whai/cherry-studio
1
1
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

fix: messages pause bug (#3343)

Browse Source 
* refactor: Simplify message resend logic and enhance abort controller handling

- Updated MessageMenubar to streamline message resend functionality.
- Improved abort controller management in BaseProvider and related services.
- Adjusted sendMessage to handle both single and multiple assistant messages.
- Enhanced logging for better debugging and tracking of message flow.

* feat: Enhance message handling and queue management

- Updated Inputbar to include mentions in dispatched messages.
- Introduced appendMessage action to manage message insertion at specific positions in the state.
- Improved sendMessage logic to handle mentions and maintain message order.
- Refactored getTopicQueue to accept options for better queue configuration.

* refactor: Improve abort handling and message operations

- Refactored useMessageOperations to streamline message pausing logic.
- Enhanced abort controller in BaseProvider to handle abort events more effectively.
- Updated OpenAIProvider to utilize new abort handling mechanism.
- Adjusted fetchChatCompletion to set message status based on abort conditions.
- Improved message dispatching in sendMessage for better queue management.

* refactor: Enhance signal promise handling in BaseProvider and OpenAIProvider

- Updated signal handling in BaseProvider to use a structured signalPromise object for better clarity and management.
- Adjusted error handling in OpenAIProvider to correctly catch and throw errors from the signalPromise.
- Improved overall abort handling logic to ensure robust message operations.

* fix:lint
This commit is contained in:
MyPrototypeWhat 2025-03-14 17:57:33 +08:00 committed by GitHub
parent aa6ecb4814
commit 8faececa4c
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
12 changed files with 194 additions and 108 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

43
src/renderer/src/hooks/useMessageOperations.ts
View File

@ -158,34 +158,35 @@ export function useMessageOperations(topic: Topic) {
/**
* 暂停消息生成
*/
const pauseMessage = useCallback(
// 存的是用户消息的id，也就是助手消息的askId
async (message: Message) => {
// 1. 调用 abort
message.askId && abortCompletion(message.askId)
// const pauseMessage = useCallback(
// // 存的是用户消息的id，也就是助手消息的askId
// async (message: Message) => {
// // 1. 调用 abort
// 2. 更新消息状态
await editMessage(message.id, { status: 'paused', content: message.content })
// // 2. 更新消息状态,
// // await editMessage(message.id, { status: 'paused', content: message.content })
// 3.更改loading状态
dispatch(setTopicLoading({ topicId: message.topicId, loading: false }))
// // 3.更改loading状态
// dispatch(setTopicLoading({ topicId: message.topicId, loading: false }))
// 4. 清理流式消息
clearStreamMessageAction(message.id)
},
[editMessage, dispatch, clearStreamMessageAction]
)
// // 4. 清理流式消息
// // clearStreamMessageAction(message.id)
// },
// [editMessage, dispatch, clearStreamMessageAction]
// )
const pauseMessages = useCallback(async () => {
// 暂停的消息不需要在这更改status,通过catch判断abort错误之后设置message.status
const streamMessages = store.getState().messages.streamMessagesByTopic[topic.id]
if (!streamMessages) return
// 不需要重复暂停
const askIds = [...new Set(Object.values(streamMessages).map((m) => m?.askId))]
if (streamMessages) {
const streamMessagesList = Object.values(streamMessages).filter((msg) => msg?.askId && msg?.id)
for (const message of streamMessagesList) {
message && (await pauseMessage(message))
}
for (const askId of askIds) {
askId && abortCompletion(askId)
}
}, [pauseMessage, topic.id])
dispatch(setTopicLoading({ topicId: topic.id, loading: false }))
}, [topic.id, dispatch])
/**
* 恢复/重发消息
@ -213,7 +214,7 @@ export function useMessageOperations(topic: Topic) {
clearStreamMessage: clearStreamMessageAction,
createNewContext,
clearTopicMessages: clearTopicMessagesAction,
pauseMessage,
// pauseMessage,
pauseMessages,
resumeMessage
}

6
src/renderer/src/pages/home/Inputbar/Inputbar.tsx
View File

@ -175,7 +175,11 @@ const Inputbar: FC<Props> = ({ assistant: _assistant, setActiveTopic, topic }) =
userMessage.usage = await estimateMessageUsage(userMessage)
currentMessageId.current = userMessage.id
dispatch(_sendMessage(userMessage, assistant, topic))
dispatch(
_sendMessage(userMessage, assistant, topic, {
mentions: mentionModels
})
)
// Clear input
setText('')

