Retry net::ERR_NETWORK_CHANGED with Node (#1228)

Author: chrmarti

src/extension/prompt/node/chatMLFetcher.ts

@@ -136,9 +136,10 @@ export class ChatMLFetcherImpl extends AbstractChatMLFetcher {
 			ignoreStatefulMarker: opts.ignoreStatefulMarker
 		});
 		let tokenCount = -1;
+		const streamRecorder = new FetchStreamRecorder(finishedCb);
+		const enableRetryOnError = opts.enableRetryOnError ?? opts.enableRetryOnFilter;
 		try {
 			let response: ChatResults | ChatRequestFailed | ChatRequestCanceled;
-			const streamRecorder = new FetchStreamRecorder(finishedCb);
 			const payloadValidationResult = isValidChatPayload(opts.messages, postOptions);
 			if (!payloadValidationResult.isValid) {
 				response = {
@@ -160,7 +161,8 @@ export class ChatMLFetcherImpl extends AbstractChatMLFetcher {
 					postOptions.n,
 					userInitiatedRequest,
 					token,
-					telemetryProperties
+					telemetryProperties,
+					opts.useFetcher,
 				));
 				tokenCount = await chatEndpoint.acquireTokenizer().countMessagesTokens(messages);
 				const extensionId = source?.extensionId ?? EXTENSION_ID;
@@ -208,6 +210,7 @@ export class ChatMLFetcherImpl extends AbstractChatMLFetcher {
 									userInitiatedRequest: false, // do not mark the retry as user initiated
 									telemetryProperties: { ...telemetryProperties, retryAfterFilterCategory: result.category ?? 'uncategorized' },
 									enableRetryOnFilter: false,
+									enableRetryOnError,
 								}, token);
 
