|
| 1 | +<p align="center"> |
| 2 | + <a href="https://sentry.io/?utm_source=github&utm_medium=logo" target="_blank"> |
| 3 | + <img src="https://sentry-brand.storage.googleapis.com/sentry-wordmark-dark-280x84.png" alt="Sentry" width="280" height="84"> |
| 4 | + </a> |
| 5 | +</p> |
| 6 | + |
| 7 | +# Official Sentry Profiling SDK for NodeJS |
| 8 | + |
| 9 | +[](https://www.npmjs.com/package/@sentry/profiling-node) |
| 10 | +[](https://www.npmjs.com/package/@sentry/profiling-node) |
| 11 | +[](https://www.npmjs.com/package/@sentry/profiling-node) |
| 12 | + |
| 13 | +## Installation |
| 14 | + |
| 15 | +Profiling works as an extension of tracing so you will need both @sentry/node and @sentry/profiling-node installed. |
| 16 | + |
| 17 | +```bash |
| 18 | +# Using yarn |
| 19 | +yarn add @sentry/node @sentry/profiling-node |
| 20 | + |
| 21 | +# Using npm |
| 22 | +npm install --save @sentry/node @sentry/profiling-node |
| 23 | +``` |
| 24 | + |
| 25 | +## Usage |
| 26 | + |
| 27 | +```javascript |
| 28 | +import * as Sentry from '@sentry/node'; |
| 29 | +import { nodeProfilingIntegration } from '@sentry/profiling-node'; |
| 30 | + |
| 31 | +Sentry.init({ |
| 32 | + dsn: 'https://7fa19397baaf433f919fbe02228d5470@o1137848.ingest.sentry.io/6625302', |
| 33 | + debug: true, |
| 34 | + tracesSampleRate: 1, |
| 35 | + profilesSampleRate: 1, // Set profiling sampling rate. |
| 36 | + integrations: [nodeProfilingIntegration()], |
| 37 | +}); |
| 38 | +``` |
| 39 | + |
| 40 | +Sentry SDK will now automatically profile all root spans, even the ones which may be started as a result of using an |
| 41 | +automatic instrumentation integration. |
| 42 | + |
| 43 | +```javascript |
| 44 | +Sentry.startSpan({ name: 'some workflow' }, () => { |
| 45 | + // The code in here will be profiled |
| 46 | +}); |
| 47 | +``` |
| 48 | + |
| 49 | +### Building the package from source |
| 50 | + |
| 51 | +Profiling uses native modules to interop with the v8 javascript engine which means that you may be required to build it |
| 52 | +from source. The libraries required to successfully build the package from source are often the same libraries that are |
| 53 | +already required to build any other package which uses native modules and if your codebase uses any of those modules, |
| 54 | +there is a fairly good chance this will work out of the box. The required packages are python, make and g++. |
| 55 | + |
| 56 | +**Windows:** If you are building on windows, you may need to install windows-build-tools |
| 57 | + |
| 58 | +**_Python:_** Python 3.12 is not supported yet so you will need a version of python that is lower than 3.12 |
| 59 | + |
| 60 | +```bash |
| 61 | + |
| 62 | +# using yarn package manager |
| 63 | +yarn global add windows-build-tools |
| 64 | +# or npm package manager |
| 65 | +npm i -g windows-build-tools |
| 66 | +``` |
| 67 | + |
| 68 | +After you have installed the toolchain, you should be able to build the binaries from source |
| 69 | + |
| 70 | +```bash |
| 71 | +# configure node-gyp using yarn |
| 72 | +yarn build:bindings:configure |
| 73 | +# or using npm |
| 74 | +npm run build:bindings:configure |
| 75 | + |
| 76 | +# compile the binaries using yarn |
| 77 | +yarn build:bindings |
| 78 | +# or using npm |
| 79 | +npm run build:bindings |
| 80 | +``` |
| 81 | + |
| 82 | +After the binaries are built, you should see them inside the profiling-node/lib folder. |
| 83 | + |
| 84 | +### Prebuilt binaries |
| 85 | + |
| 86 | +We currently ship prebuilt binaries for a few of the most common platforms and node versions (v18-24). |
| 87 | + |
| 88 | +- macOS x64 |
| 89 | +- Linux ARM64 (musl) |
| 90 | +- Linux x64 (glibc) |
| 91 | +- Windows x64 |
| 92 | + |
| 93 | +For a more detailed list, see job_compile_bindings_profiling_node job in our build.yml github action workflow. |
| 94 | + |
| 95 | +### Bundling |
| 96 | + |
| 97 | +If you are looking to squeeze some extra performance or improve cold start in your application (especially true for |
| 98 | +serverless environments where modules are often evaluates on a per request basis), then we recommend you look into |