8
src/renderer/src/pages/home/Messages/MessageMenubar.tsx
View File

@ -92,10 +92,10 @@ const MessageMenubar: FC<Props> = (props) => {
// Resend all grouped messages
if (!isEmpty(groupdMessages)) {
for (const assistantMessage of groupdMessages) {
const _model = assistantMessage.model || assistantModel
await resendMessage({ ...assistantMessage, model: _model }, assistant)
}
// for (const assistantMessage of groupdMessages) {
// const _model = assistantMessage.model || assistantModel
await resendMessage(message, assistant)
// }
return
}

21
src/renderer/src/providers/AnthropicProvider.ts
View File

@ -10,7 +10,6 @@ import { isReasoningModel } from '@renderer/config/models'
import { getStoreSetting } from '@renderer/hooks/useSettings'
import i18n from '@renderer/i18n'
import { getAssistantSettings, getDefaultModel, getTopNamingModel } from '@renderer/services/AssistantService'
import { EVENT_NAMES } from '@renderer/services/EventService'
import {
filterContextMessages,
filterEmptyMessages,
@ -241,13 +240,13 @@ export default class AnthropicProvider extends BaseProvider {
return new Promise<void>((resolve, reject) => {
const toolCalls: ToolUseBlock[] = []
let hasThinkingContent = false
const stream = this.sdk.messages
this.sdk.messages
.stream({ ...body, stream: true }, { signal })
.on('text', (text) => {
if (window.keyv.get(EVENT_NAMES.CHAT_COMPLETION_PAUSED)) {
stream.controller.abort()
return resolve()
}
// if (window.keyv.get(EVENT_NAMES.CHAT_COMPLETION_PAUSED)) {
// stream.controller.abort()
// return resolve()
// }
if (time_first_token_millsec == 0) {
time_first_token_millsec = new Date().getTime() - start_time_millsec
@ -357,15 +356,13 @@ export default class AnthropicProvider extends BaseProvider {
resolve()
})
.on('error', (error) => reject(error))
.on('abort', () => {
reject(new Error('Request was aborted.'))
})
})
}
await processStream(body, 0)
.catch((error) => {
// 不加这个错误抛不出来
throw error
})
.finally(cleanup)
await processStream(body, 0).finally(cleanup)
}
/**

37
src/renderer/src/providers/BaseProvider.ts
View File

@ -160,24 +160,45 @@ export default abstract class BaseProvider {
)
}
protected createAbortController(messageId?: string) {
protected createAbortController(messageId?: string, isAddEventListener?: boolean) {
const abortController = new AbortController()
const abortFn = () => abortController.abort()
if (messageId) {
addAbortController(messageId, () => abortController.abort())
addAbortController(messageId, abortFn)
}
const cleanup = () => {
if (messageId) {
removeAbortController(messageId)
signalPromise.resolve?.(undefined)
removeAbortController(messageId, abortFn)
}
}
const signalPromise: {
resolve: (value: unknown) => void
promise: Promise<unknown>
} = {
resolve: () => {},
promise: Promise.resolve()
}
abortController.signal.addEventListener('abort', () => {
// 兼容
cleanup()
})
if (isAddEventListener) {
signalPromise.promise = new Promise((resolve, reject) => {
signalPromise.resolve = resolve
if (abortController.signal.aborted) {
reject(new Error('Request was aborted.'))
}
// 捕获abort事件,有些abort事件必须
abortController.signal.addEventListener('abort', () => {
reject(new Error('Request was aborted.'))
})
})
return {
abortController,
cleanup,
signalPromise
}
}
return {
abortController,
cleanup

2
src/renderer/src/providers/GeminiProvider.ts
View File

@ -326,7 +326,7 @@ export default class GeminiProvider extends BaseProvider {
})
const newChat = geminiModel.startChat({ history })
const newStream = await newChat.sendMessageStream(fcRespParts, { signal })
await processStream(newStream, idx + 1).finally(cleanup)
await processStream(newStream, idx + 1)
}
}

