@autter/runtime-node 1.3.2 → 1.3.4

package/README.md

@@ -1,5 +1,11 @@
 # @autter/runtime-node
 
+The server tracker exports per-instance RSS, heap, memory limit, and GC
+metrics by default. See [memory pressure incidents](../../docs/MEMORY-PRESSURE.md)
+for ECS/Kubernetes OOM event forwarding and optional heap profiles. Set
+`memoryMetrics: false` only if another exporter already emits equivalent
+process metrics.
+
 Autter Runtime for Node.js — two halves in one package:
 
 1. **Same-origin browser relay** for `@autter/runtime-browser`

package/dist/index.cjs

@@ -83,6 +83,7 @@ var EVENT_TYPES = /* @__PURE__ */ new Set([
   "track_event",
   "outcome",
   "request_failure",
+  "csp_violation",
   "timing"
 ]);
 var SEVERITIES = /* @__PURE__ */ new Set(["fatal", "error", "warning", "info"]);
@@ -255,7 +256,7 @@ function createBrowserRelayHandler(opts) {
 // src/server.ts
 var import_node_http = require("http");
 var import_node_crypto = require("crypto");
-var import_api = require("@opentelemetry/api");
+var import_api2 = require("@opentelemetry/api");
 var import_core2 = require("@opentelemetry/core");
 var import_exporter_metrics_otlp_http = require("@opentelemetry/exporter-metrics-otlp-http");
 var import_exporter_trace_otlp_http = require("@opentelemetry/exporter-trace-otlp-http");
@@ -707,6 +708,51 @@ function makeRedactor(options) {
   return (attributes) => redactWith(attributes, r);
 }
 
