@sentry/node

Important Changes

• Rename diagnostic_channel, fetch and gRPC span origins

  Some span origins changed their names to match the other v11 origins:

  • auto.graphql.diagnostic_channel is now auto.graphql.graphql.

  • auto.db.redis.diagnostic_channel, auto.db.mysql2.diagnostic_channel and auto.db.mongoose.diagnostic_channel now end in .native_channel.

  • auto.http.wintercg_fetch is now auto.http.fetch, as on Bun, Deno and Cloudflare.

  • auto.grpc.serverless is now auto.grpc.google_cloud.

  If you have ignoreSpans rules, queries or alerts on the old origins, change them to use the new origins.

• deno: Add incoming request options to denoServeIntegration

  denoServeIntegration now accepts ignoreRequestBody, ignoreIncomingRequests, ignoreStaticAssets, spans, onSpanCreated and onSpanEnd, with the same meaning as in denoHttpIntegration. The callbacks receive the web Request.

  ignoreStaticAssets defaults to true, so Deno.serve no longer creates spans for GET requests to common static assets such as favicon.ico, robots.txt or *.js files. To keep these spans, set denoServeIntegration({ ignoreStaticAssets: false }). Deno.serve spans also follow the client's tracing configuration by default, so the integration does not start spans when tracing is disabled.

• deno: Record request sessions for Deno.serve

  denoServeIntegration now records a request session for each incoming request, as denoHttpIntegration does for node:http. This gives release health and crash-free rates for Deno.serve apps. Sessions are sent only when a release is set. To turn them off, set denoServeIntegration({ sessions: false }). sessionFlushingDelayMS sets how long the SDK waits before it sends the session aggregate.

• deno: Register channel injection hooks from init()

  Sentry.init() in @sentry/deno now registers the diagnostics-channel injection hooks, as @sentry/node does. The channel integrations now record spans without --preload=npm:@sentry/deno/import, for modules that you import after init(). To also instrument modules that load before init(), such as static imports in your entry module, keep the preload. With debug: true, the SDK logs a warning when no injection is active.

  On Node.js, tracing-hooks messages now show with debug: true instead of with DEBUG=@apm-js-collab/*.

• nextjs: Add a tunnel route handler that sends only the event payload to Sentry

  The tunnelRoute option can now send events through a route handler in your app. The handler forwards only the event payload to Sentry, and no request headers. To use it, add this file next to your tunnelRoute option:

  // app/sentry-tunnel/route.ts (or src/app/sentry-tunnel/route.ts)
  export { POST } from '@sentry/nextjs/tunnel';

  The handler accepts events for the DSN of your server SDK. To accept other DSNs, use createSentryTunnelRoute({ allowedDsns }) from the same import. Without the file, the tunnel route works as before.

  Also, the server no longer sends spans of tunnel requests when span streaming is on, because these spans only show the SDK's own traffic.

• nextjs: Never emit trace meta tags from a prerender

  Pageloads of apps with cacheComponents enabled are connected to their server request again. Pages that Next.js prerenders no longer carry trace information, so a cached page can no longer connect visitors to a request that was not theirs. This also applies to ISR pages. The pageload waits for the trace information when it arrives with the part of the page that streams in after a cached shell.

  If you set up OpenTelemetry yourself and register SentryPropagator, register NextSentryPropagator instead and pass your tracer provider to markPrerendersOnTracerProvider, so that your prerendered pages carry no trace information either.

• nextjs: Send tunnel route requests with credentials: 'omit'

  With tunnelRoute and no tunnel route handler, @sentry/nextjs now sends events from the browser with credentials: 'omit', because the tunnel route does not need cookies. If a layer in front of your app checks cookies before it lets requests through (for example Vercel Deployment Protection), you can send them again in your browser Sentry.init call:

  // instrumentation-client.ts
  Sentry.init({
    transportOptions: { fetchOptions: { credentials: 'same-origin' } },
  });

• nitro: Report a single http.server span per request

  A Nitro request now creates only one http.server span. The nested http.server spans with the origins auto.http.nitro.srvx and auto.http.nitro.h3 are removed. h3 route handlers now get a handler span, and middleware spans are direct children of the http.server span. If you have queries or alerts on the removed spans, change them to use the http.server span or the handler span.

