sentry-performance-tracing（哨兵性能跟踪）

Sentry Performance Tracing

Overview

Sentry performance monitoring captures distributed traces across your application stack, measuring latency, identifying bottlenecks, and tracking Web Vitals. The v8 SDK uses a span-based API where Sentry.startSpan() replaces the deprecated startTransaction(). Auto-instrumentation covers HTTP, database queries, and framework routes out of the box. Manual spans let you measure business-critical operations. Combined with profiling (profilesSampleRate), you get function-level flamegraphs attached to traces.

Prerequisites

• Sentry SDK v8+ installed (@sentry/node >= 8.0.0 or sentry-sdk >= 2.0.0)
• tracesSampleRate > 0 set in Sentry.init() — performance data is not collected at zero
• Performance monitoring enabled in your Sentry project settings (Settings > Performance)
• For distributed tracing: all participating services must have Sentry SDK initialized

Instructions

Step 1 — Configure Tracing and Profiling in SDK Init

Set tracesSampleRate to control what percentage of requests generate traces. Use tracesSampler for dynamic, per-endpoint sampling. Add profilesSampleRate to attach function-level flamegraphs to sampled transactions.

TypeScript (@sentry/node):

import * as Sentry from '@sentry/node';

Sentry.init({
  dsn: process.env.SENTRY_DSN,
  tracesSampleRate: 0.2, // 20% of transactions in production

  // Profiling — profiles 10% of sampled transactions
  profilesSampleRate: 0.1,

  // Dynamic sampling overrides tracesSampleRate when defined
  tracesSampler: (samplingContext) => {
    const { name, attributes } = samplingContext;

    // Drop health checks entirely — no trace data
    if (name === 'GET /health') return 0;

    // Always trace payment flows
    if (name?.includes('/api/payment')) return 1.0;

    // Higher sampling for API routes
    if (name?.startsWith('GET /api/') || name?.startsWith('POST /api/')) return 0.2;

    // Default: 5% for everything else
    return 0.05;
  },
});

Python (sentry-sdk):

import os
import sentry_sdk

sentry_sdk.init(
    dsn=os.environ["SENTRY_DSN"],
    traces_sample_rate=0.2,       # 20% of transactions
    profiles_sample_rate=0.1,     # 10% of sampled transactions get profiled

    # Dynamic sampling via traces_sampler (overrides traces_sample_rate)
    traces_sampler=lambda ctx: (
        0.0 if ctx.get("transaction_context", {}).get("name") == "GET /health"
        else 1.0 if "/api/payment" in ctx.get("transaction_context", {}).get("name", "")
        else 0.2
    ),
)

Key decisions:

• Start at tracesSampleRate: 0.2 and adjust based on volume and budget
• tracesSampler takes priority when defined — tracesSampleRate becomes the fallback
• profilesSampleRate is relative to sampled transactions (0.1 means 10% of the 20% that are sampled)
• Return 0 from tracesSampler to explicitly drop a transaction, not false

Step 2 — Create Custom Spans for Business Logic

Auto-instrumentation covers HTTP and database calls, but business-critical operations need manual spans. The v8 API provides three span creation methods for different use cases.

Sentry.startSpan() — auto-ending spans (most common):

import * as Sentry from '@sentry/node';

const result = await Sentry.startSpan(
  {
    name: 'order.process',
    op: 'task',
    attributes: {
      'order.id': orderId,
      'order.items': items.length,
    },
  },
  async (span) => {
    // Nested spans automatically become children of the parent
    const validated = await Sentry.startSpan(
      { name: 'order.validate', op: 'validation' },
      async () => validateOrder(order)
    );

    const charged = await Sentry.startSpan(
      { name: 'payment.charge', op: 'http.client' },
      async () => chargePayment(order.total)
    );

    // Set span status based on outcome
    if (!charged.success) {
      span.setStatus({ code: 2, message: 'payment_failed' });
    }

    // Add custom measurements visible in Performance dashboard
    Sentry.setMeasurement('order.item_count', items.length, 'none');
    Sentry.setMeasurement('order.total_cents', order.total, 'none');

    return { validated, charged };
  }
);
// Span automatically ends when callback resolves or rejects

Sentry.startSpanManual() — for spans that cross callback boundaries:

