@rangojs/router 0.0.0-experimental.8 → 0.0.0-experimental.80daa23e

package/src/router/telemetry-otel.ts

@@ -0,0 +1,270 @@
+/**
+ * OpenTelemetry adapters for the router.
+ *
+ * Two adapters, two surfaces — matching the split in instrument.ts:
+ *
+ *   - createOTelTracing(tracer): the OTel adapter for the `tracing` SLOT — the
+ *     canonical phase-span layer. It bridges observePhase's callback boundary
+ *     onto OTel's callback-bound `startActiveSpan`, so the router's phase spans
+ *     (rango.request/middleware/action/loader/handler/render/ssr/response/
+ *     background) nest by async context and the loader's own OTel spans
+ *     (db/fetch) land under rango.loader. This is the OTel equivalent of
+ *     createCloudflareTracing — pass it to `createRouter({ tracing })`.
+ *
+ *   - createOTelSink(tracer): the OTel adapter for the `telemetry` SLOT — a
+ *     TelemetrySink for the EVENT-shaped facts (handler errors, cache decisions,
+ *     revalidation decisions, timeouts, origin rejections). It emits one instant
+ *     OTel span per fact. It deliberately does NOT emit request/loader phase
+ *     spans: those are owned by the tracing slot (createOTelTracing) so the two
+ *     layers don't produce duplicate rango.request / rango.loader spans.
+ *
+ * The core router stays OTel-agnostic — these adapters accept a standard OTel
+ * Tracer (structurally typed, no import needed) and bridge the gap.
+ *
+ * Usage:
+ *   import { trace } from "@opentelemetry/api";
+ *   import { createRouter, createOTelTracing, createOTelSink } from "@rangojs/router";
+ *
+ *   const tracer = trace.getTracer("my-app");
+ *   const router = createRouter({
+ *     tracing: createOTelTracing(tracer),   // phase spans (callback-bound)
+ *     telemetry: createOTelSink(tracer),     // discrete-fact instant spans
+ *   });
+ *
+ * Faithful nesting requires an OTel async context manager
+ * (AsyncLocalStorageContextManager) configured in your OTel setup — standard for
+ * any startActiveSpan-based instrumentation.
+ */
+
+import type { TelemetrySink, TelemetryEvent } from "./telemetry.js";
+import { runThenSettle } from "./tracing.js";
+import type {
+  RouterTracingConfig,
+  SpanRunner,
+  TracingToggleOptions,
+} from "./tracing.js";
+
+// ---------------------------------------------------------------------------
+// Minimal OTel-compatible types (structurally typed, no import needed)
+// ---------------------------------------------------------------------------
+
+/**
+ * Minimal Span interface compatible with @opentelemetry/api Span.
+ * Only the methods used by the adapters are declared.
+ */
+export interface OTelSpan {
+  setAttribute(key: string, value: string | number | boolean): OTelSpan | void;
+  setStatus(status: { code: number; message?: string }): OTelSpan | void;
+  recordException(exception: Error): void;
+  end(): void;
+}
+
+/**
+ * Minimal Tracer interface for the EVENT sink (createOTelSink): only `startSpan`
+ * is used (one instant span per discrete fact). Kept narrow so a custom,
+ * event-only tracer that does not implement `startActiveSpan` still satisfies it.
+ * The real `@opentelemetry/api` Tracer has both methods, so it satisfies this and
+ * `OTelActiveSpanTracer` below.
+ */
+export interface OTelTracer {
+  startSpan(
+    name: string,
+    options?: {
+      attributes?: Record<string, string | number | boolean>;
+    },
+  ): OTelSpan;
+}
+
+/**
+ * Minimal Tracer interface for the PHASE-SPAN adapter (createOTelTracing): only
+ * `startActiveSpan` is used (callback-bound, so the span is active for the work
+ * and child spans nest). Declared separately from `OTelTracer` so each factory
+ * requires exactly the method it calls.
