采用更保守的重连时间策略，防止雪崩

blivedm

@@ -1 +1 @@
-Subproject commit b2c26b5f16ee4fd0f4a1ded8c94f9c37e1e2ff6a
+Subproject commit 5525bf5419e282f4d0cc7b03cc1e0fd22cee7a73

frontend/src/api/chat/ChatClientOfficialBase/index.js

@@ -51,6 +51,7 @@ export default class ChatClientOfficialBase {
     this.needInitRoom = true
     this.websocket = null
     this.retryCount = 0
+    this.totalRetryCount = 0
     this.isDestroying = false
     this.heartbeatTimerId = null
     this.receiveTimeoutTimerId = null
@@ -185,15 +186,17 @@ export default class ChatClientOfficialBase {
       return
     }
     this.retryCount++
-    console.warn('掉线重连中', this.retryCount)
+    this.totalRetryCount++
+    console.warn(`掉线重连中 retryCount=${this.retryCount}, totalRetryCount=${this.totalRetryCount}`)
     window.setTimeout(this.wsConnect.bind(this), this.getReconnectInterval())
   }
 
   getReconnectInterval() {
-    return Math.min(
-      1000 + ((this.retryCount - 1) * 2000),
-      10 * 1000
-    )
+    // 不用retryCount了，防止意外的连接成功，导致retryCount重置
+    let interval = Math.min(1000 + ((this.totalRetryCount - 1) * 2000), 20 * 1000)
+    // 加上随机延迟，防止同时请求导致雪崩
+    interval += Math.random() * 3000
+    return interval
   }
 
   onWsMessage(event) {

frontend/src/api/chat/ChatClientRelay.js

@@ -25,6 +25,7 @@ export default class ChatClientRelay {
 
     this.websocket = null
     this.retryCount = 0
+    this.totalRetryCount = 0
     this.isDestroying = false
     this.receiveTimeoutTimerId = null
   }
@@ -92,15 +93,18 @@ export default class ChatClientRelay {
     if (this.isDestroying) {
       return
     }
-    console.warn(`掉线重连中${++this.retryCount}`)
+    this.retryCount++
+    this.totalRetryCount++
+    console.warn(`掉线重连中 retryCount=${this.retryCount}, totalRetryCount=${this.totalRetryCount}`)
     window.setTimeout(this.wsConnect.bind(this), this.getReconnectInterval())
   }
 
   getReconnectInterval() {
-    return Math.min(
-      1000 + ((this.retryCount - 1) * 2000),
-      10 * 1000
-    )
+    // 不用retryCount了，防止意外的连接成功，导致retryCount重置
+    let interval = Math.min(1000 + ((this.totalRetryCount - 1) * 2000), 20 * 1000)
+    // 加上随机延迟，防止同时请求导致雪崩
+    interval += Math.random() * 3000
+    return interval
   }
 
   onWsMessage(event) {

frontend/src/views/Room.vue

@@ -76,7 +76,12 @@ export default {
   },
   mounted() {
     if (document.visibilityState === 'visible') {
-      this.init()
+      if (this.roomKeyValue === null) {
+        this.init()
+      } else {
+        // 正式房间要随机延迟加载，防止同时请求导致雪崩
+        window.setTimeout(this.init, Math.random() * 3000)
+      }
     } else {
       // 当前窗口不可见，延迟到可见时加载，防止OBS中一次并发太多请求（OBS中浏览器不可见时也会加载网页，除非显式设置）
       document.addEventListener('visibilitychange', this.onVisibilityChange)
@@ -496,7 +501,7 @@ export default {
             // 获取失败了默认为0
             img.onerror = resolve
             // 超时保底
-            setTimeout(resolve, 5000)
+            window.setTimeout(resolve, 5000)
             img.src = urlInClosure
           }
         ))

services/chat.py

@@ -2,6 +2,7 @@
 import asyncio
 import enum
 import logging
+import random
 import uuid
 from typing import *
 
@@ -10,7 +11,6 @@ import api.open_live as api_open_live
 import blivedm.blivedm as blivedm
 import blivedm.blivedm.models.open_live as dm_open_models
 import blivedm.blivedm.models.web as dm_web_models
-import blivedm.blivedm.utils as dm_utils
 import config
 import services.avatar
 import services.translate
@@ -114,7 +114,12 @@ class LiveClientManager:
         client_room_manager.del_room(room_key)
 
 
-RECONNECT_POLICY = dm_utils.make_linear_retry_policy(1, 2, 10)
+def _get_reconnect_interval(_retry_count: int, total_retry_count: int):
+    # 不用retry_count了，防止意外的连接成功，导致retry_count重置
+    interval = min(1 + (total_retry_count - 1) * 2, 20)
+    # 加上随机延迟，防止同时请求导致雪崩
+    interval += random.uniform(0, 3)
+    return interval
 
 
 class WebLiveClient(blivedm.BLiveClient):
@@ -128,7 +133,7 @@ class WebLiveClient(blivedm.BLiveClient):
             session=utils.request.http_session,
             heartbeat_interval=self.HEARTBEAT_INTERVAL,
         )
-        self.set_reconnect_policy(RECONNECT_POLICY)
+        self.set_reconnect_policy(_get_reconnect_interval)
 
     @property
     def room_key(self):
@@ -158,7 +163,7 @@ class OpenLiveClient(blivedm.OpenLiveClient):
             session=utils.request.http_session,
             heartbeat_interval=self.HEARTBEAT_INTERVAL,
         )
-        self.set_reconnect_policy(RECONNECT_POLICY)
+        self.set_reconnect_policy(_get_reconnect_interval)
 
     @property
     def room_key(self):