diff --git a/app/(chat)/api/chat/_lib/image-failover.ts b/app/(chat)/api/chat/_lib/image-failover.ts index c8c661634..70f77063c 100644 --- a/app/(chat)/api/chat/_lib/image-failover.ts +++ b/app/(chat)/api/chat/_lib/image-failover.ts @@ -264,6 +264,29 @@ export function chatMessagesHaveImageInput(messages: ChatMessage[]) { return countChatMessageImageInputs(messages) > 0; } +export function canUseImageFallback( + candidate: { model: any; provider: any }, + imageParams: { imageSize?: unknown; aspectRatio?: unknown; imageCount?: number }, + messages: ChatMessage[], + balanceBlockedProviders: ReadonlySet, +) { + if (balanceBlockedProviders.has(candidate.provider?.id)) return false; + const config = candidate.model?.imageGenConfig; + const size = String(imageParams.imageSize || "").toUpperCase(); + const ratio = String(imageParams.aspectRatio || ""); + if (size && Array.isArray(config?.supportedImageSizes) && + !config.supportedImageSizes.includes(size)) return false; + if (ratio && ratio !== "auto" && Array.isArray(config?.supportedAspectRatios) && + !config.supportedAspectRatios.includes(ratio)) return false; + if (imageParams.imageCount && typeof config?.maxImages === "number" && + imageParams.imageCount > config.maxImages) return false; + const referenceCount = countChatMessageImageInputs(messages); + if (referenceCount && modelLacksReferenceImageInputSupport(candidate.model, candidate.provider)) return false; + if (referenceCount && typeof config?.maxReferenceImages === "number" && + referenceCount > config.maxReferenceImages) return false; + return true; +} + export function getImageFailoverConvertProvider(providerType?: string | null) { switch (providerType) { case "bfl-image": diff --git a/app/(chat)/api/chat/_lib/image-generation-failure.ts b/app/(chat)/api/chat/_lib/image-generation-failure.ts index 3344b7852..a4a5bdf9c 100644 --- a/app/(chat)/api/chat/_lib/image-generation-failure.ts +++ b/app/(chat)/api/chat/_lib/image-generation-failure.ts @@ -1,8 +1,11 @@ import { saveMessages } from "@/lib/db/queries"; import type { ImageGenerationStatusStreamPayload } from "@/lib/canvas/image-generation-status"; import { formatRequestFailure, getRequestFailureDetails } from "@/lib/request-error"; +import { createLogger } from "@/lib/observability/logger"; import { safeWriteStreamChunk, type StreamWriter, writeAssistantTextChunk, writeStructuredDataChunk } from "./stream"; +const logger = createLogger("chat.imageGenerationFailure"); + export async function reportImageGenerationFailure(params: { writer: StreamWriter; chatId: string; @@ -12,6 +15,12 @@ export async function reportImageGenerationFailure(params: { label: string; }) { const failure = getRequestFailureDetails(params.error, "图片生成失败"); + logger.warn("Image request failed", { + requestId: params.messageId, + userMessageId: params.userMessageId, + providerRequestId: (params.error as { providerRequestId?: string } | null)?.providerRequestId, + code: failure.code, + }); const status: ImageGenerationStatusStreamPayload = { phase: "failed", userMessageId: params.userMessageId, diff --git a/app/(chat)/api/chat/route.ts b/app/(chat)/api/chat/route.ts index c34f819d8..087f5957b 100644 --- a/app/(chat)/api/chat/route.ts +++ b/app/(chat)/api/chat/route.ts @@ -337,6 +337,7 @@ import { } from "./_lib/image-media"; import { buildImageFallbackCandidates, + canUseImageFallback, countChatMessageImageInputs, generateImageWithFailoverCandidate, getImageCrossProviderCandidates, @@ -9216,8 +9217,11 @@ export async function POST(request: Request) { // 🆕 