Merge pull request #200 from pikasTech/fix/agentrun-event-otel-error-spans

fix: 将 runner 错误事件写入 OTel trace

src/mgr/server.ts

@@ -759,9 +759,17 @@ async function route({ method, url, body, store, sourceCommit, authSummary, runn
   }
   const eventsAppendMatch = path.match(/^\/api\/v1\/runs\/([^/]+)\/events$/u);
   if (method === "POST" && eventsAppendMatch) {
+    const startedAt = Date.now();
+    const runId = eventsAppendMatch[1] ?? "";
     const record = asRecord(body, "event");
     const type = typeof record.type === "string" ? record.type as RunEvent["type"] : "backend_status";
-    return await store.appendEvent(eventsAppendMatch[1] ?? "", type, asRecord(record.payload ?? {}, "event.payload")) as unknown as JsonValue;
+    const payload = asRecord(record.payload ?? {}, "event.payload");
+    const [run, event] = await Promise.all([
+      store.getRun(runId),
+      store.appendEvent(runId, type, payload),
+    ]);
+    emitRunEventOtelSpan(type, payload, run, startedAt);
+    return event as unknown as JsonValue;
   }
   const statusMatch = path.match(/^\/api\/v1\/runs\/([^/]+)\/status$/u);
   if (method === "PATCH" && statusMatch) {
@@ -962,6 +970,52 @@ function commandIsTerminal(command: CommandRecord): boolean {
   return command.state === "completed" || command.state === "failed" || command.state === "cancelled";
 }
 
+function emitRunEventOtelSpan(type: RunEvent["type"], payload: JsonRecord, run: RunRecord, startedAt: number): void {
+  const phase = stringJsonValue(payload.phase);
+  const terminalStatus = stringJsonValue(payload.terminalStatus);
+  const failureKind = stringJsonValue(payload.failureKind);
+  const eventName = runEventOtelSpanName(type, phase, terminalStatus, failureKind);
+  if (!eventName) return;
+  const isError = type === "error" || terminalStatus === "failed" || terminalStatus === "blocked";
+  void emitAgentRunOtelSpan(eventName, run, process.env, {
+    startTimeMs: startedAt,
+    kind: 2,
+    status: isError ? "error" : "ok",
+    error: isError ? boundedJsonString(payload.message, 300) ?? failureKind ?? eventName : undefined,
+    attributes: {
+      "http.method": "POST",
+      "http.route": "/api/v1/runs/:runId/events",
+      "http.status_code": 200,
+      eventType: type,
+      phase,
+      terminalStatus,
+      failureKind,
+      commandId: stringJsonValue(payload.commandId),
+      attemptId: stringJsonValue(payload.attemptId),
+      runnerId: stringJsonValue(payload.runnerId),
+      threadId: stringJsonValue(payload.threadId),
+      turnId: stringJsonValue(payload.turnId),
+      willRetry: typeof payload.willRetry === "boolean" ? payload.willRetry : null,
+      message: boundedJsonString(payload.message, 300),
+    },
+  });
+}
+
+function runEventOtelSpanName(type: RunEvent["type"], phase: string | null, terminalStatus: string | null, failureKind: string | null): string | null {
+  if (type === "error") return failureKind ? `runner_error.${otelNamePart(failureKind)}` : "runner_error";
+  if (type === "terminal_status") return terminalStatus ? `runner_terminal.${otelNamePart(terminalStatus)}` : "runner_terminal";
+  if (type !== "backend_status") return null;
+  if (!phase) return null;
+  if (phase === "cancel-requested" || phase === "turn-cancelled" || phase === "command-terminal") return `runner_${otelNamePart(phase)}`;
+  if (phase.startsWith("runner-claim-") || phase.startsWith("turn/interrupt:")) return `runner_${otelNamePart(phase)}`;
+  return null;
+}
+
+function otelNamePart(value: string): string {
+  const normalized = value.toLowerCase().replace(/[^a-z0-9]+/gu, "_").replace(/^_+|_+$/gu, "");
+  return normalized.length > 0 ? normalized : "unknown";
+}
+
 function optionalString(value: JsonValue | undefined): string | null {
   return typeof value === "string" && value.trim().length > 0 ? value.trim() : null;
 }