5G网络下，生成内容会中断

[liuchao0807]

📦 部署环境

Docker

📌 软件版本

1.28.5

💻 系统环境

iOS

🌐 浏览器

Chrome

🐛 问题描述

在5G网络条件下，生成长的内容几乎必现中断，导致生成不全
而切到WiFi下，则能顺利生成
除了网络条件，其余条件均一致

📷 复现步骤

见视频

1039381501.mp4

🚦 期望结果

任何网络条件下，都能生成完整回复

📝 补充信息

会不会和丢包率有关？
或者和超时时间有关？

[lobehubbot]

👀 @liuchao0807

Thank you for raising an issue. We will investigate into the matter and get back to you as soon as possible.
Please make sure you have given us as much context as possible.
非常感谢您提交 issue。我们会尽快调查此事，并尽快回复您。 请确保您已经提供了尽可能多的背景信息。

[lobehubbot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

📦 Deployment environment

Docker

📌 Software version

1.28.5

💻 System environment

iOS

🌐 Browser

Chrome

🐛 Problem description

Under 5G network conditions, generating long content will almost inevitably be interrupted, resulting in incomplete generation.
And switch to WiFi, you can successfully generate
Except for network conditions, all other conditions are the same.

📷 Steps to reproduce

See video

Uploading F9D0042B-7A52-4D64-8DB8-12CEF90709BC.mp4…

🚦 Expected results

Ability to generate complete responses under any network conditions

📝 Supplementary information

Could it be related to packet loss rate?
Or is it related to the timeout?

[Kac001]

这肯定是你部署的服务器网络问题

[lobehubbot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

This is definitely a problem with the server network you deployed

[liuchao0807]

这肯定是你部署的服务器网络问题

额，我知道肯定和网络有关，但是不知道你看我的实验视频没有，唯一的变量是我把手机的5G网络，变成了手机连wifi
服务器的网络完全没变啊

[lobehubbot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

This must be a network problem on the server you deployed

Um, I know it must be related to the network, but I don’t know if you saw my experimental video. The only variable is that I changed the 5G network of the mobile phone into the mobile phone connected to wifi.
The server network has not changed at all.

[liuchao0807]

这肯定是你部署的服务器网络问题

变化的是客户端的网络连接

[lobehubbot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

This must be a network problem on the server you deployed

What changes is the client’s network connection

[Kac001]

你要明白，wifi和5g走的路由不是一样的，5g的trace下路由看看是不是那个节点出问题。

[lobehubbot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

You have to understand that the routes taken by WiFi and 5G are not the same. Run the 5G trace route to see if there is a problem with that node.

[liuchao0807]

你要明白，wifi和5g走的路由不是一样的，5g的trace下路由看看是不是那个节点出问题。

我是搞嵌入式的，对网络这块不是很熟哈
以我的认知，有几个疑问

1. 5G和wifi即使路由不同，但是都可以连接，说明链路没问题
2. 5G网络条件下，只会生成一半，是否说明网络中途有丢包？或者中途连接断了？
3. 客户端和服务端的连接是UDP还是TCP， 如果中途丢包或断连，是否可以续传？

[lobehubbot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

You need to understand that the routes taken by WiFi and 5G are not the same. Run the 5G trace route to see if there is a problem with that node.

I work in the embedded field and am not very familiar with the network.
From my understanding, I have several questions

1. Even if the routes are different, both 5G and wifi can be connected, which means there is no problem with the link.
2. Under 5G network conditions, only half of the packets are generated. Does this mean that there is packet loss in the network? Or the connection was lost midway?
3. Is the connection between the client and the server UDP or TCP? If the packet is lost or disconnected midway, can the transmission be resumed?

[Kac001]

问题在5g到服务器这段链路上，这种大概率是中途有节点延迟高甚至丢包断连，最好试试换运营商或者用4G，应该是用的http tcp协议

[lobehubbot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

The problem lies in the link from 5G to the server. There is a high probability that there is a node with high delay or even packet loss and disconnection in the middle. It is best to try changing operators or using 4G, which should use the http tcp protocol.

[liuchao0807]

lobe

问题在5g到服务器这段链路上，这种大概率是中途有节点延迟高甚至丢包断连，最好试试换运营商或者用4G，应该是用的http tcp协议

如果是tcp协议，那理论上应该是有重传机制的？除非是超时了？
对于这种丢包或者延时大的网络条件，是否可以增加超时时间，从而增加传输可靠性？

[lobehubbot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

lobe

The problem is on the link from 5g to the server. There is a high probability that there is a node with high delay or even packet loss and disconnection in the middle. It is best to try changing operators or using 4G. It should use the http tcp protocol.