尝试备用模型(支持跨服务商切换) let fallbackSuccess = false; + const balanceBlockedProviders = new Set(); + if (isProviderBalanceError(primaryError) && currentModelConfig?.providerId) balanceBlockedProviders.add(currentModelConfig.providerId); const triedFallbackModelIds = new Set(); for (const candidate of fallbackCandidates) { + if (!canUseImageFallback(candidate, effectiveImageGenerationParams, buildMessagesForGemini(), balanceBlockedProviders)) continue; try { triedFallbackModelIds.add(String(candidate.model.id || "")); chatApiLogger.info(`[Chat API] 🔄 Trying fallback model: ${candidate.model.displayName} (${candidate.provider.type})`); @@ -9239,6 +9243,7 @@ export async function POST(request: Request) { chatApiLogger.info(`[Chat API] ✅ Fallback model ${fallbackResult.displayName} succeeded!`); break; } catch (fallbackError) { + if (isProviderBalanceError(fallbackError)) balanceBlockedProviders.add(candidate.provider.id); const fallbackErrorMessage = normalizeFailoverErrorMessage(fallbackError); if (shouldCooldownImageFallbackModel(fallbackErrorMessage)) { setImageFallbackModelCooldown(candidate.model.id, fallbackErrorMessage); @@ -9292,6 +9297,7 @@ export async function POST(request: Request) { }).filter((candidate) => !triedFallbackModelIds.has(String(candidate.model.id || ""))); for (const candidate of crossProviderCandidates) { + if (!canUseImageFallback(candidate, effectiveImageGenerationParams, buildMessagesForGemini(), balanceBlockedProviders)) continue; try { triedFallbackModelIds.add(String(candidate.model.id || "")); chatApiLogger.info(`[Chat API] 🔁 Gemini cross-provider failover: ${candidate.model.displayName} (${candidate.provider.type})`); @@ -9314,6 +9320,7 @@ export async function POST(request: Request) { chatApiLogger.info(`[Chat API] ✅ Gemini cross-provider fallback model ${fallbackResult.displayName} succeeded!`); break; } catch (fallbackError) { + if (isProviderBalanceError(fallbackError)) balanceBlockedProviders.add(candidate.provider.id); const fallbackErrorMessage = normalizeFailoverErrorMessage(fallbackError); if (shouldCooldownImageFallbackModel(fallbackErrorMessage)) { setImageFallbackModelCooldown(candidate.model.id, fallbackErrorMessage); @@ -9984,7 +9991,7 @@ export async function POST(request: Request) { const balanceBlockedProviders = new Set(); if (isProviderBalanceError(primaryError) && currentModelEntry?.provider.id) balanceBlockedProviders.add(currentModelEntry.provider.id); for (const candidate of fallbackCandidates) { - if (balanceBlockedProviders.has(candidate.provider.id)) continue; + if (!canUseImageFallback(candidate, effectiveImageGenerationParams, messagesForAlibaba, balanceBlockedProviders)) continue; try { triedFallbackModelIds.add(String(candidate.model.id || "")); chatApiLogger.info(`[Chat API] 🔄 Alibaba fallback model: ${candidate.model.displayName} (${candidate.provider.type})`); @@ -10060,7 +10067,7 @@ export async function POST(request: Request) { }).filter((candidate) => !triedFallbackModelIds.has(String(candidate.model.id || "")) && !balanceBlockedProviders.has(candidate.provider.id)); for (const candidate of crossProviderCandidates) { - if (balanceBlockedProviders.has(candidate.provider.id)) continue; + if (!canUseImageFallback(candidate, effectiveImageGenerationParams, messagesForAlibaba, balanceBlockedProviders)) continue; try { triedFallbackModelIds.add(String(candidate.model.id || "")); chatApiLogger.info(`[Chat API] 🔁 Cross-provider failover attempt: ${candidate.model.displayName} (${candidate.provider.type})`); @@ -10448,7 +10455,10 @@ export async function