 								pendingLoggedChatRequest?.resolve(retryResult, streamRecorder.deltas);
@@ -235,6 +238,7 @@ export class ChatMLFetcherImpl extends AbstractChatMLFetcher {
 						requestId: ourRequestId,
 						model: chatEndpoint.model,
 						apiType: chatEndpoint.apiType,
+						...(telemetryProperties.retryAfterErrorCategory ? { retryAfterErrorCategory: telemetryProperties.retryAfterErrorCategory } : {}),
 						...(telemetryProperties.retryAfterFilterCategory ? { retryAfterFilterCategory: telemetryProperties.retryAfterFilterCategory } : {}),
 					}, {
 						totalTokenMax: chatEndpoint.modelMaxPromptTokens ?? -1,
@@ -258,6 +262,32 @@ export class ChatMLFetcherImpl extends AbstractChatMLFetcher {
 		} catch (err: unknown) {
 			const timeToError = Date.now() - baseTelemetry.issuedTime;
 			const processed = this.processError(err, ourRequestId);
+			if (['darwin', 'linux'].includes(process.platform) && processed.type === ChatFetchResponseType.NetworkError && processed.reason.indexOf('net::ERR_NETWORK_CHANGED') !== -1) {
+
+				if (enableRetryOnError) {
+					this._logService.info('Retrying chat request with node-fetch after net::ERR_NETWORK_CHANGED error.');
+					streamRecorder.callback('', 0, { text: '', retryReason: 'network_error' });
+
+					// Retry with other fetchers
+					const retryResult = await this.fetchMany({
+						debugName: 'retry-error-' + debugName,
+						messages,
+						finishedCb,
+						location,
+						endpoint: chatEndpoint,
+						source,
+						requestOptions,
+						userInitiatedRequest: false, // do not mark the retry as user initiated
+						telemetryProperties: { ...telemetryProperties, retryAfterErrorCategory: 'electron-network-changed' },
+						enableRetryOnFilter: opts.enableRetryOnFilter,
+						enableRetryOnError: false,
+						useFetcher: 'node-fetch',
+					}, token);
+
+					pendingLoggedChatRequest?.resolve(retryResult, streamRecorder.deltas);
+					return retryResult;
+				}
+			}
 			if (processed.type === ChatFetchResponseType.Canceled) {
 				this._sendCancellationTelemetry({
 					source: telemetryProperties.messageSource ?? 'unknown',
@@ -329,6 +359,7 @@ export class ChatMLFetcherImpl extends AbstractChatMLFetcher {
 				"timeToCancelled": { "classification": "SystemMetaData", "purpose": "PerformanceAndHealth", "comment": "Time to first token", "isMeasurement": true },
 				"isVisionRequest": { "classification": "SystemMetaData", "purpose": "PerformanceAndHealth", "comment": "Whether the request was for a vision model", "isMeasurement": true },
 				"isBYOK": { "classification": "SystemMetaData", "purpose": "FeatureInsight", "comment": "Whether the request was for a BYOK model", "isMeasurement": true },
+				"retryAfterErrorCategory": { "classification": "SystemMetaData", "purpose": "FeatureInsight", "comment": "If the response failed and this is a retry attempt, this contains the error category." },
 				"retryAfterFilterCategory": { "classification": "SystemMetaData", "purpose": "FeatureInsight", "comment": "If the response was filtered and this is a retry attempt, this contains the original filtered content category." }
 			}
 		*/
@@ -379,6 +410,7 @@ export class ChatMLFetcherImpl extends AbstractChatMLFetcher {
 				"timeToFirstTokenEmitted": { "classification": "SystemMetaData", "purpose": "PerformanceAndHealth", "comment": "Time to first token emitted (visible text)", "isMeasurement": true },
 				"isVisionRequest": { "classification": "SystemMetaData", "purpose": "PerformanceAndHealth", "comment": "Whether the request was for a vision model", "isMeasurement": true },
 				"isBYOK": { "classification": "SystemMetaData", "purpose": "FeatureInsight", "comment": "Whether the request was for a BYOK model", "isMeasurement": true },
+				"retryAfterErrorCategory": { "classification": "SystemMetaData", "purpose": "FeatureInsight", "comment": "If the response failed and this is a retry attempt, this contains the error category." },
 				"retryAfterFilterCategory": { "classification": "SystemMetaData", "purpose": "FeatureInsight", "comment": "If the response was filtered and this is a retry attempt, this contains the original filtered content category." }
 			}
 		*/
@@ -391,6 +423,7 @@ export class ChatMLFetcherImpl extends AbstractChatMLFetcher {
 			apiType: chatEndpointInfo.apiType,
 			reasoningEffort: requestBody.reasoning?.effort,
 			reasoningSummary: requestBody.reasoning?.summary,
+			...(telemetryProperties?.retryAfterErrorCategory ? { retryAfterErrorCategory: telemetryProperties.retryAfterErrorCategory } : {}),
 			...(telemetryProperties?.retryAfterFilterCategory ? { retryAfterFilterCategory: telemetryProperties.retryAfterFilterCategory } : {})
 		}, {
 			totalTokenMax: chatEndpointInfo.modelMaxPromptTokens ?? -1,
@@ -447,6 +480,7 @@ export class ChatMLFetcherImpl extends AbstractChatMLFetcher {
 					"timeToComplete": { "classification": "SystemMetaData", "purpose": "PerformanceAndHealth", "comment": "Time to complete the request", "isMeasurement": true },
 					"isVisionRequest": { "classification": "SystemMetaData", "purpose": "PerformanceAndHealth", "comment": "Whether the request was for a vision model", "isMeasurement": true },
 					"isBYOK": { "classification": "SystemMetaData", "purpose": "FeatureInsight", "comment": "Whether the request was for a BYOK model", "isMeasurement": true },
+					"retryAfterErrorCategory": { "classification": "SystemMetaData", "purpose": "FeatureInsight", "comment": "If the response failed and this is a retry attempt, this contains the error category." },
 					"retryAfterFilterCategory": { "classification": "SystemMetaData", "purpose": "FeatureInsight", "comment": "If the response was filtered and this is a retry attempt, this contains the original filtered content category." }
 				}
 			*/
@@ -460,6 +494,7 @@ export class ChatMLFetcherImpl extends AbstractChatMLFetcher {
 				requestId,
 				reasoningEffort: requestBody.reasoning?.effort,
 				reasoningSummary: requestBody.reasoning?.summary,
+				...(baseTelemetry?.properties.retryAfterErrorCategory ? { retryAfterErrorCategory: baseTelemetry.properties.retryAfterErrorCategory } : {}),
 				...(baseTelemetry?.properties.retryAfterFilterCategory ? { retryAfterFilterCategory: baseTelemetry.properties.retryAfterFilterCategory } : {}),
 			}, {
 				totalTokenMax: chatEndpointInfo?.modelMaxPromptTokens ?? -1,