+ */
+export interface OTelActiveSpanTracer {
+  startActiveSpan<T>(name: string, fn: (span: OTelSpan) => T): T;
+}
+
+// OTel SpanStatusCode constants (mirrors @opentelemetry/api values)
+const STATUS_ERROR = 2;
+
+// ---------------------------------------------------------------------------
+// Tracing adapter: phase spans via startActiveSpan (the `tracing` slot)
+// ---------------------------------------------------------------------------
+
+/**
+ * Options for createOTelTracing. Alias of the shared {@link TracingToggleOptions}
+ * (`enabled` master switch + per-phase `spans` toggles); the name is public API.
+ */
+export type OTelTracingOptions = TracingToggleOptions;
+
+/**
+ * Create the tracing config that maps the router's phases onto OTel spans via
+ * `tracer.startActiveSpan`, which runs the work inside the span's active context
+ * (so child spans nest) and returns the work's value unchanged. The span ends
+ * when that value — or, for async work, the returned promise — settles, matching
+ * observePhase's contract. When the work throws or rejects, the exception is
+ * recorded and the span status is set to ERROR before it ends, so a failed phase
+ * stays visible in the trace (the old createOTelSink error path is preserved here
+ * now that phase spans live in this adapter). Pass the result to
+ * `createRouter({ tracing })`.
+ *
+ * @see createCloudflareTracing (`@rangojs/router/cloudflare`) for the same slot
+ *   using Cloudflare Workers native custom spans.
+ */
+export function createOTelTracing(
+  tracer: OTelActiveSpanTracer,
+  options: OTelTracingOptions = {},
+): RouterTracingConfig {
+  const runner: SpanRunner = <T>(name: string, fn: (span: OTelSpan) => T): T =>
+    tracer.startActiveSpan(
+      name,
+      (span): T =>
+        runThenSettle(
+          () => fn(span),
+          (error) => {
+            // On failure record the exception + ERROR status before ending, so a
+            // failed phase stays visible in the trace.
+            if (error !== undefined) {
+              if (error instanceof Error) {
+                span.recordException(error);
+                span.setStatus({ code: STATUS_ERROR, message: error.message });
+              } else {
+                span.setStatus({ code: STATUS_ERROR });
+              }
+            }
+            span.end();
+          },
+        ),
+    );
+
+  return {
+    runner,
+    enabled: options.enabled ?? true,
+    spans: options.spans,
+  };
+}
+
+// ---------------------------------------------------------------------------
+// Telemetry sink: discrete-fact instant spans (the `telemetry` slot)
+// ---------------------------------------------------------------------------
+
+/**
+ * Create a TelemetrySink that maps the router's discrete-fact events to instant
+ * OTel spans. One span per fact; no duration spans.
+ *
+ * Fact mapping:
+ *   - handler.error              -> "rango.handler.error" (error)
+ *   - cache.decision             -> "rango.cache.decision"
+ *   - revalidation.decision      -> "rango.revalidation.decision"
+ *   - request.timeout            -> "rango.request.timeout" (error)
+ *   - request.origin-rejected    -> "rango.request.origin-rejected" (error)
+ *
+ * Request and loader PHASE spans are intentionally NOT emitted here — they are
+ * owned by the tracing slot (createOTelTracing) so the two layers cannot produce
+ * duplicate rango.request / rango.loader spans. request.start/end and
+ * loader.start/end events are no-ops for this sink.
+ *
+ * Attributes use the `rango.*` namespace for router-specific data and
+ * `http.method` / `http.route` for HTTP semantics.
