@apifuse/provider-sdk 2.2.0-beta.62 → 2.2.0-beta.64

package/src/runtime/stt-telemetry.ts

@@ -0,0 +1,282 @@
+import type { SttContext } from "../types.js";
+import {
+	add,
+	capabilityErrorText,
+	capabilityFacts,
+	capabilityMetadata,
+	capabilityText,
+	inputBytes,
+	integer,
+	observeCapability,
+	registerCapabilityInput,
+} from "./capability-telemetry.js";
+import {
+	type ClosedEnum,
+	closedEnum,
+	type SpanIndex,
+	type TelemetryContributor,
+} from "./request-telemetry.js";
+export type SttTelemetryBackend = "cloudflare-workers-ai" | "custom" | "unavailable";
+export type SttTelemetryEngine = "workers-ai" | "custom";
+export type SttTelemetryErrorCode =
+	| "STT_UNAVAILABLE"
+	| "UNSUPPORTED_STT_BACKEND"
+	| "STT_UPSTREAM_FAILED"
+	| "STT_AUDIO_TOO_LARGE"
+	| "UNSUPPORTED_STT_OPTION"
+	| "INVALID_STT_AUDIO"
+	| "INVALID_STT_VERIFICATION_CODE_OPTIONS"
+	| "NO_CODE_FOUND"
+	| "AMBIGUOUS_CODE"
+	| "transport_network_error"
+	| "transport_timeout"
+	| "other";
+export type SttTelemetryModel = "whisper-large-v3-turbo";
+const ERROR_CODES: ReadonlySet<string> = new Set([
+	"STT_UNAVAILABLE",
+	"UNSUPPORTED_STT_BACKEND",
+	"STT_UPSTREAM_FAILED",
+	"STT_AUDIO_TOO_LARGE",
+	"UNSUPPORTED_STT_OPTION",
+	"INVALID_STT_AUDIO",
+	"INVALID_STT_VERIFICATION_CODE_OPTIONS",
+	"NO_CODE_FOUND",
+	"AMBIGUOUS_CODE",
+	"transport_network_error",
+	"transport_timeout",
+	"other",
+]);
+const MODELS = new Map<string, SttTelemetryModel>([
+	["@cf/openai/whisper-large-v3-turbo", "whisper-large-v3-turbo"],
+]);
+function errorCode(error: unknown): SttTelemetryErrorCode {
+	const code = error && typeof error === "object" && "code" in error ? error.code : undefined;
+	return typeof code === "string" && ERROR_CODES.has(code)
+		? (code as SttTelemetryErrorCode)
+		: "other";
+}
+export type SttTelemetryEvent = {
+	backend: SttTelemetryBackend;
+	engine: SttTelemetryEngine;
+	model?: string;
+	ms: number;
+	status?: number;
+	audioBytes?: number;
+	durationMs?: number;
+	usage?: number;
+	warnings?: number;
+	errorCode?: SttTelemetryErrorCode;
+	diagnostics?: string;
+};
+export interface SttTelemetrySink {
+	record(event: SttTelemetryEvent): void;
+	markTelemetryFailed?(): void;
+}
+export type SttTelemetryLogPayload = {
+	telemetryFailed?: true;
+	diagnostics?: string[];
+	backend: SttTelemetryBackend;
+	engine: SttTelemetryEngine;
+	model?: string;
+	ms: number;
+	status?: number;
+	audioBytes: number;
+	durationMs: number;
+	usage: number;
+	warnings: number;
+	lastErrorCode?: SttTelemetryErrorCode;
+
+	samples?: Array<Omit<SttTelemetryEvent, "backend" | "engine" | "model" | "diagnostics">>;
+	samplesDropped?: number;
+};
+export type SttTelemetryHeaderPayload = {
+	backend: ClosedEnum<SttTelemetryBackend>;
+	engine: ClosedEnum<SttTelemetryEngine>;
+	model?: ClosedEnum<SttTelemetryModel>;
+	ms: number;
+	status?: number;
+	audioBytes: number;
+	durationMs: number;
+	usage: number;
+	warnings: number;
+	lastErrorCode?: ClosedEnum<SttTelemetryErrorCode>;
+
+	samples?: Array<{
+		ms: number;
+		status?: number;
+		audioBytes?: number;
+		durationMs?: number;
+		usage?: number;
+		warnings?: number;
+		errorCode?: ClosedEnum<SttTelemetryErrorCode>;
+	}>;
+	samplesDropped?: number;
+};
+export class SttTelemetryCollector
+	implements
+		SttTelemetrySink,
+		TelemetryContributor<SttTelemetryLogPayload, SttTelemetryHeaderPayload>
+{
+	readonly key = "stt" as const;
+	readonly #redact?: (text: string) => string;
+	#samples: NonNullable<SttTelemetryLogPayload["samples"]> = [];
+	#summary: Omit<SttTelemetryLogPayload, "samples" | "samplesDropped"> | undefined;
+	#diagnostics: string[] = [];
+	#dropped = 0;
+	#failed = false;
+	constructor(options: { redact?: (text: string) => string } = {}) {
+		this.#redact = options.redact;
+	}
+	markTelemetryFailed(): void {
+		this.#failed = true;
+		this.#summary ??= {
+			backend: "custom",
+			engine: "custom",
+			ms: 0,
+			audioBytes: 0,
+			durationMs: 0,
+			usage: 0,
+			warnings: 0,
+		};
+	}
+	record(event: SttTelemetryEvent): void {
+		const sample = {
+			ms: integer(event.ms),
+			...(event.status === undefined ? {} : { status: integer(event.status) }),
+			...(event.audioBytes === undefined ? {} : { audioBytes: integer(event.audioBytes) }),
+			...(event.durationMs === undefined ? {} : { durationMs: integer(event.durationMs) }),
+			...(event.usage === undefined ? {} : { usage: integer(event.usage) }),
+			...(event.warnings === undefined ? {} : { warnings: integer(event.warnings) }),
+			...(event.errorCode ? { errorCode: event.errorCode } : {}),
+
+			...(event.diagnostics === undefined
+				? {}
+				: { diagnostics: capabilityText(event.diagnostics, this.#redact, true) }),
+		};