8
src/renderer/src/providers/OpenAIProvider.ts
View File

@ -357,7 +357,6 @@ export default class OpenAIProvider extends BaseProvider {
}
const userMessages: ChatCompletionMessageParam[] = []
const _messages = filterUserRoleStartMessages(
filterEmptyMessages(filterContextMessages(takeRight(messages, contextCount + 1)))
)
@ -414,7 +413,7 @@ export default class OpenAIProvider extends BaseProvider {
let time_first_content_millsec = 0
const start_time_millsec = new Date().getTime()
const lastUserMessage = _messages.findLast((m) => m.role === 'user')
const { abortController, cleanup } = this.createAbortController(lastUserMessage?.id)
const { abortController, cleanup, signalPromise } = this.createAbortController(lastUserMessage?.id, true)
const { signal } = abortController
mcpTools = filterMCPTools(mcpTools, lastUserMessage?.enabledMCPs)
@ -425,7 +424,6 @@ export default class OpenAIProvider extends BaseProvider {
) as ChatCompletionMessageParam[]
const toolResponses: MCPToolResponse[] = []
const processStream = async (stream: any, idx: number) => {
if (!isSupportStreamOutput()) {
const time_completion_millsec = new Date().getTime() - start_time_millsec
@ -593,6 +591,10 @@ export default class OpenAIProvider extends BaseProvider {
)
await processStream(stream, 0).finally(cleanup)
// 捕获signal的错误
await signalPromise?.promise?.catch((error) => {
throw error
})
}
/**

16
src/renderer/src/services/ApiService.ts
View File

@ -34,12 +34,6 @@ export async function fetchChatCompletion({
const webSearchProvider = WebSearchService.getWebSearchProvider()
const AI = new AiProvider(provider)
// store.dispatch(setGenerating(true))
// onResponse({ ...message })
// addAbortController(message.askId ?? message.id)
try {
let _messages: Message[] = []
let isFirstChunk = true
@ -70,7 +64,6 @@ export async function fetchChatCompletion({
}
const allMCPTools = await window.api.mcp.listTools()
await AI.completions({
messages: filterUsefulMessages(messages),
assistant,
@ -127,9 +120,12 @@ export async function fetchChatCompletion({
}
}
} catch (error: any) {
if (isAbortError(error)) return
message.status = 'error'
message.error = formatMessageError(error)
if (isAbortError(error)) {
message.status = 'paused'
} else {
message.status = 'error'
message.error = formatMessageError(error)
}
}
// Emit chat completion event

2
src/renderer/src/services/MessagesService.ts
View File

@ -53,7 +53,7 @@ export function filterEmptyMessages(messages: Message[]): Message[] {
}
export function filterUsefulMessages(messages: Message[]): Message[] {
const _messages = messages
const _messages = [...messages]
const groupedMessages = getGroupedMessages(messages)
Object.entries(groupedMessages).forEach(([key, messages]) => {