+ */
+export function createOTelSink(tracer: OTelTracer): TelemetrySink {
+  const instant = (
+    name: string,
+    attributes: Record<string, string | number | boolean>,
+  ): OTelSpan => tracer.startSpan(name, { attributes });
+
+  return {
+    emit(event: TelemetryEvent): void {
+      switch (event.type) {
+        case "handler.error": {
+          const attrs: Record<string, string | number | boolean> = {
+            "rango.handled_by_boundary": event.handledByBoundary,
+          };
+          if (event.segmentId) attrs["rango.segment_id"] = event.segmentId;
+          if (event.segmentType)
+            attrs["rango.segment_type"] = event.segmentType;
+          if (event.pathname) attrs["http.route"] = event.pathname;
+          if (event.routeKey) attrs["rango.route_key"] = event.routeKey;
+          if (event.params)
+            attrs["rango.params"] = JSON.stringify(event.params);
+
+          const span = instant("rango.handler.error", attrs);
+          span.recordException(event.error);
+          span.setStatus({ code: STATUS_ERROR, message: event.error.message });
+          span.end();
+          break;
+        }
+
+        case "cache.decision": {
+          const attrs: Record<string, string | number | boolean> = {
+            "http.route": event.pathname,
+            "rango.route_key": event.routeKey,
+            "rango.cache.hit": event.hit,
+            "rango.cache.should_revalidate": event.shouldRevalidate,
+          };
+          if (event.source) attrs["rango.cache.source"] = event.source;
+          instant("rango.cache.decision", attrs).end();
+          break;
+        }
+
+        case "revalidation.decision": {
+          instant("rango.revalidation.decision", {
+            "rango.segment_id": event.segmentId,
+            "http.route": event.pathname,
+            "rango.route_key": event.routeKey,
+            "rango.revalidate": event.shouldRevalidate,
+          }).end();
+          break;
+        }
+
+        case "request.timeout": {
+          const attrs: Record<string, string | number | boolean> = {
+            "rango.phase": event.phase,
+            "http.route": event.pathname,
+            "rango.duration_ms": event.durationMs,
+            "rango.timeout.custom_handler": event.customHandler,
+          };
+          if (event.routeKey) attrs["rango.route_key"] = event.routeKey;
+          if (event.actionId) attrs["rango.action_id"] = event.actionId;
+          if (event.render) {
+            attrs["rango.render.mode"] = event.render.mode;
+            attrs["rango.render.phase"] = event.render.phase;
+            attrs["rango.render.state"] = event.render.state;
+            attrs["rango.render.completed"] = event.render.completed;
+            attrs["rango.render.total"] = event.render.total;
+            if (event.render.phaseDurationMs !== undefined) {
+              attrs["rango.render.phase_duration_ms"] =
+                event.render.phaseDurationMs;
+            }
+          }
+          const span = instant("rango.request.timeout", attrs);
+          span.setStatus({
+            code: STATUS_ERROR,
+            message: `timeout: ${event.phase}`,
+          });
+          span.end();
+          break;
+        }
+
+        case "request.origin-rejected": {
+          const attrs: Record<string, string | number | boolean> = {
+            "http.method": event.method,
+            "http.route": event.pathname,
+            "rango.phase": event.phase,
+          };
+          if (event.origin) attrs["rango.origin"] = event.origin;
+          if (event.host) attrs["http.host"] = event.host;
+          const span = instant("rango.request.origin-rejected", attrs);
+          span.setStatus({ code: STATUS_ERROR, message: "origin rejected" });
+          span.end();
+          break;
+        }
+
+        // request.start/end/error and loader.start/end/error are phase events;
+        // their spans are owned by the tracing slot (createOTelTracing), so this
+        // sink ignores them to avoid duplicate rango.request / rango.loader spans.
+      }
+    },
+  };
+}

package/src/router/telemetry.ts

@@ -0,0 +1,386 @@
+/**
+ * Router Telemetry Sink
+ *
+ * Internal event model for structured lifecycle events.
+ * The sink is optional and zero-cost when not configured.
