grpc协议下UDP通过cloudflare会断？

[AlistairEds]

Xray版本1.4.2

问题描述：
像GTA5这种UDP联机的在线游戏，使用grpc就会无法匹配到其他玩家（能正常连接上游戏服务器，正常上网），但WS就没有问题。
不使用cf时，grpc倒是能连上一两个玩家，也可能是因为裸连线路不好（套CF线路很好），连接效果很差（正常的话至少能连十几个玩家）。

配置：
netch使用socks5代理到v2rayN--->cf--->fallback--->nginx--->grpc

xray日志：
error.log

nginx配置复制的xray-example里的VLESS-grpc

服务器配置：

{
"log": {
"access": "/var/log/xray/access.log",
"error": "/var/log/xray/error.log",
"loglevel": "info"
},
"dns": {
"servers": [
// "https+local://1.1.1.1/dns-query",
// "localhost"
"8.8.8.8",
"8.8.4.4"
]
},
"routing": {
"domainStrategy": "AsIs",
"rules": [
// prevent server local loop
{
"type": "field",
"ip": [
"geoip:private"
],
"outboundTag": "block"
},
{
"type": "field",
"domain": [
"geosite:category-ads-all"
],
"outboundTag": "block"
},
{
"type": "field",
"domain": [
"geosite:cn"
],
"outboundTag": "block"
}
]
},
"inbounds": [
{
"port": 443,
"protocol": "vless",
"settings": {
"clients": [
{
"id": "",
"level": 0,
},
{
"id": "",
"level": 0,
}
],
"decryption": "none",
"fallbacks": [
{
"dest": 8080
},
{
"name": "",
"path": "/", // websocket
"dest": 10956,
"xver": 1
},
{
"name": "",
"path": "/", // vmess+websocket
"dest": 49650,
"xver": 1
},
//fallback to nginx
{
"name": "",
"alpn": "h2",
"dest": 8081,
"xver": 1
},
{
"name": "",
"dest": 8080,
"xver": 1
}
]
},
"streamSettings": {
"network": "tcp",
"security": "xtls",
"xtlsSettings": {
"alpn": [
"h2",
"http/1.1"
],
"certificates": [
{
"certificateFile": "/",
"keyFile": "/"
}
]
}
},
"sniffing": {
"enabled": true,
"destOverride": ["http","tls"]
}
},
{
"port": 10956,
"listen": "127.0.0.1",
"protocol": "vless",
"settings": {
"clients": [
{
"id": "************",
"level": 0,
"email": ""
}
],
"decryption": "none"
},
"streamSettings": {
"network": "ws",
"security": "none",
"wsSettings": {
"acceptProxyProtocol": true,
"path": "/"
}
}
},
{
"port": 49650,
"protocol": "vmess",
"settings": {
"clients": [
{
"email": "",
"id": "",
"level": 0,
"alterId": 0
}
]
},
"streamSettings": {
"network": "ws",
"wsSettings": {
"acceptProxyProtocol": true,
"path": "/"
}
},
"sniffing": {
"enabled": true,
"destOverride": [
"http",
"tls"
]
}
},
{
"port": 10957,
"listen": "127.0.0.1",
"protocol": "vmess",
"settings": {
"clients": [
{
"id": "*********", // UUID
"email": "",
"alterId": 0
}
]
//"decryption": "none"
},
"streamSettings": {
"network": "grpc",
"grpcSettings": {
"serviceName": ********// ServiceName
}
},
"sniffing": {
"enabled": true,
"destOverride": [
"http",
"tls"
]
}
}
],
"outbounds": [
{
"protocol": "freedom"
},
{
"protocol": "blackhole",
"tag":"block"
}
]
}

[AlistairEds]

Nginx的error.log中有大量的“upstream prematurely closed connection while reading upstream”，这是CF关闭了连接？

[xqzr]

Nginx的error.log中有大量的“upstream prematurely closed connection while reading upstream”，这是CF关闭了连接？

是的，cf 限制 client_max_body_size（100M），因此，不建议 gRPC 通过 cf。

[MuenYu]

跟 UDP 是没啥关系的（毕竟 udp over tcp）。
在 套 cf 使用 grpc 的情况下，似乎 cf 会自动断掉连接时长较长的 tcp。

本地测试是使用代理 ssh 连接主机，大概 30s-5min 会断连。

在此附上以前关于这一点的发现。