src/extension/prompt/node/pseudoStartStopConversationCallback.ts

@@ -155,7 +155,9 @@ export class PseudoStopStartResponseProcessor implements IResponseProcessor {
 			this.stagedDeltasToApply = [];
 			this.currentStartStop = undefined;
 			this.nonReportedDeltas = [];
-			if (delta.retryReason === FilterReason.Copyright) {
+			if (delta.retryReason === 'network_error') {
+				progress.clearToPreviousToolInvocation(ChatResponseClearToPreviousToolInvocationReason.NoReason);
+			} else if (delta.retryReason === FilterReason.Copyright) {
 				progress.clearToPreviousToolInvocation(ChatResponseClearToPreviousToolInvocationReason.CopyrightContentRetry);
 			} else {
 				progress.clearToPreviousToolInvocation(ChatResponseClearToPreviousToolInvocationReason.FilteredContentRetry);

src/platform/networking/common/fetch.ts

@@ -138,7 +138,7 @@ export interface IResponseDelta {
 	_deprecatedCopilotFunctionCalls?: ICopilotFunctionCall[];
 	copilotConfirmation?: ICopilotConfirmation;
 	thinking?: ThinkingDelta | EncryptedThinkingDelta;
-	retryReason?: FilterReason;
+	retryReason?: FilterReason | 'network_error';
 	/** Marker for the current response, which should be presented in `IMakeChatRequestOptions` on the next call */
 	statefulMarker?: string;
 }

src/platform/networking/common/fetcherService.ts

@@ -45,6 +45,8 @@ export class Response {
 	}
 }
 
+export type FetcherId = 'electron-fetch' | 'node-fetch' | 'node-http';
+
 /** These are the options we currently use, for ease of reference. */
 export interface FetchOptions {
 	headers?: { [name: string]: string };
@@ -55,6 +57,7 @@ export interface FetchOptions {
 	signal?: IAbortSignal;
 	retryFallbacks?: boolean;
 	expectJSON?: boolean;
+	useFetcher?: FetcherId;
 }
 
 export interface IAbortSignal {

src/platform/networking/common/networking.ts

@@ -18,7 +18,7 @@ import { ILogService } from '../../log/common/logService';
 import { ITelemetryService, TelemetryProperties } from '../../telemetry/common/telemetry';
 import { TelemetryData } from '../../telemetry/common/telemetryData';
 import { FinishedCallback, OpenAiFunctionTool, OpenAiResponsesFunctionTool, OptionalChatRequestParams } from './fetch';
-import { FetchOptions, IAbortController, IFetcherService, Response } from './fetcherService';
+import { FetcherId, FetchOptions, IAbortController, IFetcherService, Response } from './fetcherService';
 import { ChatCompletion, RawMessageConversionCallback, rawMessageToCAPI } from './openai';
 
 /**
@@ -145,6 +145,10 @@ export interface IMakeChatRequestOptions {
 	telemetryProperties?: TelemetryProperties;
 	/** Enable retrying the request when it was filtered due to snippy. Note- if using finishedCb, requires supporting delta.retryReason, eg with clearToPreviousToolInvocation */
 	enableRetryOnFilter?: boolean;
+	/** Enable retrying the request when it failed. Defaults to enableRetryOnFilter. Note- if using finishedCb, requires supporting delta.retryReason, eg with clearToPreviousToolInvocation */
+	enableRetryOnError?: boolean;
+	/** Which fetcher to use, overrides the default. */
+	useFetcher?: FetcherId;
 }
 
 export interface ICreateEndpointBodyOptions extends IMakeChatRequestOptions {
@@ -263,7 +267,8 @@ function networkRequest(
 	requestId: string,
 	body?: IEndpointBody,
 	additionalHeaders?: Record<string, string>,
-	cancelToken?: CancellationToken
+	cancelToken?: CancellationToken,
+	useFetcher?: FetcherId,
 ): Promise<Response> {
 	// TODO @lramos15 Eventually don't even construct this fake endpoint object.
 	const endpoint = typeof endpointOrUrl === 'string' || 'type' in endpointOrUrl ? {
@@ -296,6 +301,7 @@ function networkRequest(
 		headers: headers,
 		json: body,
 		timeout: requestTimeoutMs,
+		useFetcher,
 	};
 