+// src/memory.ts
+var import_node_perf_hooks = require("perf_hooks");
+var import_node_v8 = require("v8");
+var import_api = require("@opentelemetry/api");
+function startMemoryMetrics() {
+  const meter = import_api.metrics.getMeter("autter-process-memory");
+  const rss = meter.createObservableGauge("autter.process.memory.rss", { unit: "By" });
+  const heapUsed = meter.createObservableGauge("autter.process.memory.heap.used", { unit: "By" });
+  const heapLimit = meter.createObservableGauge("autter.process.memory.heap.limit", { unit: "By" });
+  const containerLimit = meter.createObservableGauge("autter.process.memory.limit", { unit: "By" });
+  const afterGc = meter.createObservableGauge("autter.process.memory.heap.after_gc", { unit: "By" });
+  const gcCount = meter.createObservableCounter("autter.process.gc.count", { unit: "{collection}" });
+  const gcDuration = meter.createObservableCounter("autter.process.gc.duration", { unit: "ms" });
+  let collections = 0;
+  let durationMs = 0;
+  let lastAfterGc = null;
+  const callbacks = [
+    [rss, (result) => result.observe(process.memoryUsage().rss)],
+    [heapUsed, (result) => result.observe(process.memoryUsage().heapUsed)],
+    [heapLimit, (result) => result.observe((0, import_node_v8.getHeapStatistics)().heap_size_limit)],
+    [containerLimit, (result) => {
+      const constrained = process.constrainedMemory?.() ?? 0;
+      if (constrained > 0) result.observe(constrained);
+    }],
+    [afterGc, (result) => {
+      if (lastAfterGc !== null) result.observe(lastAfterGc);
+    }],
+    [gcCount, (result) => result.observe(collections)],
+    [gcDuration, (result) => result.observe(durationMs)]
+  ];
+  for (const [instrument, callback] of callbacks) instrument.addCallback(callback);
+  const observer = new import_node_perf_hooks.PerformanceObserver((list) => {
+    for (const entry of list.getEntries()) {
+      collections++;
+      durationMs += entry.duration;
+      lastAfterGc = process.memoryUsage().heapUsed;
+    }
+  });
+  observer.observe({ entryTypes: ["gc"] });
+  return () => {
+    observer.disconnect();
+    for (const [instrument, callback] of callbacks) instrument.removeCallback(callback);
+  };
+}
+
 // src/server.ts
 var LLM_MARKER_ATTRIBUTES = [
   "gen_ai.operation.name",
@@ -769,14 +815,14 @@ var RETENTION_MAX_BUFFERED_SPANS = 5e3;
 var RETENTION_SWEEP_INTERVAL_MS = 1e4;
 var RETENTION_TRACE_TTL_MS = 3e4;
 function spanIndicatesError(span) {
-  if (span.status.code === import_api.SpanStatusCode.ERROR) return true;
+  if (span.status.code === import_api2.SpanStatusCode.ERROR) return true;
   return span.events.some((event) => event.name === "exception");
 }
 var TAIL_RETAINED_ATTR = "autter.tail_retained";
 function withSampledFlag(span) {
   const spanContext = {
     ...span.spanContext(),
-    traceFlags: span.spanContext().traceFlags | import_api.TraceFlags.SAMPLED
+    traceFlags: span.spanContext().traceFlags | import_api2.TraceFlags.SAMPLED
   };
   return {
     name: span.name,
@@ -816,11 +862,11 @@ var ErrorTraceRetentionProcessor = class {
   }
   onEnd(span) {
     const ctx = span.spanContext();
-    if ((ctx.traceFlags & import_api.TraceFlags.SAMPLED) !== 0) return;
+    if ((ctx.traceFlags & import_api2.TraceFlags.SAMPLED) !== 0) return;
     const entry = this.entryFor(ctx.traceId);
     if (entry.retained) {
       this.forward(span);
-    } else if (this.retainErrors && spanIndicatesError(span) || this.slowThresholdMs > 0 && span.kind === import_api.SpanKind.SERVER && span.duration[0] * 1e3 + span.duration[1] / 1e6 >= this.slowThresholdMs) {
+    } else if (this.retainErrors && spanIndicatesError(span) || this.slowThresholdMs > 0 && span.kind === import_api2.SpanKind.SERVER && span.duration[0] * 1e3 + span.duration[1] / 1e6 >= this.slowThresholdMs) {
       entry.retained = true;
       this.flush(entry);
       this.forward(span);
@@ -835,8 +881,8 @@ var ErrorTraceRetentionProcessor = class {
   }
   /** Promote the active trace, if any — called on captured errors. */
   retainActiveTrace() {
-    const active2 = import_api.trace.getActiveSpan()?.spanContext();
-    if (!active2 || (active2.traceFlags & import_api.TraceFlags.SAMPLED) !== 0) return;
+    const active2 = import_api2.trace.getActiveSpan()?.spanContext();
+    if (!active2 || (active2.traceFlags & import_api2.TraceFlags.SAMPLED) !== 0) return;
     const entry = this.entryFor(active2.traceId);
     if (entry.retained) return;
     entry.retained = true;
@@ -925,7 +971,7 @@ function usageAttributes(usage) {
 }
 async function runLlmSpan(tracer, info, fn) {
   const span = tracer.startSpan(`${info.operation ?? "chat"} ${info.model}`, {
-    kind: import_api.SpanKind.CLIENT,
+    kind: import_api2.SpanKind.CLIENT,
     attributes: llmBaseAttributes(info)
   });
   const setResponseModel = (model) => span.setAttribute("gen_ai.response.model", model);
@@ -937,11 +983,11 @@ async function runLlmSpan(tracer, info, fn) {
     setAttributes: (attributes) => span.setAttributes(attributes)
   };
   try {
-    const result = await import_api.context.with(
-      import_api.trace.setSpan(import_api.context.active(), span),
+    const result = await import_api2.context.with(