POST(request: Request) { throw primaryError; } let fallbackSuccess = false; + const balanceBlockedProviders = new Set(); + if (isProviderBalanceError(primaryError) && volcanoModelConfig?.providerId) balanceBlockedProviders.add(volcanoModelConfig.providerId); for (const candidate of fallbackCandidates) { + if (!canUseImageFallback(candidate, effectiveImageGenerationParams, messagesForVolcano, balanceBlockedProviders)) continue; try { chatApiLogger.info(`[Chat API] 🔄 Volcano fallback model: ${candidate.model.displayName} (${candidate.provider.type})`); const fallbackResult = await generateImageWithFailoverCandidate({ @@ -10470,6 +10480,7 @@ export async function POST(request: Request) { chatApiLogger.info(`[Chat API] ✅ Volcano fallback model ${fallbackResult.displayName} succeeded!`); break; } catch (fallbackError) { + if (isProviderBalanceError(fallbackError)) balanceBlockedProviders.add(candidate.provider.id); const fallbackErrorMessage = normalizeFailoverErrorMessage(fallbackError); if (shouldCooldownImageFallbackModel(fallbackErrorMessage)) { setImageFallbackModelCooldown(candidate.model.id, fallbackErrorMessage); @@ -10505,6 +10516,7 @@ export async function POST(request: Request) { }); for (const candidate of crossProviderCandidates) { + if (!canUseImageFallback(candidate, effectiveImageGenerationParams, messagesForVolcano, balanceBlockedProviders)) continue; try { chatApiLogger.info(`[Chat API] 🔁 Volcano cross-provider failover attempt: ${candidate.model.displayName} (${candidate.provider.type})`); if ( @@ -10585,6 +10597,7 @@ export async function POST(request: Request) { chatApiLogger.info(`[Chat API] ✅ Volcano cross-provider failover succeeded: ${candidate.model.displayName} (${candidate.provider.type})`); break; } catch (crossError) { + if (isProviderBalanceError(crossError)) balanceBlockedProviders.add(candidate.provider.id); const crossErrorMessage = normalizeFailoverErrorMessage(crossError); if (shouldCooldownImageFallbackModel(crossErrorMessage)) { setImageFallbackModelCooldown(candidate.model.id, crossErrorMessage); @@ -11328,7 +11341,10 @@ export async function POST(request: Request) { } let fallbackSuccess = false; + const balanceBlockedProviders = new Set(); + if (isProviderBalanceError(primaryError) && currentModelConfig?.providerId) balanceBlockedProviders.add(currentModelConfig.providerId); for (const candidate of fallbackCandidates) { + if (!canUseImageFallback(candidate, effectiveImageGenerationParams, messagesForCrossProvider, balanceBlockedProviders)) continue; try { chatApiLogger.info(`[Chat API] 🔄 OpenAI-compatible fallback model: ${candidate.model.displayName} (${candidate.provider.type})`); const shouldUseOpenAICompatFallback = @@ -11370,6 +11386,7 @@ export async function POST(request: Request) { chatApiLogger.info(`[Chat API] ✅ OpenAI-compatible fallback model ${candidate.model.displayName} succeeded!`); break; } catch (fallbackError) { + if (isProviderBalanceError(fallbackError)) balanceBlockedProviders.add(candidate.provider.id); const fallbackErrorMessage = normalizeFailoverErrorMessage(fallbackError); if (shouldCooldownImageFallbackModel(fallbackErrorMessage)) { setImageFallbackModelCooldown(candidate.model.id, fallbackErrorMessage); @@ -11406,6 +11423,7 @@ export async function POST(request: Request) { }); for (const candidate of crossProviderCandidates) { + if (!canUseImageFallback(candidate, effectiveImageGenerationParams, messagesForCrossProvider, balanceBlockedProviders)) continue; try { chatApiLogger.info(`[Chat API] 🔁 OpenAI-compatible cross-provider