address comments left by @cryptobrains on #2472 (Feat: automatic retry for Gemini 2.5 Pro 503 Server Busy error (#2471));

.changeset/proud-pumas-brush.md

@@ -0,0 +1,5 @@
+---
+"claude-dev": patch
+---
+
+Automatically but respectfully retries Gemini Pro 2.5 API calls

package-lock.json

@@ -1,12 +1,12 @@
 {
 	"name": "claude-dev",
-	"version": "3.8.0",
+	"version": "3.8.3",
 	"lockfileVersion": 3,
 	"requires": true,
 	"packages": {
 		"": {
 			"name": "claude-dev",
-			"version": "3.8.0",
+			"version": "3.8.3",
 			"license": "Apache-2.0",
 			"dependencies": {
 				"@anthropic-ai/bedrock-sdk": "^0.12.4",

src/core/Cline.ts

@@ -72,6 +72,7 @@ import {
 	checkIsOpenRouterContextWindowError,
 } from "./context-management/context-error-handling"
 import { AnthropicHandler } from "../api/providers/anthropic"
+// Removed incorrect import for getRandomInt
 
 const cwd = vscode.workspace.workspaceFolders?.map((folder) => folder.uri.fsPath).at(0) ?? path.join(os.homedir(), "Desktop") // may or may not exist but fs checking existence would immediately ask for permission which would be bad UX, need to come up with a better solution
 
@@ -3749,4 +3750,47 @@ export class Cline {
 
 		return `<environment_details>\n${details.trim()}\n</environment_details>`
 	}
+
+	async retryFailedRequest(failedMessageTs: number) {
+		console.log(`Attempting to retry request associated with message ts: ${failedMessageTs}`)
+
+		// 1. Find the index of the failed 'api_req_started' or 'api_req_failed' message in clineMessages
+		const failedClineMessageIndex = this.clineMessages.findIndex((msg) => msg.ts === failedMessageTs)
+		if (failedClineMessageIndex === -1) {
+			console.error(`Could not find cline message with ts ${failedMessageTs} to retry.`)
+			return
+		}
+		const failedClineMessage = this.clineMessages[failedClineMessageIndex]
+
+		// 2. Find the index of the corresponding 'user' message in apiConversationHistory
+		// The 'api_req_started' message's conversationHistoryIndex points to the 'user' message that triggered it.
+		const userMessageApiIndex = failedClineMessage.conversationHistoryIndex
+		if (
+			userMessageApiIndex === undefined ||
+			userMessageApiIndex < 0 ||
+			userMessageApiIndex >= this.apiConversationHistory.length
+		) {
+			console.error(`Invalid conversationHistoryIndex (${userMessageApiIndex}) found for message ts ${failedMessageTs}.`)
+			return
+		}
+
+		const userMessage = this.apiConversationHistory[userMessageApiIndex]
+		if (userMessage.role !== "user") {
+			console.error(`Expected user message at index ${userMessageApiIndex}, but found ${userMessage.role}.`)
+			return
+		}
+
+		// 3. Clean up clineMessages: Remove the failed message and any subsequent messages
+		const messagesToKeep = this.clineMessages.slice(0, failedClineMessageIndex)
+		await this.overwriteClineMessages(messagesToKeep)
+
+		// 4. Clean up apiConversationHistory: Remove the user message that caused the failure and any subsequent assistant message
+		const historyToKeep = this.apiConversationHistory.slice(0, userMessageApiIndex)
+		await this.overwriteApiConversationHistory(historyToKeep)
+
+		// 5. Re-initiate the task loop with the user content that caused the failure
+		console.log(`Retrying with user content from index ${userMessageApiIndex}`)
+		await this.say("api_req_retried") // Inform UI about the retry
+		await this.initiateTaskLoop(userMessage.content as UserContent, false) // isNewTask is false
+	}
 }

src/core/webview/ClineProvider.ts

@@ -938,6 +938,15 @@ export class ClineProvider implements vscode.WebviewViewProvider {
 						this.postMessageToWebview({ type: "relinquishControl" })
 						break
 					}
+					case "retryApiRequest": {
+						if (message.ts && this.cline) {
+							console.log(`Received retry request for message ts: ${message.ts}`)
+							// Call a method on the Cline instance to handle the retry
+							// We'll need to implement this method in Cline.ts
+							await this.cline.retryFailedRequest(message.ts)
+						}
+						break
+					}
 					// Add more switch case statements here as more webview message commands
 					// are created within the webview context (i.e. inside media/main.js)
 				}