If it is a tcp protocol, then in theory there should be a retransmission mechanism? Unless it's a timeout?
For such network conditions with large packet loss or delay, can the timeout be increased to increase transmission reliability?

[Kac001]

这就要开发者回答这个问题了

[lobehubbot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

It’s up to developers to answer this question

[Sun-drenched]

应该还是网络问题——不仅是与部署lobechat的服务器之间的网络连接，还有可能和服务器（或者客户端）与API接口（看你有没有开“客户端请求模式”）的网络连接有关。

如果是使用的是二道贩子那的中转接口，可以尝试换一家，有些二道贩子的接口在国内网络环境下不稳定。

[lobehubbot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

It should still be a network problem - not only the network connection with the server where lobechat is deployed, but also the network connection between the server (or client) and the API interface (see whether you have turned on "client request mode").

If you are using the transit interface from a second-tier dealer, you can try to change it. Some second-tier dealers’ interfaces are unstable in the domestic network environment.

[liuchao0807]

应该还是网络问题——不仅是与部署lobechat的服务器之间的网络连接，还有可能和服务器（或者客户端）与API接口（看你有没有开“客户端请求模式”）的网络连接有关。

如果是使用的是二道贩子那的中转接口，可以尝试换一家，有些二道贩子的接口在国内网络环境下不稳定。

没有使用客户端请求模式，所以实际发起请求的都是我的服务器，对api请求来说，在5G和WiFi模式下是一样的网络条件

[lobehubbot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

It should still be a network problem - not only the network connection with the server where lobechat is deployed, but also the network connection between the server (or client) and the API interface (see whether you have turned on "client request mode") .

If you are using the transit interface from a second-tier dealer, you can try to change it. Some second-tier dealers’ interfaces are unstable in the domestic network environment.

The client request mode is not used, so my server actually initiates the request. For api requests, the network conditions are the same in 5G and WiFi modes.

[drsanwujiang]

分别使用电信和联通 5G 网络测试，Lobe Chat 工作正常。

这种情况通常建议考虑服务器的网络问题，例如服务器所在地是否距离太远（位于美国、欧洲等）。可以使用一台电脑连接手机热点，在电脑上使用 tracert 等工具查看一下路由。

最后，此问题应该与 Lobe Chat 本身无关，建议移至讨论区进行讨论。

[lobehubbot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

Tested using China Telecom and China Unicom 5G networks respectively, Lobe Chat works normally.

In this case, it is usually recommended to consider the server's network problems, such as whether the server is located too far away (in the United States, Europe, etc.). You can use a computer to connect to the mobile phone hotspot, and use tools such as tracert on the computer to check the route.

Finally, this issue should have nothing to do with Lobe Chat itself and it is recommended to move it to the discussion forum for discussion.

[liuchao0807]

分别使用电信和联通 5G 网络测试，Lobe Chat 工作正常。

这种情况通常建议考虑服务器的网络问题，例如服务器所在地是否距离太远（位于美国、欧洲等）。可以使用一台电脑连接手机热点，在电脑上使用 tracert 等工具查看一下路由。

最后，此问题应该与 Lobe Chat 本身无关，建议移至讨论区进行讨论。

服务器是香港的
提出这个问题的本意并不是说lobe chat本身有问题，而是看是否可以增加一些技术手段，应对稍差一些的网络条件，提高鲁棒性

[liuchao0807]

理论上浏览器针对 sse 请求应该会做自动重试的，不太确定还需要的鲁棒性是什么？

针对我发的视频这种断连的情况，从代码上可以解释吗，什么情况下会出？比如丢包率达到多少？或者连续丢包多长时间？ 这种断连从技术上可以规避吗？

[arvinxx]

代码上似乎无法解释，有没有试过换个手机网络？

因为我不是搞网络出身，所以不知道丢包率、丢包时长这些更加底层的数据。我们现在的通信协议是 SSE https://developer.mozilla.org/zh-CN/docs/Web/API/Server-sent_events/Using_server-sent_events

我们目前的请求会符合标准的 SSE 格式，不过没有加 retry 参数。这是我目前想到的唯一可尝试的方式

retry
重新连接的时间。如果与服务器的连接丢失，浏览器将等待指定的时间，然后尝试重新连接。这必须是一个整数，以毫秒为单位指定重新连接的时间。如果指定了一个非整数值，该字段将被忽略。

如果你愿意来试的话，我可以告诉你怎么改来测试验证。

[liuchao0807]

代码上似乎无法解释，有没有试过换个手机网络？

因为我不是搞网络出身，所以不知道丢包率、丢包时长这些更加底层的数据。我们现在的通信协议是 SSE https://developer.mozilla.org/zh-CN/docs/Web/API/Server-sent_events/Using_server-sent_events