Sentry.startSpanManual(
  { name: 'queue.process', op: 'queue.task' },
  (span) => {
    queue.on('message', async (msg) => {
      try {
        await processMessage(msg);
        span.setStatus({ code: 1 }); // OK
      } catch (error) {
        span.setStatus({ code: 2, message: 'processing_failed' });
        Sentry.captureException(error);
      } finally {
        span.end(); // REQUIRED — must call end() manually
      }
    });
  }
);

Sentry.startInactiveSpan() — background work without changing active context:

const span = Sentry.startInactiveSpan({
  name: 'cache.warmup',
  op: 'cache',
});

await warmCache(); // Other spans created here won't be children of this span

span.end();

Span attributes and measurements:

await Sentry.startSpan(
  { name: 'search.query', op: 'db.query' },
  async (span) => {
    const start = Date.now();
    const results = await searchIndex(query);

    // Attributes — appear in span details, filterable in Sentry UI
    span.setAttribute('search.query', query);
    span.setAttribute('search.results_count', results.length);
    span.setAttribute('search.index', indexName);

    // Measurements — appear in Performance dashboard charts
    Sentry.setMeasurement('search.duration_ms', Date.now() - start, 'millisecond');
    Sentry.setMeasurement('search.result_count', results.length, 'none');

    return results;
  }
);

Python equivalent:

import sentry_sdk

with sentry_sdk.start_span(op="task", name="process_order") as span:
    span.set_data("order_id", order_id)
    span.set_data("item_count", len(items))

    with sentry_sdk.start_span(op="validation", name="validate_input"):
        validate(input_data)

    with sentry_sdk.start_span(op="http.client", name="charge_payment"):
        result = charge(payment)

    if not result.success:
        span.set_status("internal_error")

Step 3 — Enable Auto-Instrumentation and Distributed Tracing

SDK v8 auto-instruments most I/O without configuration. For distributed tracing across services, Sentry propagates sentry-trace and baggage headers automatically on HTTP calls. Custom propagation is needed only for non-HTTP transports (message queues, gRPC, etc.).

Auto-instrumented integrations (Node.js v8):

Express with custom middleware spans:

import express from 'express';
import * as Sentry from '@sentry/node';

const app = express();

// Sentry auto-instruments all Express routes
// Add custom spans for specific middleware:
app.use('/api', async (req, res, next) => {
  await Sentry.startSpan(
    { name: 'middleware.auth', op: 'middleware' },
    async () => {
      req.user = await authenticateRequest(req);
    }
  );
  next();
});

// Parameterized route names prevent cardinality explosion
// Sentry automatically uses '/api/users/:id' not '/api/users/12345'
app.get('/api/users/:id', async (req, res) => {
  const user = await Sentry.startSpan(
    { name: 'db.getUser', op: 'db.query' },
    () => db.users.findById(req.params.id)
  );
  res.json(user);
});

// Must be after all routes
Sentry.setupExpressErrorHandler(app);

Django/Flask auto-instrumentation (Python):

import sentry_sdk
from sentry_sdk.integrations.django import DjangoIntegration

sentry_sdk.init(
    dsn=os.environ["SENTRY_DSN"],
    integrations=[DjangoIntegration()],
    traces_sample_rate=0.2,
    profiles_sample_rate=0.1,
)
# All Django views, middleware, and template rendering are traced automatically

# Flask equivalent
from sentry_sdk.integrations.flask import FlaskIntegration

sentry_sdk.init(
    dsn=os.environ["SENTRY_DSN"],
    integrations=[FlaskIntegration()],
    traces_sample_rate=0.2,
)

# FastAPI equivalent
from sentry_sdk.integrations.fastapi import FastApiIntegration
from sentry_sdk.integrations.starlette import StarletteIntegration

sentry_sdk.init(
    dsn=os.environ["SENTRY_DSN"],
    integrations=[FastApiIntegration(), StarletteIntegration()],
    traces_sample_rate=0.2,
)

Distributed tracing — custom header propagation:

When Sentry cannot automatically propagate headers (non-HTTP transports, custom fetch wrappers), extract and inject manually:

// Service A: Extract trace headers from the active span
const activeSpan = Sentry.getActiveSpan();
const traceHeaders = {
  'sentry-trace': Sentry.spanToTraceHeader(activeSpan),
  'baggage': Sentry.spanToBaggageHeader(activeSpan),
};