122
src/renderer/src/store/messages.ts
View File

@ -6,6 +6,7 @@ import { EVENT_NAMES, EventEmitter } from '@renderer/services/EventService'
import { getAssistantMessage, resetAssistantMessage } from '@renderer/services/MessagesService'
import type { AppDispatch, RootState } from '@renderer/store'
import type { Assistant, Message, Topic } from '@renderer/types'
import { Model } from '@renderer/types'
import { clearTopicQueue, getTopicQueue, waitForTopicQueue } from '@renderer/utils/queue'
import { throttle } from 'lodash'
@ -105,6 +106,29 @@ const messagesSlice = createSlice({
state.messagesByTopic[topicId].push(messages)
}
},
appendMessage: (
state,
action: PayloadAction<{ topicId: string; messages: Message | Message[]; position?: number }>
) => {
const { topicId, messages, position } = action.payload
if (!state.messagesByTopic[topicId]) {
state.messagesByTopic[topicId] = []
}
// 确保消息数组存在并且拿到引用
const messagesList = state.messagesByTopic[topicId]
// 要插入的消息
const messagesToInsert = Array.isArray(messages) ? messages : [messages]
if (position !== undefined && position >= 0 && position <= messagesList.length) {
// 如果指定了位置，在特定位置插入消息
messagesList.splice(position, 0, ...messagesToInsert)
} else {
// 否则默认添加到末尾
messagesList.push(...messagesToInsert)
}
},
updateMessage: (
state,
action: PayloadAction<{ topicId: string; messageId: string; updates: Partial<Message> }>
@ -233,8 +257,9 @@ export const sendMessage =
assistant: Assistant,
topic: Topic,
options?: {
resendAssistantMessage?: Message
resendAssistantMessage?: Message | Message[]
isMentionModel?: boolean
mentions?: Model[]
}
) =>
async (dispatch: AppDispatch, getState: () => RootState) => {
@ -255,17 +280,27 @@ export const sendMessage =
if (options?.resendAssistantMessage) {
// 直接使用传入的助手消息，进行重置
const messageToReset = options.resendAssistantMessage
const { model, id } = messageToReset
const resetMessage = resetAssistantMessage(messageToReset, model)
// 更新状态
dispatch(updateMessage({ topicId: topic.id, messageId: id, updates: resetMessage }))
// 使用重置后的消息
assistantMessages.push(resetMessage)
if (Array.isArray(messageToReset)) {
assistantMessages = messageToReset.map((m) => {
const { model, id } = m
const resetMessage = resetAssistantMessage(m, model)
// 更新状态
dispatch(updateMessage({ topicId: topic.id, messageId: id, updates: resetMessage }))
// 使用重置后的消息
return resetMessage
})
} else {
const { model, id } = messageToReset
const resetMessage = resetAssistantMessage(messageToReset, model)
// 更新状态
dispatch(updateMessage({ topicId: topic.id, messageId: id, updates: resetMessage }))
// 使用重置后的消息
assistantMessages.push(resetMessage)
}
} else {
// 不是重发情况
if (userMessage.mentions?.length) {
// 为每个被 mention 的模型创建一个助手消息
assistantMessages = userMessage.mentions.map((m) => {
// 为每个被 mention 的模型创建一个助手消息
if (options?.mentions?.length) {
assistantMessages = options?.mentions.map((m) => {
const assistantMessage = getAssistantMessage({ assistant: { ...assistant, model: m }, topic })
assistantMessage.model = m
assistantMessage.askId = userMessage.id
@ -280,19 +315,36 @@ export const sendMessage =
assistantMessages.push(assistantMessage)
}
// 获取当前消息列表
const currentMessages = getState().messages.messagesByTopic[topic.id]
// 最后一个具有相同askId的助手消息，在其后插入
let position: number | undefined
if (options?.isMentionModel) {
const lastAssistantIndex = currentMessages.findLastIndex(
(m) => m.role === 'assistant' && m.askId === userMessage.id
)
if (lastAssistantIndex !== -1) {
position = lastAssistantIndex + 1
}
}
dispatch(
addMessage({
appendMessage({
topicId: topic.id,
messages: !options?.isMentionModel ? [userMessage, ...assistantMessages] : assistantMessages
messages: !options?.isMentionModel ? [userMessage, ...assistantMessages] : assistantMessages,
position
})
)
}
for (const assistantMessage of assistantMessages) {
// for of会收到await 影响,在暂停的时候会因为异步的原因有概率拿不到数据
dispatch(setStreamMessage({ topicId: topic.id, message: assistantMessage }))
}
const queue = getTopicQueue(topic.id)
for (const assistantMessage of assistantMessages) {
// Set as stream message instead of adding to messages
dispatch(setStreamMessage({ topicId: topic.id, message: assistantMessage }))
// Sync user message with database
const state = getState()
@ -303,7 +355,7 @@ export const sendMessage =
}
// 保证请求有序，防止请求静态，限制并发数量
await queue.add(async () => {
queue.add(async () => {
try {
const messages = getState().messages.messagesByTopic[topic.id]
if (!messages) {
@ -324,13 +376,30 @@ export const sendMessage =
// 节流
const throttledDispatch = throttle(handleResponseMessageUpdate, 100, { trailing: true }) // 100ms的节流时间应足够平衡用户体验和性能
// 寻找当前正在处理的消息在消息列表中的位置
// const messageIndex = messages.findIndex((m) => m.id === assistantMessage.id)
const handleMessages = (): Message[] => {
// 找到对应的用户消息位置
const userMessageIndex = messages.findIndex((m) => m.id === assistantMessage.askId)
const messageIndex = messages.findIndex((m) => m.id === assistantMessage.id)
if (userMessageIndex !== -1) {
// 先截取到用户消息为止的所有消息，再进行过滤
const messagesUpToUser = messages.slice(0, userMessageIndex + 1)
return messagesUpToUser.filter((m) => !m.status?.includes('ing'))
}
// 如果找不到对应的用户消息，使用原有逻辑
// 按理说不会找不到 先注释掉看看
// if (messageIndex !== -1) {
// const messagesUpToAssistant = messages.slice(0, messageIndex)
// return messagesUpToAssistant.filter((m) => !m.status?.includes('ing'))
// }
// 没有找到消息索引的情况，过滤所有消息
return messages.filter((m) => !m.status?.includes('ing'))
}
await fetchChatCompletion({
message: { ...assistantMessage },
messages: messages
.filter((m) => !m.status?.includes('ing'))
.slice(0, messageIndex !== -1 ? messageIndex : undefined),
messages: handleMessages(),
assistant: assistantWithModel,
onResponse: async (msg) => {
// 允许在回调外维护一个最新的消息状态，每次都更新这个对象，但只通过节流函数分发到Redux
@ -362,13 +431,14 @@ export const sendMessage =
}
})
}
// 等待所有请求完成,设置loading
await queue.onIdle()
dispatch(setTopicLoading({ topicId: topic.id, loading: false }))
} catch (error: any) {
console.error('Error in sendMessage:', error)
dispatch(setError(error.message))
dispatch(setTopicLoading({ topicId: topic.id, loading: false }))
} finally {
// 等待所有请求完成,设置loading
await waitForTopicQueue(topic.id)
dispatch(setTopicLoading({ topicId: topic.id, loading: false }))
}
}
@ -385,7 +455,7 @@ export const resendMessage =
// 如果是用户消息，直接重发
if (message.role === 'user') {
// 查找此用户消息对应的助手消息
const assistantMessage = topicMessages.find((m) => m.role === 'assistant' && m.askId === message.id)
const assistantMessage = topicMessages.filter((m) => m.role === 'assistant' && m.askId === message.id)
return dispatch(
sendMessage(message, assistant, topic, {
resendAssistantMessage: assistantMessage,
@ -408,6 +478,7 @@ export const resendMessage =
return dispatch(sendMessage(userMessage, assistant, topic, { isMentionModel }))
}
console.log('assistantMessage', message)
dispatch(
sendMessage(userMessage, assistant, topic, {
resendAssistantMessage: message
@ -521,7 +592,8 @@ export const {
loadTopicMessages,
setStreamMessage,
commitStreamMessage,
clearStreamMessage
clearStreamMessage,
appendMessage
} = messagesSlice.actions
export default messagesSlice.reducer