• Use auto.http.http_server origin for all node:http server spans

  On Bun and Deno, node:http server spans now have the origin auto.http.http_server instead of auto.http.server, as on Node.js. On Node.js, nothing changes. Bun.serve spans keep auto.http.bun.serve and Deno.serve spans keep auto.http.deno. If you have ignoreSpans or beforeSendSpan rules, queries or alerts on auto.http.server, change them to use auto.http.http_server.

Other Changes

• browser: Add LoaderSentry type export

• cloudflare: Add isStreamingResponse option and detect streamed RSC responses

• core: Add dataCollection.mcp option

• core: Limit the body size of tunnel requests

• nextjs: Link use cache hits to their fill across instances

• nextjs: Wait for the resumed part of a document before starting the pageload

• nitro: Merge the nitro integrations into a single nitroIntegration

• server-utils: Emit execute_tool spans for Anthropic's beta.messages.toolRunner()

• astro: Defer to @sentry/cloudflare/vite on Cloudflare Workers

• aws-serverless: End the function.aws span when a callback-style handler calls back

• browser: Skip profiling for unsampled active root spans

• bundler-plugins: Check globalThis before global in the injected snippet

• bundler-plugins: Inject the release only into chunks that load at startup

• cloudflare: auto-instrument a re-exported default export

• cloudflare: Deliver check-ins and events from forked isolation scopes

• cloudflare: Don't hold Durable Object invocations open with the segment-end flush timer

• cloudflare: Keep default exports intact when auto-instrumenting

• core: Use trace_metric data category for metrics rate limits and client reports

• deno: Re-export portable helpers that Cloudflare exposes

• elysia: Set SDK metadata on the options passed to init

• elysia: Use Node transport on Node runtimes

• hono: Read connection info from the runtime instead of Hono helpers

• nextjs: Only inject applicationKey metadata into browser bundles

• nitro: Use forward slashes for resolved plugin paths on Windows

• node,vercel-edge,bundler-plugins: Ignore commit env vars set to their own name

• node: Guard user callbacks in the node:http server instrumentation

• node: Only read JavaScript and TypeScript source files in ContextLines

• node: Restore tsx runtime instrumentation

• nuxt: Delete Sentry-enabled source maps from Nitro output directories

• nuxt: Keep sourcesContent in Nuxt 5 server source maps

• react: Don't crash ErrorBoundary when error.cause is not an Error

• react: Match React Router navigations against root routes only

• server-runtime-injection: Load ESM hook without the build/esm package.json marker

• server-utils: Bundle the CommonJS dependencies of instrumented packages on Vite 8

• server-utils: Keep OpenAI decision confidence on Vercel AI decide spans

• server-utils: Nest LangChain run spans under their parent run

• server-utils: Stop the orchestrion Vite plugin from bundling the CJS build of @sentry/server-utils

• solidstart: Read server action route from the request's root span

• tanstackstart-react: Delete only the source maps that Sentry enabled

• vercel-edge: Expose portable tracing and feature flag helpers

Internal Changes

Work in this release was contributed by @4ktLuffy, @adityareddy-dev, @dasilvaluis, @faizansaiyed123, @jpdefo, @Jxxunnn, @mikamikasuki, @philipatkinson, @rubas, and @Shubham-Padkonde. Thank you for your contributions!

• bun: Add incoming request options and request sessions to bunServerIntegration

  bunServerIntegration now accepts the same options as denoServeIntegration: ignoreRequestBody, ignoreIncomingRequests, ignoreStaticAssets, spans, onSpanCreated, onSpanEnd, sessions and sessionFlushingDelayMS. The callbacks receive the web Request.

  Bun.serve now records a request session for each incoming request. Sessions are sent only when a release is set. To turn them off, set bunServerIntegration({ sessions: false }).

  ignoreStaticAssets defaults to true, so Bun.serve no longer creates spans for GET requests to common static assets such as favicon.ico, robots.txt or *.js files. To keep these spans, set bunServerIntegration({ ignoreStaticAssets: false }). Bun.serve spans also follow the client's tracing configuration by default, so the integration does not start spans when tracing is disabled. Errors thrown for OPTIONS and HEAD requests are now captured.