+		const prior = this.#summary;
+		this.#summary = {
+			backend: event.backend,
+			engine: event.engine,
+			...(event.model === undefined ? {} : { model: capabilityText(event.model, this.#redact) }),
+			ms: add(prior?.ms ?? 0, sample.ms),
+			...(sample.status === undefined ? {} : { status: sample.status }),
+			audioBytes: add(prior?.audioBytes ?? 0, sample.audioBytes ?? 0),
+			durationMs: add(prior?.durationMs ?? 0, sample.durationMs ?? 0),
+			usage: add(prior?.usage ?? 0, sample.usage ?? 0),
+			warnings: add(prior?.warnings ?? 0, sample.warnings ?? 0),
+			...(event.errorCode ? { lastErrorCode: event.errorCode } : {}),
+		};
+		if (this.#samples.length < 24) {
+			const { diagnostics, ...closedSample } = sample;
+			this.#samples.push(closedSample);
+			if (diagnostics !== undefined) this.#diagnostics.push(diagnostics);
+		} else this.#dropped = add(this.#dropped, 1);
+	}
+	toLogPayload(_spans: SpanIndex): SttTelemetryLogPayload | undefined {
+		if (!this.#summary) return undefined;
+		return {
+			...this.#summary,
+			...(this.#diagnostics.length ? { diagnostics: [...this.#diagnostics] } : {}),
+			...(this.#failed ? { telemetryFailed: true as const } : {}),
+			samples: this.#samples.map((s) => ({ ...s })),
+			...(this.#dropped ? { samplesDropped: this.#dropped } : {}),
+		};
+	}
+	toHeaderPayload(log: SttTelemetryLogPayload): SttTelemetryHeaderPayload {
+		return {
+			backend: closedEnum(log.backend),
+			engine: closedEnum(log.engine),
+			...(log.model && MODELS.has(log.model) ? { model: closedEnum(MODELS.get(log.model)!) } : {}),
+			ms: log.ms,
+			...(log.status === undefined ? {} : { status: log.status }),
+			audioBytes: log.audioBytes,
+			durationMs: log.durationMs,
+			usage: log.usage,
+			warnings: log.warnings,
+			...(log.lastErrorCode ? { lastErrorCode: closedEnum(log.lastErrorCode) } : {}),
+
+			...(log.samples
+				? {
+						samples: log.samples.map((s) => ({
+							ms: s.ms,
+							...(s.status === undefined ? {} : { status: s.status }),
+							...(s.audioBytes === undefined ? {} : { audioBytes: s.audioBytes }),
+							...(s.durationMs === undefined ? {} : { durationMs: s.durationMs }),
+							...(s.usage === undefined ? {} : { usage: s.usage }),
+							...(s.warnings === undefined ? {} : { warnings: s.warnings }),
+							...(s.errorCode ? { errorCode: closedEnum(s.errorCode) } : {}),
+						})),
+					}
+				: {}),
+			...(log.samplesDropped ? { samplesDropped: log.samplesDropped } : {}),
+		};
+	}
+}
+export function bindSttTelemetry(
+	context: SttContext,
+	sink: SttTelemetrySink,
+	options: { backend?: SttTelemetryBackend; engine?: SttTelemetryEngine; model?: string } = {},
+): SttContext {
+	const metadata = capabilityMetadata(context);
+	const backend =
+		options.backend ?? (metadata?.backend as SttTelemetryBackend | undefined) ?? "custom";
+	const engine = options.engine ?? (metadata?.engine as SttTelemetryEngine | undefined) ?? "custom";
+	const model = options.model ?? metadata?.model;
+	async function run<
+		T extends {
+			text: string;
+			durationMs?: number;
+			usage?: { audioBytes?: number; audioDurationMs?: number };
+			warnings?: readonly unknown[];
+		},
+	>(input: { kind: string; data?: string; url?: string }, operation: () => Promise<T>): Promise<T> {
+		const started = performance.now();
+		const facts: { status?: number; finishReason?: string; diagnostics?: string } = {};
+		observeCapability(sink, () => registerCapabilityInput(input));
+		let result: T;
+		try {
+			result = await capabilityFacts.run(facts, operation);
+		} catch (error) {
+			observeCapability(sink, () => {
+				const status =
+					error &&
+					typeof error === "object" &&
+					"status" in error &&
+					typeof error.status === "number"
+						? error.status
+						: undefined;
+				return sink.record({
+					backend,
+					engine,
+					model,
+					ms: performance.now() - started,
+					status: facts.status ?? status,
+					audioBytes: inputBytes(input),
+					errorCode: errorCode(error),
+
+					diagnostics: `${capabilityErrorText(error)} ${facts.diagnostics ?? ""}`,
+				});
+			});
+			throw error;
+		}
+		observeCapability(sink, () =>
+			sink.record({
+				backend,
+				engine,
+				model: model,
+				ms: performance.now() - started,
+				status: facts.status,
+				audioBytes: result.usage?.audioBytes ?? inputBytes(input),
+				durationMs: result.durationMs,
+				usage: result.usage?.audioDurationMs,
+				warnings: result.warnings?.length ?? 0,
+			}),
+		);
+		return result;
+	}
+	return {
+		...context,
+		transcribe: (request) => run(request.audio, () => context.transcribe(request)),
+		extractVerificationCode: (text, opts) => context.extractVerificationCode(text, opts),
+	};
+}

