From be20f865ace5831ea7c6e79a05a2b4cec3923b27 Mon Sep 17 00:00:00 2001 From: bentrd Date: Mon, 2 Mar 2026 08:40:55 +0100 Subject: [PATCH] fix: recover from 413 Request Entity Too Large via auto-compaction (#14707) Co-authored-by: Noam Bressler --- packages/opencode/src/provider/error.ts | 3 +- packages/opencode/src/session/compaction.ts | 113 ++++++++++++++++---- packages/opencode/src/session/message-v2.ts | 44 +++++--- packages/opencode/src/session/processor.ts | 48 +++++---- packages/opencode/src/session/prompt.ts | 2 + 5 files changed, 152 insertions(+), 58 deletions(-) diff --git a/packages/opencode/src/provider/error.ts b/packages/opencode/src/provider/error.ts index 86990a205..945d29f97 100644 --- a/packages/opencode/src/provider/error.ts +++ b/packages/opencode/src/provider/error.ts @@ -19,6 +19,7 @@ export namespace ProviderError { /context window exceeds limit/i, // MiniMax /exceeded model token limit/i, // Kimi For Coding, Moonshot /context[_ ]length[_ ]exceeded/i, // Generic fallback + /request entity too large/i, // HTTP 413 ] function isOpenAiErrorRetryable(e: APICallError) { @@ -177,7 +178,7 @@ export namespace ProviderError { export function parseAPICallError(input: { providerID: string; error: APICallError }): ParsedAPICallError { const m = message(input.providerID, input.error) - if (isOverflow(m)) { + if (isOverflow(m) || input.error.statusCode === 413) { return { type: "context_overflow", message: m, diff --git a/packages/opencode/src/session/compaction.ts b/packages/opencode/src/session/compaction.ts index 924542605..a0f1c55c4 100644 --- a/packages/opencode/src/session/compaction.ts +++ b/packages/opencode/src/session/compaction.ts @@ -104,8 +104,31 @@ export namespace SessionCompaction { sessionID: string abort: AbortSignal auto: boolean + overflow?: boolean }) { const userMessage = input.messages.findLast((m) => m.info.id === input.parentID)!.info as MessageV2.User + + let messages = input.messages + let replay: MessageV2.WithParts | undefined + if (input.overflow) { + const idx = input.messages.findIndex((m) => m.info.id === input.parentID) + for (let i = idx - 1; i >= 0; i--) { + const msg = input.messages[i] + if (msg.info.role === "user" && !msg.parts.some((p) => p.type === "compaction")) { + replay = msg + messages = input.messages.slice(0, i) + break + } + } + const hasContent = replay && messages.some( + (m) => m.info.role === "user" && !m.parts.some((p) => p.type === "compaction"), + ) + if (!hasContent) { + replay = undefined + messages = input.messages + } + } + const agent = await Agent.get("compaction") const model = agent.model ? await Provider.getModel(agent.model.providerID, agent.model.modelID) @@ -185,7 +208,7 @@ When constructing the summary, try to stick to this template: tools: {}, system: [], messages: [ - ...MessageV2.toModelMessages(input.messages, model), + ...MessageV2.toModelMessages(messages, model, { stripMedia: true }), { role: "user", content: [ @@ -199,29 +222,71 @@ When constructing the summary, try to stick to this template: model, }) + if (result === "compact") { + processor.message.error = new MessageV2.ContextOverflowError({ + message: replay + ? "Conversation history too large to compact - exceeds model context limit" + : "Session too large to compact - context exceeds model limit even after stripping media", + }).toObject() + processor.message.finish = "error" + await Session.updateMessage(processor.message) + return "stop" + } + if (result === "continue" && input.auto) { - const continueMsg = await Session.updateMessage({ - id: Identifier.ascending("message"), - role: "user", - sessionID: input.sessionID, - time: { - created: Date.now(), - }, - agent: userMessage.agent, - model: userMessage.model, - }) - await Session.updatePart({ - id: Identifier.ascending("part"), - messageID: continueMsg.id, - sessionID: input.sessionID, - type: "text", - synthetic: true, - text: "Continue if you have next steps, or stop and ask for clarification if you are unsure how to proceed.", - time: { - start: Date.now(), - end: Date.now(), - }, - }) + if (replay) { + const original = replay.info as MessageV2.User + const replayMsg = await Session.updateMessage({ + id: Identifier.ascending("message"), + role: "user", + sessionID: input.sessionID, + time: { created: Date.now() }, + agent: original.agent, + model: original.model, + format: original.format, + tools: original.tools, + system: original.system, + variant: original.variant, + }) + for (const part of replay.parts) { + if (part.type === "compaction") continue + const replayPart = + part.type === "file" && MessageV2.isMedia(part.mime) + ? { type: "text" as const, text: `[Attached ${part.mime}: ${part.filename ?? "file"}]` } + : part + await Session.updatePart({ + ...replayPart, + id: Identifier.ascending("part"), + messageID: replayMsg.id, + sessionID: input.sessionID, + }) + } + } else { + const continueMsg = await Session.updateMessage({ + id: Identifier.ascending("message"), + role: "user", + sessionID: input.sessionID, + time: { created: Date.now() }, + agent: userMessage.agent, + model: userMessage.model, + }) + const text = + (input.overflow + ? "The previous request exceeded the provider's size limit due to large media attachments. The conversation was compacted and media files were removed from context. If the user was asking about attached images or files, explain that the attachments were too large to process and suggest they try again with smaller or fewer files.\n\n" + : "") + "Continue if you have next steps, or stop and ask for clarification if you are unsure how to proceed." + await Session.updatePart({ + id: Identifier.ascending("part"), + messageID: continueMsg.id, + sessionID: input.sessionID, + type: "text", + synthetic: true, + text, + time: { + start: Date.now(), + end: Date.now(), + }, + }) + } } if (processor.message.error) return "stop" Bus.publish(Event.Compacted, { sessionID: input.sessionID }) @@ -237,6 +302,7 @@ When constructing the summary, try to stick to this template: modelID: z.string(), }), auto: z.boolean(), + overflow: z.boolean().optional(), }), async (input) => { const msg = await Session.updateMessage({ @@ -255,6 +321,7 @@ When constructing the summary, try to stick to this template: sessionID: msg.sessionID, type: "compaction", auto: input.auto, + overflow: input.overflow, }) }, ) diff --git a/packages/opencode/src/session/message-v2.ts b/packages/opencode/src/session/message-v2.ts index 178751a22..5b4e7bdbc 100644 --- a/packages/opencode/src/session/message-v2.ts +++ b/packages/opencode/src/session/message-v2.ts @@ -17,6 +17,10 @@ import { type SystemError } from "bun" import type { Provider } from "@/provider/provider" export namespace MessageV2 { + export function isMedia(mime: string) { + return mime.startsWith("image/") || mime === "application/pdf" + } + export const OutputLengthError = NamedError.create("MessageOutputLengthError", z.object({})) export const AbortedError = NamedError.create("MessageAbortedError", z.object({ message: z.string() })) export const StructuredOutputError = NamedError.create( @@ -196,6 +200,7 @@ export namespace MessageV2 { export const CompactionPart = PartBase.extend({ type: z.literal("compaction"), auto: z.boolean(), + overflow: z.boolean().optional(), }).meta({ ref: "CompactionPart", }) @@ -488,7 +493,11 @@ export namespace MessageV2 { }) export type WithParts = z.infer - export function toModelMessages(input: WithParts[], model: Provider.Model): ModelMessage[] { + export function toModelMessages( + input: WithParts[], + model: Provider.Model, + options?: { stripMedia?: boolean }, + ): ModelMessage[] { const result: UIMessage[] = [] const toolNames = new Set() // Track media from tool results that need to be injected as user messages @@ -562,13 +571,21 @@ export namespace MessageV2 { text: part.text, }) // text/plain and directory files are converted into text parts, ignore them - if (part.type === "file" && part.mime !== "text/plain" && part.mime !== "application/x-directory") - userMessage.parts.push({ - type: "file", - url: part.url, - mediaType: part.mime, - filename: part.filename, - }) + if (part.type === "file" && part.mime !== "text/plain" && part.mime !