• node: Add onSpanEnd to httpIntegration

  httpIntegration and httpServerSpansIntegration now accept onSpanEnd, a hook that runs when the response for an incoming request finishes, before its span ends. Use it to change the span based on the response. Deno and Bun already have this option.

• bun: Inject debug IDs and upload source maps in sentryBunPlugin

  sentryBunPlugin from @sentry/bun/plugin now accepts the same options as sentryEsbuildPlugin (org, project, authToken, release, sourcemaps, ...). When Bun.build writes source maps to outdir, the plugin injects the debug ID that Bun writes into each bundle, injects the release, and uploads the source maps at the end of the build. You no longer need to run sentry sourcemap inject and sentry sourcemap upload after the build.

  Like the other bundler plugins, sentryBunPlugin now sends build telemetry and injects the release it detects (for example, from git). To turn off telemetry, set telemetry: false.

Important Changes

• bundler-plugins: Release @sentry/webpack-plugin, @sentry/vite-plugin, @sentry/rollup-plugin and @sentry/esbuild-plugin from the SDK repository

  The four standalone bundler plugins now release together with the SDK, so their version jumps from 5.4.1 to the SDK version. Each package re-exports the matching @sentry/bundler-plugins entry point, which we recommend for new projects. The plugin options do not change, but the update brings these breaking changes:

  • Node.js 20.19.0 or higher is required.

  • @sentry/webpack-plugin requires webpack 5.1 or higher. The @sentry/webpack-plugin/webpack5 entry point still works and exports the same plugin as @sentry/webpack-plugin.

  • The plugins use the sentry CLI SDK instead of @sentry/cli. Sentry self-hosted 26.4.2 or higher is supported. Lower versions may continue to work.

  • Deploys that the plugins create on Vercel use the value of VERCEL_TARGET_ENV (production, preview or a custom name) as their environment, instead of vercel-production or vercel-preview. To keep the old value, set release.deploy.env.

• server-utils: Instrument the OpenAI decisions API as gen_ai.decide spans

  OpenAI decisions.create calls now create gen_ai.decide spans.

Other Changes

• core: Support SuppressedError in LinkedErrors integration

• astro: Delete generated source maps for static output

• astro: Upload source maps of Astro 7's prerender build

• browser: Apply urlQueryParams to the Referer header

• cloudflare: auto-instrument when wrangler main is a package

• hono: Expose runtime-specific SDK options

• node: Instrument Prisma 8 CollectionBase from rc.15

• nuxt: Skip 3xx/4xx Nuxt errors in the vue:error hook

• server-utils: Read pi-durable 1.2.0 tool results from output

• server-utils: Transform modules once when two Sentry Vite plugins add orchestrion

• sveltekit: Add op and origin to Kit's sveltekit.endpoint span

• tanstackstart-react: Keep parametrized route names in Nitro 3 builds

• tanstackstart-react: Rewrite source paths of split route modules

Internal Changes

• license: Add cli FSL notice to package licenses

Work in this release was contributed by @kunal649 and @mketiku. Thank you for your contributions!

Important Changes

• cloudflare: Add opt-in cron monitoring for Cron Triggers

  The new cronTriggersIntegration sends check-ins for Cron Triggers. Each check-in includes the schedule from controller.cron, so Sentry creates the monitor on the first run. The integration is opt-in, because monitors are billed.

  Cron Triggers have no names, so map each cron expression of your wrangler config to a monitor slug. With the Sentry Cloudflare Vite plugin, add the integration to src/instrument.server.ts:

  // src/monitorSlugs.ts
  export const monitorSlugs: Record<string, string> = {
    '30 9 * * MON-FRI': 'daily-report',
  };
  
  // src/instrument.server.ts
  import * as Sentry from '@sentry/cloudflare';
  import { monitorSlugs } from './monitorSlugs';
  
  export default Sentry.defineCloudflareOptions(env => ({
    dsn: env.SENTRY_DSN,
    integrations: [Sentry.cronTriggersIntegration({ monitorSlug: cron => monitorSlugs[cron] })],
  }));
  
  // src/index.ts
  import { monitorSlugs } from './monitorSlugs';
  
  export default {
    async scheduled(controller, env) {
      switch (monitorSlugs[controller.cron]) {
        case 'daily-report':
          // Your daily report job
          break;
      }
    },
  };

  The monitorSlug function can also return monitor settings, for example { monitorSlug: 'daily-report', maxRuntime: 30 }, or undefined to send no check-ins for a trigger. Without monitorSlug, the SDK derives the slug from the cron expression. If you deploy with wrangler directly, pass the same integration to Sentry.withSentry.