29
src/renderer/src/utils/abortController.ts
View File

@ -1,25 +1,22 @@
export const abortMap = new Map<string, () => void>()
export const abortMap = new Map<string, (() => void)[]>()
export const addAbortController = (id: string, abortFn: () => void) => {
let callback = abortFn
const existingCallback = abortMap.get(id)
if (existingCallback) {
callback = () => {
existingCallback?.()
abortFn()
}
}
abortMap.set(id, callback)
abortMap.set(id, [...(abortMap.get(id) || []), abortFn])
}
export const removeAbortController = (id: string) => {
abortMap.delete(id)
export const removeAbortController = (id: string, abortFn: () => void) => {
const callbackArr = abortMap.get(id)
if (abortFn) {
callbackArr?.splice(callbackArr?.indexOf(abortFn), 1)
} else abortMap.delete(id)
}
export const abortCompletion = (id: string) => {
const abortFn = abortMap.get(id)
if (abortFn) {
abortFn()
removeAbortController(id)
const abortFns = abortMap.get(id)
if (abortFns?.length) {
for (const fn of [...abortFns]) {
fn()
removeAbortController(id, fn)
}
}
}

8
src/renderer/src/utils/queue.ts
View File

@ -8,13 +8,9 @@ const requestQueues: { [topicId: string]: PQueue } = {}
* @param topicId The ID of the topic
* @returns A PQueue instance for the topic
*/
export const getTopicQueue = (topicId: string): PQueue => {
export const getTopicQueue = (topicId: string, options = {}): PQueue => {
if (!requestQueues[topicId]) {
requestQueues[topicId] = new PQueue({
concurrency: 4,
timeout: 1000 * 60 * 5, // 5 minutes
throwOnTimeout: false
})
requestQueues[topicId] = new PQueue(options)
}
return requestQueues[topicId]
}