+      import_api2.trace.setSpan(import_api2.context.active(), span),
       () => fn(handle)
     );
-    span.setStatus({ code: import_api.SpanStatusCode.OK });
+    span.setStatus({ code: import_api2.SpanStatusCode.OK });
     return result;
   } catch (err) {
     if (err instanceof Error) {
@@ -949,7 +995,7 @@ async function runLlmSpan(tracer, info, fn) {
       span.setAttribute("error.type", err.name);
     }
     span.setStatus({
-      code: import_api.SpanStatusCode.ERROR,
+      code: import_api2.SpanStatusCode.ERROR,
       message: err instanceof Error ? err.message : String(err)
     });
     throw err;
@@ -964,7 +1010,7 @@ function recordLlmCall(tracer, call) {
   const span = tracer.startSpan(
     `${call.operation ?? "chat"} ${call.model}`,
     {
-      kind: import_api.SpanKind.CLIENT,
+      kind: import_api2.SpanKind.CLIENT,
       startTime,
       attributes: { ...llmBaseAttributes(call), ...usageAttributes(call) }
     }
@@ -975,27 +1021,27 @@ function recordLlmCall(tracer, call) {
       span.setAttribute("error.type", call.error.name);
     }
     span.setStatus({
-      code: import_api.SpanStatusCode.ERROR,
+      code: import_api2.SpanStatusCode.ERROR,
       message: call.error instanceof Error ? call.error.message : String(call.error)
     });
   } else {
-    span.setStatus({ code: import_api.SpanStatusCode.OK });
+    span.setStatus({ code: import_api2.SpanStatusCode.OK });
   }
   span.end(endTime);
 }
 async function runWithSpan(tracer, name, fn, attributes) {
   const span = tracer.startSpan(name, { attributes });
   try {
-    const result = await import_api.context.with(
-      import_api.trace.setSpan(import_api.context.active(), span),
+    const result = await import_api2.context.with(
+      import_api2.trace.setSpan(import_api2.context.active(), span),
       fn
     );
-    span.setStatus({ code: import_api.SpanStatusCode.OK });
+    span.setStatus({ code: import_api2.SpanStatusCode.OK });
     return result;
   } catch (err) {
     if (err instanceof Error) span.recordException(err);
     span.setStatus({
-      code: import_api.SpanStatusCode.ERROR,
+      code: import_api2.SpanStatusCode.ERROR,
       message: err instanceof Error ? err.message : String(err)
     });
     throw err;
@@ -1013,7 +1059,7 @@ function expressRouteOf(req) {
 }
 function captureExpressRoute(_span, response) {
   if (!(response instanceof import_node_http.ServerResponse)) return;
-  const rpcMetadata = (0, import_core2.getRPCMetadata)(import_api.context.active());
+  const rpcMetadata = (0, import_core2.getRPCMetadata)(import_api2.context.active());
   if (rpcMetadata?.type !== import_core2.RPCType.HTTP) return;
   const setRoute = () => {
     if (rpcMetadata.route) return;
@@ -1039,7 +1085,7 @@ function initAutterServer(options) {
   activeRedactor = makeRedactor(options.redactAttributes ?? true);
   const resource = new import_resources.Resource({
     [import_semantic_conventions.ATTR_SERVICE_NAME]: options.service,
-    "service.instance.id": (0, import_node_crypto.randomUUID)(),
+    "service.instance.id": (options.instanceId || process.env.AUTTER_RUNTIME_INSTANCE_ID || (0, import_node_crypto.randomUUID)()).slice(0, 128),
     ...options.release ? { [import_semantic_conventions.ATTR_SERVICE_VERSION]: options.release } : {},
     "deployment.environment": environment,
     // Tells the ingester request metrics arrive on the metrics pipe, so
@@ -1104,6 +1150,7 @@ function initAutterServer(options) {
     ]
   });
   sdk.start();
+  const stopMemoryMetrics = options.memoryMetrics === false ? null : startMemoryMetrics();
   const alwaysOnProvider = new import_sdk_trace_base.BasicTracerProvider({
     resource,
     sampler: new import_sdk_trace_base.AlwaysOnSampler(),
@@ -1125,7 +1172,7 @@ function initAutterServer(options) {
     const isError = error instanceof Error;
     const message = isError ? error.message : String(error);
     const redacted = activeRedactor(attributes);
-    const activeSpan = import_api.trace.getActiveSpan();
+    const activeSpan = import_api2.trace.getActiveSpan();
     const reuseActive = activeSpan?.isRecording() === true;
     const span = reuseActive ? activeSpan : errorTracer.startSpan(isError ? error.name : "Error");
     span.setAttributes({ "autter.severity": "error", ...redacted });
@@ -1150,7 +1197,7 @@ function initAutterServer(options) {
         ...redacted["autter.handled"] === true ? { "autter.handled": true, "autter.sampled": redacted["autter.sampled"] === true } : {}
       });
     }
-    span.setStatus({ code: import_api.SpanStatusCode.ERROR, message });
+    span.setStatus({ code: import_api2.SpanStatusCode.ERROR, message });
     if (!reuseActive) span.end();
   }
   function captureMessage2(message, severity = "warning", attributes) {
@@ -1168,13 +1215,13 @@ function initAutterServer(options) {
       ...stack ? { "exception.stacktrace": stack } : {},
       "autter.severity": severity
     });
-    span.setStatus({ code: import_api.SpanStatusCode.ERROR, message });
+    span.setStatus({ code: import_api2.SpanStatusCode.ERROR, message });
     span.end();
   }
   function reportOutcome2(name, message, attributes) {
     errorTraceBuffer?.retainActiveTrace();