• node: Add neonIntegration for @neondatabase/serverless (#25164, #25176)

  The new neonIntegration creates db spans for queries of the @neondatabase/serverless driver: for the WebSocket Client and Pool, and for the HTTP neon() function. This includes the neon-http and neon-serverless adapters of Drizzle. The integration is on by default when tracing is enabled.

• node: Stop sending the process session once request sessions are tracked

  Node servers sent release health data two times: as request session aggregates, and as the process session of processSessionIntegration. An error in a request also marked the process session as errored, so the SDK counted the error two times. Now the SDK drops the process session when it tracks the first request session. Scripts, CLIs and workers that do not serve requests keep the process session, and so do servers with httpServerIntegration({ sessions: false }). This changes the release health numbers of Node servers, because they no longer send the errored or crashed update of the process session.

• server-utils: Record decide calls as gen_ai.decide spans (#25270, #25272, #25274, #25281)

  Vercel AI experimental_decide calls, Mastra classifier evaluations, TypeSafe and LangChain Jev calls, and Workers AI calls to TypeSafe Jev and Cloudflare Clef models now create decide <model> spans with the op gen_ai.decide. Before, these spans had the op gen_ai.evaluate, which is now deprecated in the Sentry conventions. If you use gen_ai.evaluate in queries, alerts or dashboards, change it to gen_ai.decide.

• browser: Apply urlQueryParams to more URLs (#25262, #25289)

  The dataCollection.urlQueryParams option now also filters the query parameters of these URLs:

  • the page URL in event.request.url

  • the browser.* and resource.* span names with traceLifecycle: 'static'

  • the LCP URL attributes lcp.url and browser.web_vital.lcp.url

  • the URLs in fetch, XHR and navigation breadcrumbs

  • the ui.element.url attribute of elementTimingIntegration

  Before this fix, these URLs could send sensitive query values, for example ?jwt=..., to Sentry.

• nextjs: Apply top-level bundleSizeOptimizations to webpack and Turbopack builds

  The top-level bundleSizeOptimizations option now removes unused SDK code in webpack and in Turbopack builds. Before, the tree-shaking options had no effect on Turbopack. The webpack.treeshake option is deprecated. It still works, but when you set both options, bundleSizeOptimizations takes precedence.

Other Changes

• nuxt: Detect the server API with @nuxt/kit 4.6

• bundler-plugins: Escape JSON in generated injection code

• core: Don't mark http client spans as errors when the caller aborts

• core: Record span client report outcomes when discarding transactions

• core: Record client outcomes when beforeSend callbacks return invalid values

• core: Record span outcomes for spans dropped when building a transaction

• core: Report actual item count in client reports of envelope send failures

• core: Strip internal custom span name from streamed spans

• nextjs: Filter query parameters in request error context

• node: Don't mark fetch spans as errors when the caller aborts

• nuxt: Delete public source maps before Nitro lists its public assets

• nuxt: Keep @sentry/server-runtime-injection external on Nitro v3

• nuxt: Remove @opentelemetry/resources dev alias

• profiling-node: Stop the trace lifecycle profiler when spans are unsampled

• react-router: Respect custom default integrations

• remix: Respect custom default integrations

• server-utils: Avoid quadratic SQL summary regex backtracking

• server-utils: End Express layer spans when the client aborts

• server-utils: Record amqplib consumer info before the first delivery and instrument the callback API

• server-utils: Stop treating previous_response_id as the OpenAI conversation id

• sveltekit: Read load event pathname untracked

• sveltekit: Stop adapter-node 6 output dir probe from deleting the build

Internal Changes

Work in this release was contributed by @Philip-Nwabuwa and @4ktLuffy. Thank you for your contributions!