+ *
+ * Emit points:
+ *   - request.start / request.end   (match-handlers.ts)
+ *   - request.error                  (match-handlers.ts catch blocks)
+ *   - request.origin-rejected        (rsc/handler.ts origin guard)
+ *   - loader.start / loader.end / loader.error  (loader-resolution.ts)
+ *   - handler.error                  (trackHandler catch, segment-resolution/helpers.ts)
+ *   - cache.decision                 (cache-lookup middleware)
+ *   - revalidation.decision          (revalidation evaluation)
+ */
+
+interface BaseEvent {
+  /** Monotonic timestamp from performance.now() */
+  timestamp: number;
+  /** Request ID (from header or generated) */
+  requestId?: string;
+}
+
+export interface RequestStartEvent extends BaseEvent {
+  type: "request.start";
+  method: string;
+  pathname: string;
+  /** "match" for full document requests, "matchPartial" for navigation */
+  transaction: "match" | "matchPartial";
+  isPartial: boolean;
+}
+
+export interface RequestEndEvent extends BaseEvent {
+  type: "request.end";
+  method: string;
+  pathname: string;
+  transaction: "match" | "matchPartial";
+  durationMs: number;
+  segmentCount: number;
+  cacheHit: boolean;
+  /**
+   * HTTP status when a Response ended the transaction — a thrown-Response
+   * short-circuit (redirect / auth gate carries the Response's status, e.g. 302),
+   * or dispatch()'s final response status. Absent for a normal render completion:
+   * the Response is built after match(), so match()/matchPartial() have no status
+   * to stamp there. Lets a sink split 3xx short-circuits from 2xx completions.
+   */
+  status?: number;
+}
+
+export interface RequestErrorEvent extends BaseEvent {
+  type: "request.error";
+  method: string;
+  pathname: string;
+  transaction: "match" | "matchPartial";
+  error: Error;
+  phase: string;
+  durationMs: number;
+}
+
+export interface LoaderStartEvent extends BaseEvent {
+  type: "loader.start";
+  segmentId: string;
+  loaderName: string;
+  pathname: string;
+}
+
+export interface LoaderEndEvent extends BaseEvent {
+  type: "loader.end";
+  segmentId: string;
+  loaderName: string;
+  pathname: string;
+  durationMs: number;
+  ok: boolean;
+}
+
+export interface LoaderErrorEvent extends BaseEvent {
+  type: "loader.error";
+  segmentId: string;
+  loaderName: string;
+  pathname: string;
+  error: Error;
+  handledByBoundary: boolean;
+}
+
+export interface HandlerErrorEvent extends BaseEvent {
+  type: "handler.error";
+  segmentId?: string;
+  segmentType?: string;
+  error: Error;
+  handledByBoundary: boolean;
+  pathname?: string;
+  routeKey?: string;
+  params?: Record<string, string>;
+}
+
+/**
+ * Per-segment (or coarse route-level) cache status carried on the
+ * cache.decision telemetry event and the X-Rango-Cache debug header.
+ *
+ * v1 is COARSE: the router's pipeline tracks cache decisions at the
+ * route/entry level (cacheHit/cacheSource/shouldRevalidate), not per
+ * individual segment. The `segments` array therefore contains a single
+ * route-level entry keyed by the route key. The shape is forward-compatible
+ * with genuine per-segment status if the pipeline later exposes it.
+ */
+export type CacheSegmentStatus =
+  | "hit"
+  | "miss"
+  | "stale"
+  | "prerendered"
+  | "passthrough";
+
+export interface CacheSegmentSignal {
+  /** Segment id (v1: the route key, since status is route-level). */
+  id: string;
+  /** Segment type (v1: "route" for the coarse route-level entry). */
+  type: string;
+  /** Resolved cache status for this segment. */
+  cacheStatus: CacheSegmentStatus;
+  /** Whether stale-while-revalidate was triggered for this segment. */
+  shouldRevalidate?: boolean;
+}
+
+export interface CacheDecisionEvent extends BaseEvent {
+  type: "cache.decision";
+  pathname: string;
+  routeKey: string;
+  hit: boolean;
+  /** Whether stale-while-revalidate was triggered */
+  shouldRevalidate: boolean;
+  source?: "runtime" | "prerender";
+  /**
+   * Optional per-segment (v1: coarse route-level) cache status. Present only
+   * when telemetry or the debug cache signal is enabled. Optional so existing
+   * sinks are unaffected.