     telemetryStats.markCaptured();
-    const activeSpan = import_api.trace.getActiveSpan();
+    const activeSpan = import_api2.trace.getActiveSpan();
     const reuseActive = activeSpan?.isRecording() === true;
     const span = reuseActive ? activeSpan : errorTracer.startSpan("OutcomeFailure");
     const safeOutcome = activeRedactor({ "autter.outcome.name": name.slice(0, 200), "autter.outcome.message": message.slice(0, 1e3) });
@@ -1186,7 +1233,7 @@ function initAutterServer(options) {
       "autter.outcome.name": safeName,
       "autter.outcome.message": safeMessage
     });
-    span.setStatus({ code: import_api.SpanStatusCode.ERROR, message: safeMessage });
+    span.setStatus({ code: import_api2.SpanStatusCode.ERROR, message: safeMessage });
     if (!reuseActive) span.end();
   }
   if (options.captureGlobalErrors !== false) {
@@ -1225,6 +1272,7 @@ function initAutterServer(options) {
     withLlmCall: (info, fn) => runLlmSpan(llmTracer, info, fn),
     trackLlmCall: (call) => recordLlmCall(llmTracer, call),
     shutdown: async () => {
+      stopMemoryMetrics?.();
       active = null;
       activeAlwaysOnProvider = null;
       autoFlushHandle?.dispose();
@@ -1243,35 +1291,35 @@ function captureException(error, attributes) {
     active.captureException(error, attributes);
     return;
   }
-  const span = import_api.trace.getTracer("autter-errors").startSpan(error instanceof Error ? error.name : "Error", { attributes });
+  const span = import_api2.trace.getTracer("autter-errors").startSpan(error instanceof Error ? error.name : "Error", { attributes });
   if (error instanceof Error) span.recordException(error);
   span.setStatus({
-    code: import_api.SpanStatusCode.ERROR,
+    code: import_api2.SpanStatusCode.ERROR,
     message: error instanceof Error ? error.message : String(error)
   });
   span.end();
 }
 function withProcessSpan(name, fn, attributes) {
   if (active) return active.withProcessSpan(name, fn, attributes);
-  return runWithSpan(import_api.trace.getTracer("autter-processes"), name, fn, attributes);
+  return runWithSpan(import_api2.trace.getTracer("autter-processes"), name, fn, attributes);
 }
 function withLlmCall(info, fn) {
   if (active) return active.withLlmCall(info, fn);
-  return runLlmSpan(import_api.trace.getTracer("autter-llm"), info, fn);
+  return runLlmSpan(import_api2.trace.getTracer("autter-llm"), info, fn);
 }
 function trackLlmCall(call) {
   if (active) {
     active.trackLlmCall(call);
     return;
   }
-  recordLlmCall(import_api.trace.getTracer("autter-llm"), call);
+  recordLlmCall(import_api2.trace.getTracer("autter-llm"), call);
 }
 function captureMessage(message, severity = "warning", attributes) {
   if (active) {
     active.captureMessage(message, severity, attributes);
     return;
   }
-  const span = import_api.trace.getTracer("autter-errors").startSpan("Message", {
+  const span = import_api2.trace.getTracer("autter-errors").startSpan("Message", {
     attributes: { "autter.severity": severity, ...attributes }
   });
   span.addEvent("exception", {
@@ -1279,12 +1327,12 @@ function captureMessage(message, severity = "warning", attributes) {
     "exception.message": message,
     "autter.severity": severity
   });
-  span.setStatus({ code: import_api.SpanStatusCode.ERROR, message });
+  span.setStatus({ code: import_api2.SpanStatusCode.ERROR, message });
   span.end();
 }
 function reportOutcome(name, message, attributes) {
   if (active) return active.reportOutcome(name, message, attributes);
-  const span = import_api.trace.getTracer("autter-outcomes").startSpan("OutcomeFailure");
+  const span = import_api2.trace.getTracer("autter-outcomes").startSpan("OutcomeFailure");
   const safeOutcome = redactAttributes({ "autter.outcome.name": name.slice(0, 200), "autter.outcome.message": message.slice(0, 1e3) });
   const safeName = String(safeOutcome["autter.outcome.name"] ?? "operation");
   const safeMessage = String(safeOutcome["autter.outcome.message"] ?? "failed");
@@ -1294,7 +1342,7 @@ function reportOutcome(name, message, attributes) {
     "autter.outcome.name": safeName,
     "autter.outcome.message": safeMessage
   });
-  span.setStatus({ code: import_api.SpanStatusCode.ERROR, message: safeMessage });
+  span.setStatus({ code: import_api2.SpanStatusCode.ERROR, message: safeMessage });
   span.end();
 }
 function makeSafeCapture(options) {
@@ -1305,7 +1353,7 @@ function makeSafeCapture(options) {
   };
 }
 function alwaysOnTracer(name) {
-  return activeAlwaysOnProvider ? activeAlwaysOnProvider.getTracer(name) : import_api.trace.getTracer(name);
+  return activeAlwaysOnProvider ? activeAlwaysOnProvider.getTracer(name) : import_api2.trace.getTracer(name);
 }
 function autterLlmTracer() {
   return alwaysOnTracer("autter-llm");
@@ -1325,7 +1373,7 @@ async function emitLlmSelftestTrace() {
       attributes: { "autter.selftest": true }
     },
     (llm) => {
-      traceId = import_api.trace.getActiveSpan()?.spanContext().traceId ?? "";
+      traceId = import_api2.trace.getActiveSpan()?.spanContext().traceId ?? "";
       llm.setUsage({ inputTokens: 1, outputTokens: 1 });
       llm.setCost(0);
     }
@@ -1335,7 +1383,7 @@ async function emitLlmSelftestTrace() {
 }
 