 	if (cancelToken) {
@@ -353,7 +359,8 @@ export function postRequest(
 	requestId: string,
 	body?: IEndpointBody,
 	additionalHeaders?: Record<string, string>,
-	cancelToken?: CancellationToken
+	cancelToken?: CancellationToken,
+	useFetcher?: FetcherId,
 ): Promise<Response> {
 	return networkRequest(fetcherService,
 		telemetryService,
@@ -365,7 +372,8 @@ export function postRequest(
 		requestId,
 		body,
 		additionalHeaders,
-		cancelToken
+		cancelToken,
+		useFetcher,
 	);
 }
 

src/platform/networking/node/fetcherFallback.ts

@@ -38,7 +38,15 @@ export async function fetchWithFallbacks(availableFetchers: readonly IFetcher[],
 		}
 		throw firstResult!.err;
 	}
-	return { response: await availableFetchers[0].fetch(url, options) };
+	const fetcher = options.useFetcher && availableFetchers.find(f => f.getUserAgentLibrary() === options.useFetcher) || availableFetchers[0];
+	if (options.useFetcher) {
+		if (options.useFetcher === fetcher.getUserAgentLibrary()) {
+			logService.debug(`FetcherService: using ${options.useFetcher} as requested.`);
+		} else {
+			logService.info(`FetcherService: could not find requested fetcher ${options.useFetcher}, using ${fetcher.getUserAgentLibrary()} instead.`);
+		}
+	}
+	return { response: await fetcher.fetch(url, options) };
 }
 
 async function tryFetch(fetcher: IFetcher, url: string, options: FetchOptions, logService: ILogService): Promise<{ ok: boolean; response: Response } | { ok: false; err: any }> {

src/platform/openai/node/fetch.ts

@@ -17,7 +17,7 @@ import { IDomainService } from '../../endpoint/common/domainService';
 import { IEnvService } from '../../env/common/envService';
 import { ILogService } from '../../log/common/logService';
 import { FinishedCallback, OptionalChatRequestParams, RequestId, getProcessingTime, getRequestId } from '../../networking/common/fetch';
-import { IFetcherService, Response } from '../../networking/common/fetcherService';
+import { FetcherId, IFetcherService, Response } from '../../networking/common/fetcherService';
 import { IChatEndpoint, IEndpointBody, postRequest, stringifyUrlOrRequestMetadata } from '../../networking/common/networking';
 import { CAPIChatMessage, ChatCompletion } from '../../networking/common/openai';
 import { sendEngineMessagesTelemetry } from '../../networking/node/chatStream';
@@ -108,7 +108,8 @@ export async function fetchAndStreamChat(
 	nChoices: number | undefined,
 	userInitiatedRequest?: boolean,
 	cancel?: CancellationToken | undefined,
-	telemetryProperties?: TelemetryProperties | undefined
+	telemetryProperties?: TelemetryProperties | undefined,
+	useFetcher?: FetcherId,
 ): Promise<ChatResults | ChatRequestFailed | ChatRequestCanceled> {
 	const logService = accessor.get(ILogService);
 	const telemetryService = accessor.get(ITelemetryService);
@@ -160,7 +161,9 @@ export async function fetchAndStreamChat(
 		location,
 		userInitiatedRequest,
 		cancel,
-		{ ...telemetryProperties, modelCallId });
+		{ ...telemetryProperties, modelCallId },
+		useFetcher,
+	);
 
 	if (cancel?.isCancellationRequested) {
 		const body = await response!.body();
@@ -465,7 +468,8 @@ async function fetchWithInstrumentation(
 	location: ChatLocation,
 	userInitiatedRequest?: boolean,
 	cancel?: CancellationToken,
-	telemetryProperties?: TelemetryProperties
+	telemetryProperties?: TelemetryProperties,
+	useFetcher?: FetcherId,
 ): Promise<Response> {
 
 	// If request contains an image, we include this header.
@@ -514,7 +518,8 @@ async function fetchWithInstrumentation(
 		ourRequestId,
 		request,
 		additionalHeaders,
-		cancel
+		cancel,
+		useFetcher,
 	).then(response => {
 		const apim = response.headers.get('apim-request-id');
 		if (apim) {