failover: ${candidate.model.displayName} (${candidate.provider.type})`); if ( @@ -11489,6 +11507,7 @@ export async function POST(request: Request) { chatApiLogger.info(`[Chat API] ✅ OpenAI-compatible cross-provider failover succeeded: ${candidate.model.displayName} (${candidate.provider.type})`); break; } catch (crossError) { + if (isProviderBalanceError(crossError)) balanceBlockedProviders.add(candidate.provider.id); const crossErrorMessage = normalizeFailoverErrorMessage(crossError); if (shouldCooldownImageFallbackModel(crossErrorMessage)) { setImageFallbackModelCooldown(candidate.model.id, crossErrorMessage); diff --git a/docs/模型参数核对-2026-11-16.md b/docs/模型参数核对-2026-11-16.md index e5002a8a8..d51a180d7 100644 --- a/docs/模型参数核对-2026-11-16.md +++ b/docs/模型参数核对-2026-11-16.md @@ -15,6 +15,8 @@ | 文字输出长度 | AI SDK 调用统一使用 maxOutputTokens,原生厂商接口保留各自标准字段。 | | Claude 采样参数 | 对固定采样协议移除 temperature、top_p、top_k,保留工具调用和输出长度。Haiku 5.5 已实测。 | | 世界生成分类 | world-generation 的类型及任务调用配置按现有世界生成接口校正,避免作为聊天模型使用。 | +| 聊天生图备用模型 | 按原请求的清晰度、比例、张数和参考图数量筛选,不把 1K/4K 请求传给固定 2K 模型;额度不足时跳过同一账户的其他模型。 | +| 额度错误与请求追踪 | 98K 的英文可用额度不足归类为服务商余额/额度问题,保留上游请求号并关联网页请求号。 | ## 验证 @@ -22,6 +24,7 @@ - 阿里云图片另外覆盖 80 个尺寸、数量、参考图、随机种子和开关组合;7 个 AtlasCloud 三维模型按各自 OpenAPI 请求规范核对。 - 19 个聊天/文字模型捕获官方 SDK 生成的请求,覆盖输出长度、采样、停止词以及适用的图片和工具调用。13 个 98K 模型同时通过当前路线契约校验;其余 6 个根据接入代码和 SDK 请求核对。 - 自动回归 118 项通过,TypeScript 检查通过。 +- 聊天回退与失败提示另通过 49 项针对性回归,覆盖固定清晰度、比例、张数、参考图上限、同账户额度隔离、未知结果不重提,以及网页提示和历史记录一致。 - 独立实测:Nano Banana 2 带参考图生成 16:9 图片,返回 1376×768 PNG;固定 4K 编辑返回 4096×4096 JPEG;Claude Haiku 5.5 返回有效工具调用。两张图片均完成七牛保存和下载验证。每项只有一次提交。 ## 生效模型清单 @@ -107,6 +110,7 @@ ## 已知问题与验证边界 +- 11 月 16 日 15:14–15:23(新加坡时间)98K 只读核查确认账户可用余额不足以预留生成费用,相关拒绝未派发上游;恢复该账户生成能力需要补充余额。此问题与网页是否刷新无关。 - 场景融合的首选模型配置指向一个视频模型,当前实现会排除该项并使用图片候选模型,因此实际没有将图片请求提交给视频接口;该首选配置尚未调整。 - 离线参数核对不等于 76 个模型逐个完成付费生成,也不证明上游持续可用、效果或速度;实际生成验证范围为上述三项。 - 自动选模时页面尚未确定最终模型,接入层仍须依据实际选中接口规范化固定参数。模型清单和接口约束会随上游调整,本报告对应本次核对时点。 diff --git a/lib/ai/providers/vip98k.ts b/lib/ai/providers/vip98k.ts index 3661bcba6..cdcc62eaa 100644 --- a/lib/ai/providers/vip98k.ts +++ b/lib/ai/providers/vip98k.ts @@ -1317,7 +1317,10 @@ async function submitTask(params: { || [502, 504, 524].includes(response.status) || /pending\s+safe\s+reconciliation|request\s+outcome\s+(?:is\s+)?unknown|final\s+outcome\s+could\s+not\s+be\s+confirmed/i.test(errorMessage)) throw new ProviderTaskUncertainError(`供应商提交结果未确认,可能仍在处理,请勿重复生成:${getUserFacingImageGenerationErrorMessage({ message: errorMessage, status: response.status })} (HTTP ${response.status})`); - throw new Error(`${errorMessage} (HTTP ${response.status})`); + throw Object.assign(new Error(`${errorMessage} (HTTP ${response.status})`), { + status: response.status, + providerRequestId: response.headers.get("x-request-id"), + }); } throw new Error("VIP98K request failed after media readiness retries"); diff --git a/lib/request-error.ts b/lib/request-error.ts index d9b44422c..36375284f 100644 --- a/lib/request-error.ts +++ b/lib/request-error.ts @@ -18,6 +18,7 @@ const PROVIDER_BALANCE_PATTERNS = [ /\b(?:exceeded_current_quota|pre_consume_token_quota_failed)\b/i, /\btoken quota is not enough\b/i, /\bquota is not enough\b/i, + /\bdoes not have enough available quota\b/i, /\b(?:not enough credits?