 // src/llm-instrument.ts
-var import_api2 = require("@opentelemetry/api");
+var import_api3 = require("@opentelemetry/api");
 var CALL_METHODS = /* @__PURE__ */ new Set([
   "create",
   // openai chat/completions/embeddings/responses, anthropic messages
@@ -1393,7 +1441,7 @@ function applyUsage(span, usage) {
 function endWithError(span, err) {
   if (err instanceof Error) span.recordException(err);
   span.setStatus({
-    code: import_api2.SpanStatusCode.ERROR,
+    code: import_api3.SpanStatusCode.ERROR,
     message: err instanceof Error ? err.message : String(err)
   });
   span.end();
@@ -1413,7 +1461,7 @@ function wrapStream(stream, span) {
     if (err !== void 0) {
       endWithError(span, err);
     } else {
-      span.setStatus({ code: import_api2.SpanStatusCode.OK });
+      span.setStatus({ code: import_api3.SpanStatusCode.OK });
       span.end();
     }
   };
@@ -1468,7 +1516,7 @@ function instrumentLlmClient(client, options) {
     const operation = /embed/i.test(callPath) ? "embeddings" : "chat";
     const userId = resolveUserId();
     const span = autterLlmTracer().startSpan(`${operation} ${request.model}`, {
-      kind: import_api2.SpanKind.CLIENT,
+      kind: import_api3.SpanKind.CLIENT,
       attributes: {
         "gen_ai.operation.name": operation,
         "gen_ai.provider.name": provider,
@@ -1490,7 +1538,7 @@ function instrumentLlmClient(client, options) {
         return wrapStream(result, span);
       }
       applyUsage(span, extractUsage(result));
-      span.setStatus({ code: import_api2.SpanStatusCode.OK });
+      span.setStatus({ code: import_api3.SpanStatusCode.OK });
       span.end();
       return result;
     }
@@ -1500,7 +1548,7 @@ function instrumentLlmClient(client, options) {
           return wrapStream(resolved, span);
         }
         applyUsage(span, extractUsage(resolved));
-        span.setStatus({ code: import_api2.SpanStatusCode.OK });
+        span.setStatus({ code: import_api3.SpanStatusCode.OK });
         span.end();
         return resolved;
       },