我们目前的请求会符合标准的 SSE 格式，不过没有加 retry 参数。这是我目前想到的唯一可尝试的方式

retry

重新连接的时间。如果与服务器的连接丢失，浏览器将等待指定的时间，然后尝试重新连接。这必须是一个整数，以毫秒为单位指定重新连接的时间。如果指定了一个非整数值，该字段将被忽略。

如果你愿意来试的话，我可以告诉你怎么改来测试验证。

我可以试一下，不过我是搞嵌入式的，基本只用过c语言，玩这个项目也只是使用了docker镜像。如何修改并编译代码可能需要你这边指导下

[arvinxx]

lobe-chat/src/libs/agent-runtime/utils/streams/protocol.ts

Lines 95 to 97 in d310931

	controller.enqueue(`id: ${id}\n`);
	controller.enqueue(`event: ${type}\n`);
	controller.enqueue(`data: ${JSON.stringify(data)}\n\n`);

这里是构造 SSE 输出的代码，我理解改成这样可能就行？

controller.enqueue(`data: ${JSON.stringify(data)}\n`);
controller.enqueue(`retry: 300\n\n`);  // 300ms 重试

然后本地 dev 的话，需要先装 node 。

// 安装 pnpm
1.  npm i -g pnpm 
// 在项目根目录安装依赖
2. pnpm i
// 启动项目
3. npm run dev

然后等待项目编译并启动；

如果dev测试没问题，要打 docker 镜像测试的话，执行 npm run self-hosting:docker-cn@database 即可打出本地镜像

lobe-chat/package.json

Lines 55 to 57 in d310931

	"self-hosting:docker": "docker build -t lobe-chat:local .",
	"self-hosting:docker-cn": "docker build -t lobe-chat-local --build-arg USE_CN_MIRROR=true .",
	"self-hosting:docker-cn@database": "docker build -t lobe-chat-database-local -f Dockerfile.database --build-arg USE_CN_MIRROR=true .",

[liuchao0807]

lobe-chat/src/libs/agent-runtime/utils/streams/protocol.ts

Lines 95 to 97 in d310931

	controller.enqueue(`id: ${id}\n`);
	controller.enqueue(`event: ${type}\n`);
	controller.enqueue(`data: ${JSON.stringify(data)}\n\n`);

这里是构造 SSE 输出的代码，我理解改成这样可能就行？

controller.enqueue(`data: ${JSON.stringify(data)}\n`);
controller.enqueue(`retry: 300\n\n`);  // 300ms 重试

然后本地 dev 的话，需要先装 node 。

// 安装 pnpm
1.  npm i -g pnpm 
// 在项目根目录安装依赖
2. pnpm i
// 启动项目
3. npm run dev

然后等待项目编译并启动；

如果dev测试没问题，要打 docker 镜像测试的话，执行 npm run self-hosting:docker-cn@database 即可打出本地镜像

lobe-chat/package.json

Lines 55 to 57 in d310931

	"self-hosting:docker": "docker build -t lobe-chat:local .",
	"self-hosting:docker-cn": "docker build -t lobe-chat-local --build-arg USE_CN_MIRROR=true .",
	"self-hosting:docker-cn@database": "docker build -t lobe-chat-database-local -f Dockerfile.database --build-arg USE_CN_MIRROR=true .",

今天花了点时间，做了以下几个实验

1. 本地dev启动项目，未修改的情况下，也无法复现问题
2. 使用非database版本的docker，也不能复现问题
3. 只有使用database版本的docker，才可以复现问题
   感觉这个问题不止一个影响因素，我已经用回非database版了，没太多时间折腾。。。
   这个问题暂时就到这把，如果你还有什么想法，可以找我复现抓取log

[lobehubbot]

Bot detected the issue body's language is not English, translate it automatically. 👯👭🏻🧑‍🤝‍🧑👫🧑🏿‍🤝‍🧑🏻👩🏾‍🤝‍👨🏿👬🏿

---

Tested using China Telecom and China Unicom 5G networks respectively, Lobe Chat works normally.

In this case, it is usually recommended to consider the network problem of the server, such as whether the server is located too far away (located in the United States, Europe, etc.). You can use a computer to connect to the mobile phone hotspot, and use tools such as tracert on the computer to check the route.

Finally, this issue should have nothing to do with Lobe Chat itself, and it is recommended to move it to the discussion area for discussion.

The server is from Hong Kong
The original intention of raising this question is not to say that there is a problem with Lobe Chat itself, but to see if some technical means can be added to cope with slightly worse network conditions and improve robustness.