src/shared/ExtensionMessage.ts

@@ -40,6 +40,7 @@ export interface ExtensionMessage {
 		| "userCreditsPayments"
 		| "totalTasksSize"
 		| "addToInput"
+		| "retryApiRequest" // Added for automatic retry
 	text?: string
 	action?:
 		| "chatButtonClicked"

src/shared/WebviewMessage.ts

@@ -65,8 +65,10 @@ export interface WebviewMessage {
 		| "fetchUserCreditsData"
 		| "optionsResponse"
 		| "requestTotalTasksSize"
+		| "retryApiRequest" // Added for automatic retry
 	// | "relaunchChromeDebugMode"
 	text?: string
+	ts?: number // Added for retryApiRequest
 	disabled?: boolean
 	askResponse?: ClineAskResponse
 	apiConfiguration?: ApiConfiguration

webview-ui/src/components/chat/ChatRow.tsx

@@ -3,6 +3,12 @@ import deepEqual from "fast-deep-equal"
 import React, { memo, useCallback, useEffect, useMemo, useRef, useState } from "react"
 import { useEvent, useSize } from "react-use"
 import styled from "styled-components"
+// Define getRandomInt locally as import path was incorrect
+const getRandomInt = (min: number, max: number): number => {
+	min = Math.ceil(min)
+	max = Math.floor(max)
+	return Math.floor(Math.random() * (max - min + 1)) + min
+}
 import {
 	ClineApiReqInfo,
 	ClineAskQuestion,
@@ -136,6 +142,7 @@ export default ChatRow
 export const ChatRowContent = ({ message, isExpanded, onToggleExpand, lastModifiedMessage, isLast }: ChatRowContentProps) => {
 	const { mcpServers, mcpMarketplaceCatalog } = useExtensionState()
 	const [seeNewChangesDisabled, setSeeNewChangesDisabled] = useState(false)
+	const [retryAttempted, setRetryAttempted] = useState(false) // State to prevent infinite retries
 
 	const [cost, apiReqCancelReason, apiReqStreamingFailedMessage] = useMemo(() => {
 		if (message.text != null && message.say === "api_req_started") {
@@ -151,6 +158,40 @@ export const ChatRowContent = ({ message, isExpanded, onToggleExpand, lastModifi
 			? lastModifiedMessage?.text
 			: undefined
 
+	// Effect for automatic retry on specific errors
+	useEffect(() => {
+		const errorText = apiRequestFailedMessage || apiReqStreamingFailedMessage
+		const shouldRetry =
+			isLast && // Only retry if it's the last message
+			!retryAttempted && // Only retry once per message instance
+			errorText && // Ensure there is an error message
+			(errorText.includes("503 Service Unavailable") || errorText.includes("model is overloaded"))
+
+		let retryTimeoutId;
+
+		if (shouldRetry) {
+			console.log("Detected overload error, scheduling retry for message:", message.ts)
+			setRetryAttempted(true) // Mark retry as attempted for this instance
+			const delay = getRandomInt(3000, 7000) // Random delay between 3 and 7 seconds
+			retryTimeoutId = setTimeout(() => {
+				console.log(`Retrying API request for message ${message.ts} after ${delay}ms delay.`)
+				vscode.postMessage({
+					type: "retryApiRequest",
+					ts: message.ts,
+				})
+			}, delay)
+		}
+		// Reset retry flag if the message or error changes (e.g., user manually retries or new message arrives)
+		// This might need refinement depending on exact interaction flow.
+		// For now, reset if the message timestamp changes.
+		return () => {
+			// Cleanup if needed, potentially reset retryAttempted if component unmounts or message changes significantly
+			if (retryTimeoutId) {
+				clearTimeout(retryTimeoutId);
+			}
+		}
+	}, [apiRequestFailedMessage, apiReqStreamingFailedMessage, isLast, message.ts, retryAttempted])
+
 	const isCommandExecuting =
 		isLast &&
 		(lastModifiedMessage?.ask === "command" || lastModifiedMessage?.say === "command") &&