": 4974.7035
+        },
+        "http_req_blocked": {
+            "p(90)": 0,
+            "p(95)": 0,
+            "avg": 1.902208615188258,
+            "min": 0,
+            "med": 0,
+            "max": 315.6305
+        },
+        "http_req_connecting": {
+            "max": 148.6131,
+            "p(90)": 0,
+            "p(95)": 0,
+            "avg": 0.895541671984684,
+            "min": 0,
+            "med": 0
+        },
+        "chat_messages_sent": {
+            "count": 1649,
+            "rate": 4.979513986546759
+        },
+        "http_req_sending": {
+            "med": 0,
+            "max": 3.377,
+            "p(90)": 0.6262200000000001,
+            "p(95)": 0.7184799999999996,
+            "avg": 0.17865022335673258,
+            "min": 0
+        },
+        "vus": {
+            "value": 2,
+            "min": 2,
+            "max": 12
+        },
+        "data_sent": {
+            "count": 251316,
+            "rate": 758.9032971758552
+        },
+        "http_req_waiting": {
+            "p(95)": 2721.5110299999997,
+            "avg": 1151.4267185705173,
+            "min": 430.2286,
+            "med": 977.2512,
+            "max": 4974.1908,
+            "p(90)": 2076.08822
+        },
+        "chat_e2e_latency": {
+            "min": 350,
+            "med": 404,
+            "max": 5422,
+            "p(90)": 3371.499999999998,
+            "p(95)": 4390.65,
+            "avg": 937.6231796116505,
+            "thresholds": {
+                "p(95)<3000": true
+            }
+        },
+        "notif_req_duration": {
+            "med": 977.7864,
+            "max": 4974.7035,
+            "p(90)": 2076.8168400000004,
+            "p(95)": 2721.69289,
+            "avg": 1151.817254243779,
+            "min": 431.0899
+        },
+        "ws_session_duration": {
+            "min": 331155.5802,
+            "med": 331155.89955,
+            "max": 331156.2189,
+            "p(90)": 331156.15502999997,
+            "p(95)": 331156.186965,
+            "avg": 331155.89955
+        },
+        "notif_errors": {
+            "passes": 0,
+            "fails": 1567,
+            "thresholds": {
+                "rate<0.10": false
+            },
+            "value": 0
+        },
+        "http_req_failed": {
+            "fails": 1567,
+            "passes": 0,
+            "value": 0
+        },
+        "iteration_duration": {
+            "avg": 1574.7665509878918,
+            "min": 431.6325,
+            "med": 982.5982,
+            "max": 331156.2189,
+            "p(90)": 2088.631680000001,
+            "p(95)": 2724.6312
+        },
+        "http_req_duration": {
+            "p(90)": 2076.8168400000004,
+            "p(95)": 2721.69289,
+            "avg": 1151.817254243779,
+            "min": 431.0899,
+            "med": 977.7864,
+            "max": 4974.7035
+        },
+        "data_received": {
+            "count": 1517803,
+            "rate": 4583.336123300564
+        },
+        "http_req_tls_handshaking": {
+            "max": 181.0315,
+            "p(90)": 0,
+            "p(95)": 0,
+            "avg": 1.003557562220804,
+            "min": 0,
+            "med": 0
+        }
+    }
+}

src/main/java/umc/cockple/demo/domain/chat/service/listener/ChatMessageSendEventListener.java

@@ -16,7 +16,7 @@ public class ChatMessageSendEventListener {
     private final ChatSendService chatSendService;
 
     @EventListener
-    @Async
+    @Async("chatExecutor")
     public void handleChatMessageSend(ChatMessageSendEvent event) {
         log.info("메시지 전송 이벤트 처리 - 채팅방: {}, 발신자: {}",
                 event.chatRoomId(), event.senderId());

src/main/java/umc/cockple/demo/domain/chat/service/listener/ChatNotificationEventListener.java

@@ -17,7 +17,7 @@ public class ChatNotificationEventListener {
     private final ChatPushNotificationService chatPushNotificationService;
 
     @TransactionalEventListener(phase = TransactionPhase.AFTER_COMMIT)
-    @Async
+    @Async("notificationExecutor")
     public void handleChatNotification(ChatNotificationEvent event) {
         log.info("채팅 알림 이벤트 처리 - 채팅방: {}, 발신자: {}", event.chatRoomId(), event.senderId());
         try {

src/main/java/umc/cockple/demo/domain/chat/service/listener/ChatRoomListUpdateEventListener.java

@@ -24,7 +24,7 @@ public class ChatRoomListUpdateEventListener {
     private final ChatRoomListUpdateBroadcaster chatRoomListUpdateBroadcaster;
 