package/src/runtime/stt.ts

@@ -1,4 +1,3 @@
-import { readDiagnosticEnv } from "./diagnostic-env.js";
 import { ProviderError, TransportError, ValidationError } from "../errors.js";
 import type {
 	Bcp47Locale,
@@ -15,6 +14,13 @@ import type {
 	VerificationCodeCandidateSource,
 	VerificationCodeExtractionResult,
 } from "../types.js";
+import { readCapabilityErrorBody } from "./capability-response.js";
+import {
+	captureCapability,
+	registerCapabilityInput,
+	tagCapability,
+} from "./capability-telemetry.js";
+import { readDiagnosticEnv, registerDiagnosticValue } from "./diagnostic-env.js";
 import { createTimeoutController, isTimeoutLikeError } from "./timeout.js";
 
 export const APIFUSE__STT__BACKEND_ENV = "APIFUSE__STT__BACKEND";
@@ -50,15 +56,18 @@ function providerError(message: string, options: { code: string; fix?: string })
 }
 
 function createErrorSttClient(options: ErrorSttClientOptions): SttContext {
-	return {
-		async transcribe() {
-			throw providerError(options.message, {
-				code: options.code,
-				fix: options.fix,
-			});
+	return tagCapability(
+		{
+			async transcribe() {
+				throw providerError(options.message, {
+					code: options.code,
+					fix: options.fix,
+				});
+			},
+			extractVerificationCode,
 		},
-		extractVerificationCode,
-	};
+		{ backend: "unavailable", engine: "custom" },
+	);
 }
 