== "application/x-directory") { + if (options?.stripMedia && isMedia(part.mime)) { + userMessage.parts.push({ + type: "text", + text: `[Attached ${part.mime}: ${part.filename ?? "file"}]`, + }) + } else { + userMessage.parts.push({ + type: "file", + url: part.url, + mediaType: part.mime, + filename: part.filename, + }) + } + } if (part.type === "compaction") { userMessage.parts.push({ @@ -618,14 +635,12 @@ export namespace MessageV2 { toolNames.add(part.tool) if (part.state.status === "completed") { const outputText = part.state.time.compacted ? "[Old tool result content cleared]" : part.state.output - const attachments = part.state.time.compacted ? [] : (part.state.attachments ?? []) + const attachments = part.state.time.compacted || options?.stripMedia ? [] : (part.state.attachments ?? []) // For providers that don't support media in tool results, extract media files // (images, PDFs) to be sent as a separate user message - const isMediaAttachment = (a: { mime: string }) => - a.mime.startsWith("image/") || a.mime === "application/pdf" - const mediaAttachments = attachments.filter(isMediaAttachment) - const nonMediaAttachments = attachments.filter((a) => !isMediaAttachment(a)) + const mediaAttachments = attachments.filter((a) => isMedia(a.mime)) + const nonMediaAttachments = attachments.filter((a) => !isMedia(a.mime)) if (!supportsMediaInToolResults && mediaAttachments.length > 0) { media.push(...mediaAttachments) } @@ -802,7 +817,8 @@ export namespace MessageV2 { msg.parts.some((part) => part.type === "compaction") ) break - if (msg.info.role === "assistant" && msg.info.summary && msg.info.finish) completed.add(msg.info.parentID) + if (msg.info.role === "assistant" && msg.info.summary && msg.info.finish && !msg.info.error) + completed.add(msg.info.parentID) } result.reverse() return result diff --git a/packages/opencode/src/session/processor.ts b/packages/opencode/src/session/processor.ts index e7532d200..67edc0ecf 100644 --- a/packages/opencode/src/session/processor.ts +++ b/packages/opencode/src/session/processor.ts @@ -279,7 +279,10 @@ export namespace SessionProcessor { sessionID: input.sessionID, messageID: input.assistantMessage.parentID, }) - if (await SessionCompaction.isOverflow({ tokens: usage.tokens, model: input.model })) { + if ( + !input.assistantMessage.summary && + (await SessionCompaction.isOverflow({ tokens: usage.tokens, model: input.model })) + ) { needsCompaction = true } break @@ -354,27 +357,32 @@ export namespace SessionProcessor { }) const error = MessageV2.fromError(e, { providerID: input.model.providerID }) if (MessageV2.ContextOverflowError.isInstance(error)) { - // TODO: Handle context overflow error - } - const retry = SessionRetry.retryable(error) - if (retry !== undefined) { - attempt++ - const delay = SessionRetry.delay(attempt, error.name === "APIError" ? error : undefined) - SessionStatus.set(input.sessionID, { - type: "retry", - attempt, - message: retry, - next: Date.now() + delay, + needsCompaction = true + Bus.publish(Session.Event.Error, { + sessionID: input.sessionID, + error, }) - await SessionRetry.sleep(delay, input.abort).catch(() => {}) - continue + } else { + const retry = SessionRetry.retryable(error) + if (retry !== undefined) { + attempt++ + const delay = SessionRetry.delay(attempt, error.name === "APIError" ? error : undefined) + SessionStatus.set(input.sessionID, { + type: "retry", + attempt, + message: retry, + next: Date.now() + delay, + }) + await SessionRetry.sleep(delay, input.abort).catch(() => {}) + continue + } + input.assistantMessage.error = error + Bus.publish(Session.Event.Error, { + sessionID: input.assistantMessage.sessionID, + error: input.assistantMessage.error, + }) + SessionStatus.set(input.sessionID, { type: "idle" }) } - input.assistantMessage.error = error - Bus.publish(Session.Event.Error, { - sessionID: input.assistantMessage.sessionID, - error: input.assistantMessage.error, - }) - SessionStatus.set(input.sessionID, { type: "idle" }) } if (snapshot) { const patch = await Snapshot.patch(snapshot) diff --git a/packages/opencode/src/session/prompt.ts b/packages/opencode/src/session/prompt.ts index 75bd3c9df..4f77920cc 100644 --- a/packages/opencode/src/session/prompt.ts +++ b/packages/opencode/src/session/prompt.ts @@ -533,6 +533,7 @@ export namespace SessionPrompt { abort, sessionID, auto: task.auto, + overflow: task.overflow, }) if (result === "stop") break continue @@ -707,6 +708,7 @@ export namespace SessionPrompt { agent: lastUser.agent, model: lastUser.model, auto: true, + overflow: !processor.message.finish, }) } continue