     @TransactionalEventListener(phase = TransactionPhase.AFTER_COMMIT)
-    @Async
+    @Async("chatExecutor")
     public void handleChatRoomListUpdate(ChatRoomListUpdateEvent event) {
         log.info("채팅방 목록 업데이트 이벤트 처리 시작 - 채팅방: {}", event.chatRoomId());
 

src/main/java/umc/cockple/demo/domain/chat/service/listener/ChatRoomRedisCleanupListener.java

@@ -21,7 +21,7 @@ public class ChatRoomRedisCleanupListener {
     private final ChatListSubscriptionStore chatListSubscriptionStore;
 
     @TransactionalEventListener(phase = TransactionPhase.AFTER_COMMIT)
-    @Async
+    @Async("chatExecutor")
     public void handleChatRoomRedisCleanup(ChatRoomRedisCleanupEvent event) {
         Long chatRoomId = event.chatRoomId();
         log.info("[채팅방 Redis 정리 시작] - chatRoomId: {}", chatRoomId);

src/main/java/umc/cockple/demo/domain/chat/service/listener/ChatSubscriptionEventListener.java

@@ -49,7 +49,7 @@ public void handleChatRoomSubscription(ChatRoomSubscriptionEvent event) {
     }
 
     @EventListener
-    @Async
+    @Async("chatExecutor")
     public void handleChatListSubscription(ChatListSubscriptionEvent event) {
         log.info("채팅방 목록 구독 이벤트 처리 시작 - 멤버: {}, 액션: {}, 채팅방 수: {}",
                 event.memberId(), event.action(), event.chatRoomIds().size());

src/main/java/umc/cockple/demo/domain/chat/service/listener/ChatUnreadStatusUpdateEventListener.java

@@ -30,7 +30,7 @@ public class ChatUnreadStatusUpdateEventListener {
     private final ChatSessionRegistry chatSessionRegistry;
 
     @TransactionalEventListener(phase = TransactionPhase.AFTER_COMMIT)
-    @Async
+    @Async("chatExecutor")
     public void handleChatUnreadStatusUpdate(ChatUnreadStatusUpdateEvent event) {
         log.info("채팅 안읽음 상태 업데이트 이벤트 처리 시작 - 대상자: {}명", event.targetMemberIds().size());
 

src/main/java/umc/cockple/demo/domain/chat/service/listener/PartyChatSystemMessageListener.java

@@ -15,7 +15,7 @@ public class PartyChatSystemMessageListener {
     private final ChatSendService chatSendService;
 
     @TransactionalEventListener(phase = TransactionPhase.AFTER_COMMIT)
-    @Async
+    @Async("chatExecutor")
     public void handlePartyMemberChanged(PartyMemberJoinedEvent event) {
         switch (event.action()) {
             case JOINED -> chatSendService.sendSystemMessage(event.partyId(),

src/main/java/umc/cockple/demo/domain/notification/events/NotificationEventListener.java

@@ -21,7 +21,7 @@ public class NotificationEventListener {
     private final MemberRepository memberRepository;
 
     @TransactionalEventListener(phase = TransactionPhase.AFTER_COMMIT)
-    @Async
+    @Async("notificationExecutor")
     @Transactional(readOnly = true, propagation = Propagation.REQUIRES_NEW)
     public void handleNotification(NotificationEvent event) {
         log.info("[NOTIFICATION] FCM 전송 이벤트 처리 - memberId: {}", event.memberId());

src/main/java/umc/cockple/demo/global/config/AsyncConfig.java

@@ -1,5 +1,7 @@
 package umc.cockple.demo.global.config;
 
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.beans.factory.annotation.Value;
 import org.springframework.boot.autoconfigure.task.TaskExecutionAutoConfiguration;
 import org.springframework.boot.task.ThreadPoolTaskExecutorBuilder;
 import org.springframework.context.annotation.Bean;
@@ -9,16 +11,32 @@
 import org.springframework.scheduling.concurrent.ThreadPoolTaskExecutor;
 import umc.cockple.demo.global.logging.MdcTaskDecorator;
 
+import java.util.concurrent.RejectedExecutionHandler;
+import java.util.concurrent.ThreadPoolExecutor;
+
+/**
+ * 지연 특성별로 벌크헤드(bulkhead) 격리
+ * {@code chatExecutor} — 실시간/저지연
+ * {@code notificationExecutor} — 외부 FCM HTTP 호출
+ * {@code applicationTaskExecutor} — qualifier 없는 @Async의 안전한 기본 착지점
+ */
 @Configuration
 @EnableAsync
+@Slf4j
 public class AsyncConfig {
 
+    /** Spring이 qualifier 없는 @Async의 기본 executor를 찾는 이름. */
+    private static final String DEFAULT_TASK_EXECUTOR_BEAN_NAME = "taskExecutor";
+
     @Bean
     public TaskDecorator mdcTaskDecorator() {
         return new MdcTaskDecorator();
     }
 