 export function createUnsupportedSttClient(reason?: string): SttContext {
@@ -273,72 +282,85 @@ export function createCloudflareWorkersAiSttClient(
 ): SttContext {
 	const model = options.model ?? DEFAULT_CLOUDFLARE_WORKERS_AI_STT_MODEL;
 	const runFetch = options.fetch ?? fetch;
-	return {
-		async transcribe(request) {
-			const warnings = [];
-			if (request.initialPrompt && effectivePromptPolicy(request) !== "custom-hint") {
-				const warning = warnOrThrowUnsupportedOption(
-					request,
-					"initialPrompt is honored only when promptPolicy is custom-hint",
-				);
-				if (warning) warnings.push(warning);
-			}
-			const audioBytes = assertBase64Audio(request.audio, request.maxAudioBytes);
-			const timeout = createTimeoutController(request.timeoutMs ?? DEFAULT_STT_TIMEOUT_MS);
-			try {
-				let response: Response;
-				try {
-					response = await runFetch(
-						`https://api.cloudflare.com/client/v4/accounts/${encodeURIComponent(options.accountId)}/ai/run/${model}`,
-						{
-							method: "POST",
-							headers: {
-								Authorization: `Bearer ${options.apiToken}`,
-								"Content-Type": "application/json",
-							},
-							body: JSON.stringify(toCloudflareInput(request)),
-							signal: timeout.controller.signal,
-						},
+	return tagCapability(
+		{
+			async transcribe(request) {
+				registerCapabilityInput(request.audio);
+				registerDiagnosticValue(options.apiToken);
+				registerDiagnosticValue(`Bearer ${options.apiToken}`);
+				const warnings = [];
+				if (request.initialPrompt && effectivePromptPolicy(request) !== "custom-hint") {
+					const warning = warnOrThrowUnsupportedOption(
+						request,
+						"initialPrompt is honored only when promptPolicy is custom-hint",
 					);
-				} catch (error) {
-					throw toSttTransportError(error);
+					if (warning) warnings.push(warning);
 				}
-				const payload = await response.json().catch(() => undefined);
-				if (!response.ok) {
-					throw new TransportError("STT upstream request failed", {
-						code: "STT_UPSTREAM_FAILED",
-						status: response.status,
-						upstreamStatus: response.status,
-					});
-				}
-				const envelope = unknownRecord(payload);
-				if (envelope?.success === false) {
-					throw new TransportError("STT upstream returned an error", {
-						code: "STT_UPSTREAM_FAILED",
-						status: 502,
-					});
-				}
-				const transcript = toSttTranscript(payload, audioBytes);
-				const withWarnings =
-					warnings.length > 0
-						? {
-								...transcript,
-								warnings: [...(transcript.warnings ?? []), ...warnings],
-							}
-						: transcript;
-				if (request.mode === "otp" || request.verificationCode) {
-					return {
-						...withWarnings,
-						verificationCode: extractVerificationCode(withWarnings.text, request.verificationCode),
-					};
+				const audioBytes = assertBase64Audio(request.audio, request.maxAudioBytes);
+				const timeout = createTimeoutController(request.timeoutMs ?? DEFAULT_STT_TIMEOUT_MS);
+				try {
+					let response: Response;
+					try {
+						response = await runFetch(
+							`https://api.cloudflare.com/client/v4/accounts/${encodeURIComponent(options.accountId)}/ai/run/${model}`,
+							{
+								method: "POST",
+								headers: {
+									Authorization: `Bearer ${options.apiToken}`,
+									"Content-Type": "application/json",
+								},
+								body: JSON.stringify(toCloudflareInput(request)),
+								signal: timeout.controller.signal,
+							},
+						);
+					} catch (error) {
+						throw toSttTransportError(error);
+					}
+					captureCapability({ status: response.status });
+					if (!response.ok) {
+						const echoed = await readCapabilityErrorBody(response, timeout.controller.signal);
+						captureCapability({ diagnostics: echoed });
+						throw new TransportError("STT upstream request failed", {
+							code: "STT_UPSTREAM_FAILED",
+							status: response.status,
+							upstreamStatus: response.status,
+						});
+					}
+					const payload = await response.json().catch(() => undefined);
+					const envelope = unknownRecord(payload);
+					if (envelope?.success === false) {
+						captureCapability({ diagnostics: JSON.stringify(payload) });
+						throw new TransportError("STT upstream returned an error", {
+							code: "STT_UPSTREAM_FAILED",
+							status: 502,
+						});
+					}
+					const transcript = toSttTranscript(payload, audioBytes);
+					const withWarnings =
+						warnings.length > 0
+							? {
+									...transcript,
+									warnings: [...(transcript.warnings ?? []), ...warnings],
+								}
+							: transcript;
+					if (request.mode === "otp" || request.verificationCode) {
+						return {
+							...withWarnings,
+							verificationCode: extractVerificationCode(
+								withWarnings.text,
+								request.verificationCode,
+							),
+						};
+					}
+					return withWarnings;
+				} finally {
+					timeout.clear();
 				}
-				return withWarnings;
-			} finally {
-				timeout.clear();
-			}
+			},
+			extractVerificationCode,
 		},
-		extractVerificationCode,
-	};
+		{ backend: "cloudflare-workers-ai", engine: "workers-ai", model },
+	);
 }
 
 const EN_DIGITS: Record<string, string> = {