package/dist/index.d.cts

@@ -100,6 +100,8 @@ interface AutterServerOptions {
     /** Ingester base URL. Default: https://otlp.autter.dev */
     endpoint?: string;
     service: string;
+    /** Stable for this process lifetime; use an ECS task/container ID or pod UID plus restart count to correlate platform events. */
+    instanceId?: string;
     environment?: string;
     /** e.g. a git SHA. Maps to service.version / release. */
     release?: string;
@@ -119,6 +121,8 @@ interface AutterServerOptions {
     retainTracesAboveMs?: number;
     /** Metric export interval. Default 60_000 ms. */
     metricIntervalMs?: number;
+    /** Collect RSS, heap, limit, and GC signals per process instance. Default true. */
+    memoryMetrics?: boolean;
     /** Capture crashing exceptions via process.uncaughtExceptionMonitor (default true). */
     captureGlobalErrors?: boolean;
     /**

package/dist/index.d.ts

@@ -100,6 +100,8 @@ interface AutterServerOptions {
     /** Ingester base URL. Default: https://otlp.autter.dev */
     endpoint?: string;
     service: string;
+    /** Stable for this process lifetime; use an ECS task/container ID or pod UID plus restart count to correlate platform events. */
+    instanceId?: string;
     environment?: string;
     /** e.g. a git SHA. Maps to service.version / release. */
     release?: string;
@@ -119,6 +121,8 @@ interface AutterServerOptions {
     retainTracesAboveMs?: number;
     /** Metric export interval. Default 60_000 ms. */
     metricIntervalMs?: number;
+    /** Collect RSS, heap, limit, and GC signals per process instance. Default true. */
+    memoryMetrics?: boolean;
     /** Capture crashing exceptions via process.uncaughtExceptionMonitor (default true). */
     captureGlobalErrors?: boolean;
     /**

package/dist/index.js

@@ -32,6 +32,7 @@ var EVENT_TYPES = /* @__PURE__ */ new Set([
   "track_event",
   "outcome",
   "request_failure",
+  "csp_violation",
   "timing"
 ]);
 var SEVERITIES = /* @__PURE__ */ new Set(["fatal", "error", "warning", "info"]);
@@ -675,6 +676,51 @@ function makeRedactor(options) {
   return (attributes) => redactWith(attributes, r);
 }
 