// Pass headers to downstream service via HTTP, message queue, etc.
await fetch('https://service-b.internal/api/process', {
  headers: { ...traceHeaders, 'Content-Type': 'application/json' },
  body: JSON.stringify(payload),
});

// Service B: Sentry SDK automatically reads sentry-trace and baggage
// from incoming request headers and continues the same trace

Browser Web Vitals (@sentry/browser):

The browser SDK automatically captures Core Web Vitals when tracing is enabled:

• LCP (Largest Contentful Paint) — loading performance
• INP (Interaction to Next Paint) — responsiveness (replaced FID in 2024)
• CLS (Cumulative Layout Shift) — visual stability
• TTFB (Time to First Byte) — server response time

These appear in the Web Vitals tab of your Sentry Performance dashboard. No additional configuration beyond tracesSampleRate > 0 in the browser SDK.

Output

• Distributed traces visible in Sentry Performance > Trace View as span waterfalls
• Auto-instrumented spans for HTTP, database, and framework operations
• Custom spans with attributes measuring business-critical operations
• Profiling flamegraphs attached to sampled transactions
• Web Vitals (LCP, INP, CLS, TTFB) tracked for frontend performance
• Custom measurements charted in Performance dashboard
• Cross-service traces linked via sentry-trace and baggage headers

Error Handling

Examples

TypeScript — Full Express API with Profiling

import * as Sentry from '@sentry/node';
import express from 'express';

Sentry.init({
  dsn: process.env.SENTRY_DSN,
  tracesSampleRate: 0.2,
  profilesSampleRate: 0.1,
});

const app = express();

app.post('/api/orders', async (req, res) => {
  const order = await Sentry.startSpan(
    { name: 'order.create', op: 'task', attributes: { 'order.source': 'api' } },
    async (span) => {
      const validated = await Sentry.startSpan(
        { name: 'order.validate', op: 'validation' },
        () => validateOrder(req.body)
      );

      const saved = await Sentry.startSpan(
        { name: 'order.save', op: 'db.query' },
        () => db.orders.create(validated)
      );

      await Sentry.startSpan(
        { name: 'notification.send', op: 'http.client' },
        () => notifyWarehouse(saved.id)
      );

      Sentry.setMeasurement('order.total_cents', saved.total, 'none');
      return saved;
    }
  );

  res.status(201).json(order);
});

Sentry.setupExpressErrorHandler(app);
app.listen(3000);

Python — FastAPI with Custom Spans

import os
import sentry_sdk
from sentry_sdk.integrations.fastapi import FastApiIntegration
from sentry_sdk.integrations.starlette import StarletteIntegration
from fastapi import FastAPI

sentry_sdk.init(
    dsn=os.environ["SENTRY_DSN"],
    integrations=[FastApiIntegration(), StarletteIntegration()],
    traces_sample_rate=0.2,
    profiles_sample_rate=0.1,
)

app = FastAPI()

@app.post("/api/orders")
async def create_order(payload: OrderRequest):
    with sentry_sdk.start_span(op="task", name="order.create") as span:
        span.set_data("order_source", "api")

        with sentry_sdk.start_span(op="validation", name="order.validate"):
            validated = validate_order(payload)

        with sentry_sdk.start_span(op="db.query", name="order.save"):
            saved = await db.orders.create(validated)

        with sentry_sdk.start_span(op="http.client", name="notification.send"):
            await notify_warehouse(saved.id)

    return {"id": saved.id, "status": "created"}

Resources

• Set Up Tracing — Node.js
• Custom Instrumentation
• Distributed Tracing
• Python Performance Monitoring
• Profiling — Node.js
• Web Vitals
• Performance Monitoring Product Guide

Next Steps

• Alerting on performance regressions: Configure Performance Alerts in Sentry to trigger when p95 latency exceeds thresholds or throughput drops
• Custom dashboards: Build dashboards in Sentry using custom measurements (Sentry.setMeasurement()) to track business KPIs alongside latency
• Span sampling in high-volume services: Use tracesSampler to selectively trace slow endpoints at higher rates while keeping fast endpoints low
• Connect to error tracking: Errors captured with Sentry.captureException() inside a traced span automatically link to that trace in the Sentry UI