| 99 | +bundling your code. Modern JS engines are much faster at parsing and compiling JS than following long module resolution |
| 100 | +chains and reading file contents from disk. Because @sentry/profiling-node is a package that uses native node modules, |
| 101 | +bundling it is slightly different than just bundling javascript. In other words, the bundler needs to recognize that a |
| 102 | +.node file is node native binding and move it to the correct location so that it can later be used. Failing to do so |
| 103 | +will result in a MODULE_NOT_FOUND error. |
| 104 | + |
| 105 | +The easiest way to make bundling work with @sentry/profiling-node and other modules which use native nodejs bindings is |
| 106 | +to mark the package as external - this will prevent the code from the package from being bundled, but it means that you |
| 107 | +will now need to rely on the package to be installed in your production environment. |
| 108 | + |
| 109 | +To mark the package as external, use the following configuration: |
| 110 | + |
| 111 | +[Next.js 13+](https://nextjs.org/docs/app/api-reference/next-config-js/serverComponentsExternalPackages) |
| 112 | + |
| 113 | +```js |
| 114 | +const { withSentryConfig } = require('@sentry/nextjs'); |
| 115 | + |
| 116 | +/** @type {import('next').NextConfig} */ |
| 117 | +const nextConfig = { |
| 118 | + experimental: { |
| 119 | + // Add the "@sentry/profiling-node" to serverComponentsExternalPackages. |
| 120 | + serverComponentsExternalPackages: ['@sentry/profiling-node'], |
| 121 | + }, |
| 122 | +}; |
| 123 | + |
| 124 | +module.exports = withSentryConfig(nextConfig, { |
| 125 | + /* ... */ |
| 126 | +}); |
| 127 | +``` |
| 128 | + |
| 129 | +[webpack](https://webpack.js.org/configuration/externals/#externals) |
| 130 | + |
| 131 | +```js |
| 132 | +externals: { |
| 133 | + "@sentry/profiling-node": "commonjs @sentry/profiling-node", |
| 134 | +}, |
| 135 | +``` |
| 136 | + |
| 137 | +[esbuild](https://esbuild.github.io/api/#external) |
| 138 | + |
| 139 | +```js |
| 140 | +{ |
| 141 | + entryPoints: ['index.js'], |
| 142 | + platform: 'node', |
| 143 | + external: ['@sentry/profiling-node'], |
| 144 | +} |
| 145 | +``` |
| 146 | + |
| 147 | +[Rollup](https://rollupjs.org/configuration-options/#external) |
| 148 | + |
| 149 | +```js |
| 150 | +{ |
| 151 | + entry: 'index.js', |
| 152 | + external: '@sentry/profiling-node' |
| 153 | +} |
| 154 | +``` |
| 155 | + |
| 156 | +[serverless-esbuild (serverless.yml)](https://www.serverless.com/plugins/serverless-esbuild#external-dependencies) |
| 157 | + |
| 158 | +```yml |
| 159 | +custom: |
| 160 | + esbuild: |
| 161 | + external: |
| 162 | + - @sentry/profiling-node |
| 163 | + packagerOptions: |
| 164 | + scripts: |
| 165 | + - npm install @sentry/profiling-node |
| 166 | +``` |
| 167 | +
|
| 168 | +[vercel-ncc](https://github.com/vercel/ncc#programmatically-from-nodejs) |
| 169 | +
|
| 170 | +```js |
| 171 | +{ |
| 172 | + externals: ["@sentry/profiling-node"], |
| 173 | +} |
| 174 | +``` |
| 175 | + |
| 176 | +[vite](https://vitejs.dev/config/ssr-options.html#ssr-external) |
| 177 | + |
| 178 | +```js |
| 179 | +ssr: { |
| 180 | + external: ['@sentry/profiling-node']; |
| 181 | +} |
| 182 | +``` |
| 183 | + |
| 184 | +Marking the package as external is the simplest and most future proof way of ensuring it will work, however if you want |
| 185 | +to bundle it, it is possible to do so as well. Bundling has the benefit of improving your script startup time as all of |
| 186 | +the code is (usually) inside a single executable .js file, which saves time on module resolution. |
| 187 | + |
| 188 | +In general, when attempting to bundle .node native file extensions, you will need to tell your bundler how to treat |
| 189 | +these, as by default it does not know how to handle them. The required approach varies between build tools and you will |
| 190 | +need to find which one will work for you. |
| 191 | + |
| 192 | +The result of bundling .node files correctly is that they are placed into your bundle output directory with their |
| 193 | +require paths updated to reflect their final location. |