+   */
+  segments?: CacheSegmentSignal[];
+}
+
+export interface RevalidationDecisionEvent extends BaseEvent {
+  type: "revalidation.decision";
+  segmentId: string;
+  pathname: string;
+  routeKey: string;
+  shouldRevalidate: boolean;
+}
+
+export interface RequestTimeoutEvent extends BaseEvent {
+  type: "request.timeout";
+  phase: import("./timeout.js").TimeoutPhase;
+  pathname: string;
+  routeKey?: string;
+  actionId?: string;
+  durationMs: number;
+  customHandler: boolean;
+  render?: import("./timeout.js").RenderTimeoutContext;
+}
+
+export interface OriginCheckRejectedEvent extends BaseEvent {
+  type: "request.origin-rejected";
+  method: string;
+  pathname: string;
+  phase: import("../rsc/origin-guard.js").OriginCheckPhase;
+  origin: string | null;
+  host: string | null;
+}
+
+export type TelemetryEvent =
+  | RequestStartEvent
+  | RequestEndEvent
+  | RequestErrorEvent
+  | LoaderStartEvent
+  | LoaderEndEvent
+  | LoaderErrorEvent
+  | HandlerErrorEvent
+  | CacheDecisionEvent
+  | RevalidationDecisionEvent
+  | RequestTimeoutEvent
+  | OriginCheckRejectedEvent;
+
+// ---------------------------------------------------------------------------
+// Cache signal derivation (coarse, route-level)
+// ---------------------------------------------------------------------------
+
+/**
+ * Derive the coarse, route-level cache status from pipeline cache state.
+ *
+ * v1 mapping (route-level — see CacheSegmentSignal):
+ *   - prerender hit                         -> "prerendered"
+ *   - runtime hit + shouldRevalidate (SWR)  -> "stale"
+ *   - runtime hit                           -> "hit"
+ *   - no hit                                -> "miss"
+ *
+ * Note: "passthrough" is a build-time prerender concept (a route opts out of
+ * being prerendered for some params). At runtime a passthrough route renders
+ * fresh and is indistinguishable from a normal miss in the pipeline state, so
+ * v1 reports it as "miss". The "passthrough" status remains in the type union
+ * for forward compatibility.
+ */
+export function deriveCacheStatus(state: {
+  cacheHit: boolean;
+  cacheSource?: "runtime" | "prerender";
+  shouldRevalidate?: boolean;
+}): CacheSegmentStatus {
+  if (state.cacheHit) {
+    if (state.cacheSource === "prerender") return "prerendered";
+    if (state.shouldRevalidate) return "stale";
+    return "hit";
+  }
+  return "miss";
+}
+
+/**
+ * Build the coarse route-level cache signal array (a single entry keyed by
+ * the route key). Used for both the cache.decision telemetry event and the
+ * X-Rango-Cache debug header.
+ */
+export function buildCacheSignalSegments(
+  routeKey: string,
+  state: {
+    cacheHit: boolean;
+    cacheSource?: "runtime" | "prerender";
+    shouldRevalidate?: boolean;
+  },
+): CacheSegmentSignal[] {
+  return [
+    {
+      id: routeKey,
+      type: "route",
+      cacheStatus: deriveCacheStatus(state),
+      shouldRevalidate: !!state.shouldRevalidate,
+    },
+  ];
+}
+
+/**
+ * Serialize cache signal segments into the X-Rango-Cache header value:
+ * `<segId>=<status>, <segId2>=<status2>`.
+ */
+export function formatCacheSignalHeader(
+  segments: CacheSegmentSignal[],
+): string {
+  return segments.map((s) => `${s.id}=${s.cacheStatus}`).join(", ");
+}
+
+/**
+ * Telemetry sink receives structured lifecycle events from the router.
+ * Implement this interface to integrate with any observability backend.
+ *
+ * All methods are fire-and-forget — exceptions are caught and logged.
+ */
+export interface TelemetrySink {
+  emit(event: TelemetryEvent): void;
+}
+
+const noopSink: TelemetrySink = {
+  emit() {},
+};
+
+/**
+ * Returns the configured sink, or the no-op singleton.
+ * Call sites use this so they don't need null checks.
+ */
+export function resolveSink(sink: TelemetrySink | undefined): TelemetrySink {
+  return sink ?? noopSink;
+}
+
+/**
+ * Safe emit — catches any error thrown by the sink to prevent
+ * telemetry failures from affecting request handling.
