Merge remote-tracking branch 'origin/farm/3f9eb0f8/widen-first-event-timeout-deepseek-v4-reasoning'

2026-06-09 19:19:07 +02:00
parent d97a89999c f5f4f2265b
commit d8d40569d6
3 changed files with 82 additions and 7 deletions
@@ -10,6 +10,10 @@

 - Added `antigravityRankingStrategy` and registered it as the default `CredentialRankingStrategy` for `google-antigravity`, so multi-account selection consumes the per-counter Antigravity usage reports (sorted ascending by `remainingFraction` in `fetchAntigravityUsage`) before falling back to round-robin — preventing the exhausted-counter credential from being chosen first when an unblocked sibling has headroom ([#2187](https://github.com/can1357/oh-my-pi/issues/2187)).

+### Fixed
+
+- Widened the OpenAI-completions first-event watchdog floor from 120s to 300s for DeepSeek V4 reasoning models hosted on the official DeepSeek API. The reasoner emits no SSE bytes until its private chain-of-thought finishes, which routinely takes longer than the generic 100s first-event budget under load — every chat then aborted with `OpenAI completions stream timed out while waiting for the first event` and silently retried. Mirrors the existing GLM coding-plan widening ([#2177](https://github.com/can1357/oh-my-pi/issues/2177)).
+
 ## [15.10.8] - 2026-06-09
 ### Added

@@ -392,17 +392,36 @@ const OPENAI_COMPLETIONS_FIRST_EVENT_TIMEOUT_MESSAGE =
 const GLM_CODING_PLAN_STREAM_IDLE_TIMEOUT_MS = 600_000;
 const GLM_CODING_PLAN_MODEL_PATTERN = /^glm-5(?:[.-]|$)/i;

-/** Returns the widened OpenAI stream watchdog floor for slow GLM coding-plan reasoning models. */
+// DeepSeek V4 reasoning models on the official api.deepseek.com emit no SSE
+// bytes while the model finishes its private chain-of-thought, which routinely
+// takes longer than the generic 100s first-event floor under load (issue
+// #2177). Mirror the GLM coding-plan widening: a 5-minute idle floor lifts the
+// first-event watchdog (it floors at idle) without changing the runtime
+// streaming behavior, so reasoning warm-ups stop aborting and retrying.
+const DEEPSEEK_REASONING_STREAM_IDLE_TIMEOUT_MS = 300_000;
+
+function isDirectDeepseekReasoningModel(model: Model<"openai-completions">): boolean {
+	if (!model.reasoning) return false;
+	if (model.provider === "deepseek") return true;
+	return model.baseUrl.toLowerCase().includes("api.deepseek.com");
+}
+
+/** Returns the widened OpenAI stream watchdog floor for slow reasoning models hosted on OpenAI-compatible endpoints. */
 export function getOpenAICompletionsStreamIdleTimeoutFallbackMs(
 	model: Model<"openai-completions">,
 ): number | undefined {
-	if (!GLM_CODING_PLAN_MODEL_PATTERN.test(model.id)) return undefined;
-	if (model.provider === "zhipu-coding-plan" || model.provider === "zai")
-		return GLM_CODING_PLAN_STREAM_IDLE_TIMEOUT_MS;
+	if (GLM_CODING_PLAN_MODEL_PATTERN.test(model.id)) {
+		if (model.provider === "zhipu-coding-plan" || model.provider === "zai")
+			return GLM_CODING_PLAN_STREAM_IDLE_TIMEOUT_MS;

-	const baseUrl = model.baseUrl.toLowerCase();
-	if (baseUrl.includes("open.bigmodel.cn") || baseUrl.includes("api.z.ai")) {
-		return GLM_CODING_PLAN_STREAM_IDLE_TIMEOUT_MS;
+		const baseUrl = model.baseUrl.toLowerCase();
+		if (baseUrl.includes("open.bigmodel.cn") || baseUrl.includes("api.z.ai")) {
+			return GLM_CODING_PLAN_STREAM_IDLE_TIMEOUT_MS;
+		}
+	}
+
+	if (isDirectDeepseekReasoningModel(model)) {
+		return DEEPSEEK_REASONING_STREAM_IDLE_TIMEOUT_MS;
 	}

 	return undefined;
@@ -103,6 +103,58 @@ describe("getOpenAICompletionsStreamIdleTimeoutFallbackMs", () => {
 		expect(getOpenAICompletionsStreamIdleTimeoutFallbackMs(model)).toBe(600_000);
 	});

+	it("widens DeepSeek V4 reasoning streams on the official DeepSeek API", () => {
+		const model = {
+			...openAICompletionsModel,
+			id: "deepseek-v4-pro",
+			name: "DeepSeek V4 Pro",
+			provider: "deepseek",
+			baseUrl: "https://api.deepseek.com",
+			reasoning: true,
+		} satisfies Model<"openai-completions">;
+
+		expect(getOpenAICompletionsStreamIdleTimeoutFallbackMs(model)).toBe(300_000);
+	});
+
+	it("widens DeepSeek reasoning streams routed through an aliased OpenAI-compatible provider id", () => {
+		const model = {
+			...openAICompletionsModel,
+			id: "deepseek-v4-pro",
+			name: "DeepSeek V4 Pro",
+			provider: "openai",
+			baseUrl: "https://api.deepseek.com/v1",
+			reasoning: true,
+		} satisfies Model<"openai-completions">;
+
+		expect(getOpenAICompletionsStreamIdleTimeoutFallbackMs(model)).toBe(300_000);
+	});
+
+	it("leaves non-reasoning DeepSeek-hosted models on the global timeout", () => {
+		const model = {
+			...openAICompletionsModel,
+			id: "deepseek-chat",
+			name: "DeepSeek Chat",
+			provider: "deepseek",
+			baseUrl: "https://api.deepseek.com",
+			reasoning: false,
+		} satisfies Model<"openai-completions">;
+
+		expect(getOpenAICompletionsStreamIdleTimeoutFallbackMs(model)).toBeUndefined();
+	});
+
+	it("does not widen DeepSeek V4 reasoning models hosted on third-party OpenAI-compatible proxies", () => {
+		const model = {
+			...openAICompletionsModel,
+			id: "deepseek-v4-pro",
+			name: "DeepSeek V4 Pro",
+			provider: "aimlapi",
+			baseUrl: "https://api.aimlapi.com/v1",
+			reasoning: true,
+		} satisfies Model<"openai-completions">;
+
+		expect(getOpenAICompletionsStreamIdleTimeoutFallbackMs(model)).toBeUndefined();
+	});
+
 	it("keeps ordinary OpenAI-compatible models on the global timeout", () => {
 		expect(getOpenAICompletionsStreamIdleTimeoutFallbackMs(openAICompletionsModel)).toBeUndefined();
 	});