|credits? is not enough)\b/i, /\b(?:billing_hard_limit_reached|payment_required)\b/i, ]; diff --git a/playwright.helpers.config.ts b/playwright.helpers.config.ts index 1628b6977..78300ddba 100644 --- a/playwright.helpers.config.ts +++ b/playwright.helpers.config.ts @@ -23,6 +23,7 @@ export default defineConfig({ /routes\/video-canvas-media-import\.test\.ts/, /routes\/video-canvas-toolbar-jobs\.test\.ts/, /routes\/image-provider-error-classification\.test\.ts/, + /routes\/chat-image-fallback-contract\.test\.ts/, /routes\/image-generation-status\.test\.ts/, /routes\/canvas-agent-runtime-image-failover\.test\.ts/, /routes\/video-canvas-audio-remote\.test\.ts/, diff --git a/tests/routes/chat-image-fallback-contract.test.ts b/tests/routes/chat-image-fallback-contract.test.ts new file mode 100644 index 000000000..54d8870a0 --- /dev/null +++ b/tests/routes/chat-image-fallback-contract.test.ts @@ -0,0 +1,45 @@ +import { expect, test } from "@playwright/test"; +import { serverLoader } from "../helpers/canvas-download-server-loader"; + +const { canUseImageFallback } = serverLoader({ + "@/lib/ai/dynamic-provider": {}, + "@/lib/ai/providers/fal-image": {}, + "./image-media": {}, +})("app/(chat)/api/chat/_lib/image-failover.ts"); + +const candidate = { + provider: { id: "account", type: "vip98k" }, + model: { + modelId: "nano-banana-2-2k-cl", supportsVision: true, + imageGenConfig: { supportedImageSizes: ["2K"], supportedAspectRatios: ["1:1", "3:4"], maxReferenceImages: 2, maxImages: 1 }, + }, +}; +const params = { imageSize: "2K", aspectRatio: "3:4", imageCount: 1 }; +const messages = [{ role: "user", parts: [ + { type: "text", text: "Replace the bookshelf" }, + { type: "file", mediaType: "image/png", url: "https://fixture.invalid/source.png" }, +] }]; + +test("a fixed 2K fallback never receives a 1K or 4K request", () => { + for (const imageSize of ["1K", "4K"]) { + expect(canUseImageFallback(candidate, { ...params, imageSize }, messages, new Set())).toBe(false); + } + expect(canUseImageFallback(candidate, params, messages, new Set())).toBe(true); +}); + +test("fallback preserves image ratio, quantity and all reference images", () => { + expect(canUseImageFallback(candidate, { ...params, aspectRatio: "16:9" }, messages, new Set())).toBe(false); + expect(canUseImageFallback(candidate, { ...params, imageCount: 2 }, messages, new Set())).toBe(false); + expect(canUseImageFallback(candidate, params, [messages[0], messages[0], messages[0]], new Set())).toBe(false); + const textOnly = { ...candidate, model: { ...candidate.model, supportsVision: false } }; + expect(canUseImageFallback(textOnly, params, messages, new Set())).toBe(false); + expect(canUseImageFallback(textOnly, params, [], new Set())).toBe(true); + expect(canUseImageFallback(candidate, { ...params, aspectRatio: "auto" }, messages, new Set())).toBe(true); +}); + +test("an exhausted account is skipped across models, while a compatible independent account is eligible", () => { + const blocked = new Set(["account"]); + expect(canUseImageFallback(candidate, params, messages, blocked)).toBe(false); + expect(canUseImageFallback({ ...candidate, model: { ...candidate.model, modelId: "another-model" } }, params, messages, blocked)).toBe(false); + expect(canUseImageFallback({ ...candidate, provider: { ...candidate.provider, id: "other-account" } }, params, messages, blocked)).toBe(true); +}); diff --git a/tests/routes/image-generation-status.test.ts b/tests/routes/image-generation-status.test.ts index f1456c7bb..e40a6b390 100644 --- a/tests/routes/image-generation-status.test.ts +++ b/tests/routes/image-generation-status.test.ts @@ -37,3 +37,26 @@ test("late failures remain associated with their original user request", () => { expect(findImageGenerationFailure(messages, "second")).toBeNull(); expect(findImageGenerationFailure([{ role: "assistant", parts: [{ type: "text", text: "请分析失败原因" }] }], "second")).toBeNull(); }); + +test("quota rejection reports the same terminal state to the browser and history and links the upstream request", async () => { + const chunks: any[] = []; + const saved: any[] = []; + const logs: any[] = []; + const load = serverLoader({ + "@/lib/db/queries": { saveMessages: async ({ messages }: any) => saved.push(...messages) }, + "@/lib/observability/logger": { createLogger: () => ({ warn: (_: string, fields: any) => logs.push(fields) }) }, + }); + const { reportImageGenerationFailure } = load("app/(chat)/api/chat/_lib/image-generation-failure.ts"); + await reportImageGenerationFailure({ + writer: { write: (chunk: any) => chunks.push(chunk) }, + chatId: "chat", userMessageId: "original-user", messageId: "local-request", + error: Object.assign(new Error("The API key does not have enough available quota. (HTTP 429)"), { + status: 429, providerRequestId: "upstream-request", + }), label: "test", + }); + const streamed = JSON.parse(chunks.find(chunk => chunk.type === "data-imageGenerationStatus").data); + expect(findImageGenerationFailure(saved, "original-user")).toEqual(streamed); + expect(streamed).toMatchObject({ phase: "failed", failure: { code: "PROVIDER_BALANCE_EXHAUSTED", outcome: "failed" } }); + expect(saved[0].parts[0].text).toContain("与您的积分余额无关"); + expect(logs).toContainEqual(expect.objectContaining({ requestId: "local-request", providerRequestId: "upstream-request" })); +}); diff --git a/tests/routes/image-provider-error-classification.test.ts b/tests/routes/image-provider-error-classification.test.ts index c7a539416..d756779b3 100644 --- a/tests/routes/image-provider-error-classification.test.ts +++ b/tests/routes/image-provider-error-classification.test.ts @@ -269,6 +269,14 @@ test("ordinary image success uses the original response directly", async () => { expect(h.calls.map(c => c.method)).toEqual(["POST"]); }); +test("a 98K quota rejection retains its request ID without another submission", async () => { + const h = imageRecoveryHarness(() => Response.json({ error: { + code: "insufficient_quota", message: "The API key does not have enough available quota.", + } }, { status: 429, headers: { "X-Request-ID": "provider-quota-request" } })); + await expect(h.run()).rejects.toMatchObject({ status: 429, providerRequestId: "provider-quota-request" }); + expect(h.calls.map(c => c.method)).toEqual(["POST"]); +}); + test("an explicit provider balance error takes precedence over HTTP 429", () => { const error = Object.assign(new Error("余额不足,请点击充值页面续费(HTTP 429)"), { status: 429 }); expect(isProviderBalanceError(error)).toBe(true); @@ -288,6 +296,19 @@ test("nested billing details and generic quota limits are kept distinct", () => expect(isProviderBalanceError(new Error("该模型额度不足(HTTP 429)"))).toBe(true); }); +test("98K wallet rejection remains a balance error instead of rate limiting", () => { + for (const message of [ + "The API key does not have enough available quota.", + "The API key does not have enough available quota for this request.", + ]) { + const error = Object.assign(new Error(`${message} (HTTP 429)`), { status: 429 }); + expect(isProviderBalanceError(error)).toBe(true); + expect(isRateLimitError(error)).toBe(false); + expect(getRequestFailureDetails(error)).toMatchObject({ code: "PROVIDER_BALANCE_EXHAUSTED" }); + expect(getUserFacingImageGenerationErrorMessage(error)).toContain("与您的积分余额无关"); + } +}); + test("confirmed balance failures remain isolated until explicitly restored", () => { const model = { healthLastEventType: "quota",