+ */
+export function safeEmit(sink: TelemetrySink, event: TelemetryEvent): void {
+  try {
+    sink.emit(event);
+  } catch (e) {
+    // Telemetry must never break request handling
+    if (process.env.NODE_ENV !== "production") {
+      console.error("[Router.telemetry] Sink error:", e);
+    }
+  }
+}
+
+const requestIds = new WeakMap<Request, string>();
+let telemetryRequestCounter = 0;
+
+/**
+ * Get or create a request ID for telemetry correlation.
+ * Checks standard headers first (x-rsc-router-request-id, x-request-id,
+ * cf-ray), then generates an internal ID when none is present.
+ * Generated IDs use format "t-{base36}" to distinguish from header values.
+ */
+export function getRequestId(request: Request): string {
+  const existing = requestIds.get(request);
+  if (existing) return existing;
+
+  const candidate =
+    request.headers.get("x-rsc-router-request-id") ??
+    request.headers.get("x-request-id") ??
+    request.headers.get("cf-ray");
+
+  let id: string;
+  if (candidate) {
+    const trimmed = candidate.trim();
+    id =
+      trimmed.length > 0
+        ? trimmed
+        : `t-${(++telemetryRequestCounter).toString(36)}`;
+  } else {
+    id = `t-${(++telemetryRequestCounter).toString(36)}`;
+  }
+
+  requestIds.set(request, id);
+  return id;
+}
+
+/**
+ * Built-in console sink that logs events in a structured format.
+ * Designed as the default sink for development / debugging.
+ */
+export function createConsoleSink(): TelemetrySink {
+  return {
+    emit(event: TelemetryEvent): void {
+      switch (event.type) {
+        case "request.start":
+          console.log(
+            `[telemetry] ${event.type} ${event.method} ${event.pathname} (${event.transaction})`,
+          );
+          break;
+        case "request.end":
+          console.log(
+            `[telemetry] ${event.type} ${event.method} ${event.pathname} ${event.durationMs.toFixed(1)}ms segments=${event.segmentCount} cache=${event.cacheHit}${event.status !== undefined ? ` status=${event.status}` : ""}`,
+          );
+          break;
+        case "request.error":
+          console.log(
+            `[telemetry] ${event.type} ${event.method} ${event.pathname} phase=${event.phase} ${event.durationMs.toFixed(1)}ms`,
+            event.error.message,
+          );
+          break;
+        case "loader.start":
+          console.log(
+            `[telemetry] ${event.type} ${event.loaderName} (${event.segmentId})`,
+          );
+          break;
+        case "loader.end":
+          console.log(
+            `[telemetry] ${event.type} ${event.loaderName} ${event.durationMs.toFixed(1)}ms ok=${event.ok}`,
+          );
+          break;
+        case "loader.error":
+          console.log(
+            `[telemetry] ${event.type} ${event.loaderName} boundary=${event.handledByBoundary}`,
+            event.error.message,
+          );
+          break;
+        case "handler.error":
+          console.log(
+            `[telemetry] ${event.type} segment=${event.segmentId ?? "unknown"} boundary=${event.handledByBoundary}${event.pathname ? ` ${event.pathname}` : ""}`,
+            event.error.message,
+          );
+          break;
+        case "cache.decision":
+          console.log(
+            `[telemetry] ${event.type} ${event.pathname} hit=${event.hit} swr=${event.shouldRevalidate}${event.source ? ` source=${event.source}` : ""}`,
+          );
+          break;
+        case "revalidation.decision":
+          console.log(
+            `[telemetry] ${event.type} ${event.segmentId} revalidate=${event.shouldRevalidate}`,
+          );
+          break;
+        case "request.timeout":
+          console.log(
+            `[telemetry] ${event.type} phase=${event.phase} ${event.pathname} ${event.durationMs.toFixed(1)}ms custom=${event.customHandler}`,
+          );
+          break;
+        case "request.origin-rejected":
+          console.log(
+            `[telemetry] ${event.type} ${event.method} ${event.pathname} phase=${event.phase} origin=${event.origin ?? "none"} host=${event.host ?? "none"}`,
+          );
+          break;
+      }
+    },
+  };
+}