-    @Bean(name = TaskExecutionAutoConfiguration.APPLICATION_TASK_EXECUTOR_BEAN_NAME)
+    @Bean(name = {
+            TaskExecutionAutoConfiguration.APPLICATION_TASK_EXECUTOR_BEAN_NAME,
+            DEFAULT_TASK_EXECUTOR_BEAN_NAME
+    })
     public ThreadPoolTaskExecutor applicationTaskExecutor(
             ThreadPoolTaskExecutorBuilder builder,
             TaskDecorator mdcTaskDecorator
@@ -28,4 +46,65 @@ public ThreadPoolTaskExecutor applicationTaskExecutor(
                 .taskDecorator(mdcTaskDecorator)
                 .build();
     }
+
+    /**
+     * 실시간 채팅용 풀-> 짧은 내부 I/O 작업을 저지연으로 처리
+     */
+    @Bean("chatExecutor")
+    public ThreadPoolTaskExecutor chatExecutor(
+            @Value("${async.chat.core-size:8}") int coreSize,
+            @Value("${async.chat.max-size:16}") int maxSize,
+            @Value("${async.chat.queue-capacity:500}") int queueCapacity,
+            @Value("${async.chat.await-termination-seconds:30}") int awaitTerminationSeconds,
+            TaskDecorator mdcTaskDecorator
+    ) {
+        // 채팅은 유실 방지가 우선 -> 포화 시 호출 스레드에서 실행해 백프레셔를 건다.
+        return buildExecutor("cockple-chat-", coreSize, maxSize, queueCapacity,
+                awaitTerminationSeconds, mdcTaskDecorator, new ThreadPoolExecutor.CallerRunsPolicy());
+    }
+
+    /**
+     * FCM 알림용 풀. 외부 Firebase HTTP 호출을 채팅과 격리
+     * 큐를 크게 잡되 유한하게 두어 Firebase 장애 시 무한 백로그(OOM)를 방지
+     */
+    @Bean("notificationExecutor")
+    public ThreadPoolTaskExecutor notificationExecutor(
+            @Value("${async.notification.core-size:4}") int coreSize,
+            @Value("${async.notification.max-size:12}") int maxSize,
+            @Value("${async.notification.queue-capacity:1000}") int queueCapacity,
+            @Value("${async.notification.await-termination-seconds:30}") int awaitTerminationSeconds,
+            TaskDecorator mdcTaskDecorator
+    ) {
+
+        return buildExecutor("cockple-noti-", coreSize, maxSize, queueCapacity,
+                awaitTerminationSeconds, mdcTaskDecorator, logAndDiscardPolicy());
+    }
+
+    /**
+     * 큐/스레드 포화로 거부된 작업을 예외 전파 없이 버림
+     */
+    private RejectedExecutionHandler logAndDiscardPolicy() {
+        // TODO: 알림 재전송 도입 시, 여기서 drop 대신 outbox/DLQ에 적재해 재시도 훅으로 사용
+        return (runnable, executor) ->
+                log.warn("[NOTIFICATION] 알림 풀 포화로 작업 drop - poolSize={}, activeCount={}, queueSize={}",
+                        executor.getPoolSize(), executor.getActiveCount(), executor.getQueue().size());
+    }
+
+    private ThreadPoolTaskExecutor buildExecutor(String threadNamePrefix, int coreSize, int maxSize,
+                                                 int queueCapacity, int awaitTerminationSeconds,
+                                                 TaskDecorator mdcTaskDecorator,
+                                                 RejectedExecutionHandler rejectedExecutionHandler) {
+        ThreadPoolTaskExecutor executor = new ThreadPoolTaskExecutor();
+        executor.setCorePoolSize(coreSize);
+        executor.setMaxPoolSize(maxSize);
+        executor.setQueueCapacity(queueCapacity);
+        executor.setThreadNamePrefix(threadNamePrefix);
+        executor.setTaskDecorator(mdcTaskDecorator);
+        executor.setRejectedExecutionHandler(rejectedExecutionHandler);
+        // graceful shutdown: 진행 중인 작업이 끝날 때까지 대기
+        executor.setWaitForTasksToCompleteOnShutdown(true);
+        executor.setAwaitTerminationSeconds(awaitTerminationSeconds);
+        // 스프링이 빈 초기화 시 afterPropertiesSet()으로 initialize()를 호출한다.
+        return executor;
+    }
 }

src/main/resources/application.yml

@@ -85,6 +85,20 @@ jwt:
   access-token-validity: 900000
   refresh-token-validity: 1209600000
 
+async:
+  # 실시간 채팅 풀 (저지연 내부 I/O)
+  chat:
+    core-size: 8
+    max-size: 16
+    queue-capacity: 500
+    await-termination-seconds: 30
+  # FCM 알림 풀
+  notification:
+    core-size: 4
+    max-size: 12
+    queue-capacity: 1000
+    await-termination-seconds: 30
+
 fcm:
   fake-latency-ms: 0