+// src/memory.ts
+import { PerformanceObserver } from "perf_hooks";
+import { getHeapStatistics } from "v8";
+import { metrics } from "@opentelemetry/api";
+function startMemoryMetrics() {
+  const meter = metrics.getMeter("autter-process-memory");
+  const rss = meter.createObservableGauge("autter.process.memory.rss", { unit: "By" });
+  const heapUsed = meter.createObservableGauge("autter.process.memory.heap.used", { unit: "By" });
+  const heapLimit = meter.createObservableGauge("autter.process.memory.heap.limit", { unit: "By" });
+  const containerLimit = meter.createObservableGauge("autter.process.memory.limit", { unit: "By" });
+  const afterGc = meter.createObservableGauge("autter.process.memory.heap.after_gc", { unit: "By" });
+  const gcCount = meter.createObservableCounter("autter.process.gc.count", { unit: "{collection}" });
+  const gcDuration = meter.createObservableCounter("autter.process.gc.duration", { unit: "ms" });
+  let collections = 0;
+  let durationMs = 0;
+  let lastAfterGc = null;
+  const callbacks = [
+    [rss, (result) => result.observe(process.memoryUsage().rss)],
+    [heapUsed, (result) => result.observe(process.memoryUsage().heapUsed)],
+    [heapLimit, (result) => result.observe(getHeapStatistics().heap_size_limit)],
+    [containerLimit, (result) => {
+      const constrained = process.constrainedMemory?.() ?? 0;
+      if (constrained > 0) result.observe(constrained);
+    }],
+    [afterGc, (result) => {
+      if (lastAfterGc !== null) result.observe(lastAfterGc);
+    }],
+    [gcCount, (result) => result.observe(collections)],
+    [gcDuration, (result) => result.observe(durationMs)]
+  ];
+  for (const [instrument, callback] of callbacks) instrument.addCallback(callback);
+  const observer = new PerformanceObserver((list) => {
+    for (const entry of list.getEntries()) {
+      collections++;
+      durationMs += entry.duration;
+      lastAfterGc = process.memoryUsage().heapUsed;
+    }
+  });
+  observer.observe({ entryTypes: ["gc"] });
+  return () => {
+    observer.disconnect();
+    for (const [instrument, callback] of callbacks) instrument.removeCallback(callback);
+  };
+}
+
 // src/server.ts
 var LLM_MARKER_ATTRIBUTES = [
   "gen_ai.operation.name",
@@ -1007,7 +1053,7 @@ function initAutterServer(options) {
   activeRedactor = makeRedactor(options.redactAttributes ?? true);
   const resource = new Resource({
     [ATTR_SERVICE_NAME]: options.service,
-    "service.instance.id": randomUUID(),
+    "service.instance.id": (options.instanceId || process.env.AUTTER_RUNTIME_INSTANCE_ID || randomUUID()).slice(0, 128),
     ...options.release ? { [ATTR_SERVICE_VERSION]: options.release } : {},
     "deployment.environment": environment,
     // Tells the ingester request metrics arrive on the metrics pipe, so
@@ -1072,6 +1118,7 @@ function initAutterServer(options) {
     ]
   });
   sdk.start();
+  const stopMemoryMetrics = options.memoryMetrics === false ? null : startMemoryMetrics();
   const alwaysOnProvider = new BasicTracerProvider({
     resource,
     sampler: new AlwaysOnSampler(),
@@ -1193,6 +1240,7 @@ function initAutterServer(options) {
     withLlmCall: (info, fn) => runLlmSpan(llmTracer, info, fn),
     trackLlmCall: (call) => recordLlmCall(llmTracer, call),
     shutdown: async () => {
+      stopMemoryMetrics?.();
       active = null;
       activeAlwaysOnProvider = null;
       autoFlushHandle?.dispose();

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@autter/runtime-node",
-	"version": "1.3.2",
+	"version": "1.3.4",
 	"description": "Autter Runtime for Node.js: same-origin browser relay handler + curated OpenTelemetry server tracker",
 	"license": "MIT",
 	"type": "module",