Add first-chunk timeout

src/main/kotlin/LargeLanguageModels.kt

@@ -44,6 +44,7 @@ object LargeLanguageModels {
 
     fun reload() {
         val timeout = PluginConfig.timeout.milliseconds
+        val firstChunkTimeout = PluginConfig.firstChunkTimeout.milliseconds
 
         // 初始化聊天模型
         if (PluginConfig.openAiApi.isNotBlank() && PluginConfig.openAiToken.isNotBlank()) {
@@ -51,6 +52,7 @@ object LargeLanguageModels {
                 baseUrl = PluginConfig.openAiApi,
                 token = PluginConfig.openAiToken,
                 timeout = timeout,
+                firstChunkTimeout = firstChunkTimeout,
                 extraBody = parseExtraBody(PluginConfig.chatModelExtraBody)
             )
         }
@@ -61,6 +63,7 @@ object LargeLanguageModels {
                 baseUrl = PluginConfig.reasoningModelApi,
                 token = PluginConfig.reasoningModelToken,
                 timeout = timeout,
+                firstChunkTimeout = firstChunkTimeout,
                 extraBody = parseExtraBody(PluginConfig.reasoningModelExtraBody)
             )
         }
@@ -71,6 +74,7 @@ object LargeLanguageModels {
                 baseUrl = PluginConfig.visualModelApi,
                 token = PluginConfig.visualModelToken,
                 timeout = timeout,
+                firstChunkTimeout = firstChunkTimeout,
                 extraBody = parseExtraBody(PluginConfig.visualModelExtraBody)
             )
         }

src/main/kotlin/ModelService.kt

@@ -14,6 +14,7 @@ import kotlinx.coroutines.currentCoroutineContext
 import kotlinx.coroutines.flow.Flow
 import kotlinx.coroutines.flow.flow
 import kotlinx.coroutines.isActive
+import kotlinx.coroutines.withTimeout
 import kotlinx.serialization.json.*
 import kotlin.time.Duration
 
@@ -21,15 +22,16 @@ class ModelService(
     val baseUrl: String,
     val token: String,
     val timeout: Duration,
+    val firstChunkTimeout: Duration,
     val extraBody: JsonObject? = null
 ) {
     val httpClient: HttpClient by lazy {
         HttpClient(OkHttp) {
             install(HttpTimeout) {
-                val millis = timeout.inWholeMilliseconds
+                // 总请求/socket 超时保持长值，允许慢速流式输出；连接握手则用短超时。
-                requestTimeoutMillis = millis
+                requestTimeoutMillis = timeout.inWholeMilliseconds
-                connectTimeoutMillis = millis
+                socketTimeoutMillis = timeout.inWholeMilliseconds
-                socketTimeoutMillis = millis
+                connectTimeoutMillis = firstChunkTimeout.inWholeMilliseconds
             }
             defaultRequest {
                 url(baseUrl)
@@ -66,17 +68,35 @@ class ModelService(
             }.let { response ->
                 val channel: ByteReadChannel = response.body()
                 try {
-                    while (currentCoroutineContext().isActive && !channel.isClosedForRead) {
+                    // 首块 data: 必须在 firstChunkTimeout 内到达，否则抛 TimeoutCancellationException
-                        val line = channel.readUTF8Line() ?: continue
+                    // 走 JChatGPT 的重试流程；之后的流式读取不再有应用层超时，由 socketTimeoutMillis 兜底。
-                        when {
+                    val firstDataLine: String? = withTimeout(firstChunkTimeout) {
-                            line.startsWith("data: [DONE]") -> break
+                        var found: String? = null
-                            line.startsWith("data: ") -> {
+                        while (currentCoroutineContext().isActive && !channel.isClosedForRead) {
-                                val chunk = json.decodeFromString<ChatCompletionChunk>(
+                            val line = channel.readUTF8Line() ?: continue
-                                    line.removePrefix("data: ")
+                            if (line.startsWith("data: ")) {
-                                )
+                                found = line
-                                emit(chunk)
+                                break
+                            }
+                            // 心跳/空行/注释行，不计为首块，继续等
+                        }
+                        found
+                    }
+
+                    if (firstDataLine != null) {
+                        if (!firstDataLine.startsWith("data: [DONE]")) {
+                            emit(json.decodeFromString(firstDataLine.removePrefix("data: ")))
+
+                            while (currentCoroutineContext().isActive && !channel.isClosedForRead) {
+                                val line = channel.readUTF8Line() ?: continue
+                                when {
+                                    line.startsWith("data: [DONE]") -> break
+                                    line.startsWith("data: ") -> {
+                                        emit(json.decodeFromString(line.removePrefix("data: ")))
+                                    }
+                                    else -> continue
+                                }
                             }
-                            else -> continue
                         }
                     }
                 } finally {

src/main/kotlin/PluginConfig.kt

@@ -77,9 +77,12 @@ object PluginConfig : AutoSavePluginConfig("Config") {
     @ValueDescription("群荣誉等级权限门槛，达到这个等级相当于自动拥有对话权限。")
     val temperaturePermission: Int by value(50)
 
-    @ValueDescription("等待响应超时时间，单位毫秒，默认60秒")
+    @ValueDescription("等待响应超时时间（整个请求的总超时与socket读超时），单位毫秒，默认60秒")
     val timeout: Long by value(60000L)
 
+    @ValueDescription("首块响应超时时间，单位毫秒，默认10秒。若连接建立后在此时间内没收到首块data:则中断走重试")
+    val firstChunkTimeout: Long by value(10000L)
+
     @Deprecated("使用外部文件而不是在配置文件内保存提示词")
     @ValueDescription("系统提示词，该字段已弃用，使用提示词文件而不是在这里修改")
     var prompt: String by value("你是一个乐于助人的助手")