| 194 | + |
| 195 | +Example of bundling @sentry/profiling-node with esbuild and .copy loader |
| 196 | + |
| 197 | +```json |
| 198 | +// package.json |
| 199 | +{ |
| 200 | + "scripts": "node esbuild.serverless.js" |
| 201 | +} |
| 202 | +``` |
| 203 | + |
| 204 | +```js |
| 205 | +// esbuild.serverless.js |
| 206 | +const { sentryEsbuildPlugin } = require('@sentry/esbuild-plugin'); |
| 207 | + |
| 208 | +require('esbuild').build({ |
| 209 | + entryPoints: ['./index.js'], |
| 210 | + outfile: './dist', |
| 211 | + platform: 'node', |
| 212 | + bundle: true, |
| 213 | + minify: true, |
| 214 | + sourcemap: true, |
| 215 | + // This is no longer necessary |
| 216 | + // external: ["@sentry/profiling-node"], |
| 217 | + loader: { |
| 218 | + // ensures .node binaries are copied to ./dist |
| 219 | + '.node': 'copy', |
| 220 | + }, |
| 221 | + plugins: [ |
| 222 | + // See https://docs.sentry.io/platforms/javascript/sourcemaps/uploading/esbuild/ |
| 223 | + sentryEsbuildPlugin({ |
| 224 | + project: '', |
| 225 | + org: '', |
| 226 | + authToken: '', |
| 227 | + release: '', |
| 228 | + sourcemaps: { |
| 229 | + // Specify the directory containing build artifacts |
| 230 | + assets: './dist/**', |
| 231 | + }, |
| 232 | + }), |
| 233 | + ], |
| 234 | +}); |
| 235 | +``` |
| 236 | + |
| 237 | +Once you run `node esbuild.serverless.js` esbuild wil bundle and output the files to ./dist folder, but note that all of |
| 238 | +the binaries will be copied. This is wasteful as you will likely only need one of these libraries to be available during |
| 239 | +runtime. |
| 240 | + |
| 241 | +To prune the other libraries, profiling-node ships with a small utility script that helps you prune unused binaries. The |
| 242 | +script can be invoked via `sentry-prune-profiler-binaries`, use `--help` to see a list of available options or |
| 243 | +`--dry-run` if you want it to log the binaries that would have been deleted. |
| 244 | + |
| 245 | +Example of only preserving a binary to run node16 on linux x64 musl. |
| 246 | + |
| 247 | +```bash |
| 248 | +sentry-prune-profiler-binaries --target_dir_path=./dist --target_platform=linux --target_node=16 --target_stdlib=musl --target_arch=x64 |
| 249 | +``` |
| 250 | + |
| 251 | +Which will output something like |
| 252 | + |
| 253 | +``` |
| 254 | +Sentry: pruned ./dist/sentry_cpu_profiler-darwin-x64-108-IFGH3SUR.node (90.41 KiB) |
| 255 | +Sentry: pruned ./dist/sentry_cpu_profiler-darwin-x64-93-Q7KBVHSP.node (74.16 KiB) |
| 256 | +Sentry: pruned ./dist/sentry_cpu_profiler-linux-arm64-glibc-108-NXSISRTB.node (52.17 KiB) |
| 257 | +Sentry: pruned ./dist/sentry_cpu_profiler-linux-arm64-glibc-83-OEQT5HUK.node (52.08 KiB) |
| 258 | +Sentry: pruned ./dist/sentry_cpu_profiler-linux-arm64-glibc-93-IIXXW2PN.node (52.06 KiB) |
| 259 | +Sentry: pruned ./dist/sentry_cpu_profiler-linux-arm64-musl-108-DSILNYHA.node (48.46 KiB) |
| 260 | +Sentry: pruned ./dist/sentry_cpu_profiler-linux-arm64-musl-83-4CNOBNC3.node (48.37 KiB) |
| 261 | +Sentry: pruned ./dist/sentry_cpu_profiler-linux-arm64-musl-93-JA5PKNWQ.node (48.38 KiB) |
| 262 | +Sentry: pruned ./dist/sentry_cpu_profiler-linux-x64-glibc-108-NXSISRTB.node (52.17 KiB) |
| 263 | +Sentry: pruned ./dist/sentry_cpu_profiler-linux-x64-glibc-83-OEQT5HUK.node (52.08 KiB) |
| 264 | +Sentry: pruned ./dist/sentry_cpu_profiler-linux-x64-glibc-93-IIXXW2PN.node (52.06 KiB) |
| 265 | +Sentry: pruned ./dist/sentry_cpu_profiler-linux-x64-musl-108-CX7SL27U.node (51.50 KiB) |
| 266 | +Sentry: pruned ./dist/sentry_cpu_profiler-linux-x64-musl-83-YD7ZQK2E.node (51.53 KiB) |
| 267 | +Sentry: pruned ./dist/sentry_cpu_profiler-win32-x64-108-P7V3URQV.node (181.50 KiB) |
| 268 | +Sentry: pruned ./dist/sentry_cpu_profiler-win32-x64-93-3PKQDSGE.node (181.50 KiB) |
| 269 | +✅ Sentry: pruned 15 binaries, saved 1.06 MiB in total. |
| 270 | +``` |
| 271 | + |
| 272 | +### Environment flags |
| 273 | + |
| 274 | +The default mode of the v8 CpuProfiler is kEagerLoggin which enables the profiler even when no profiles are active - |
| 275 | +this is good because it makes calls to startProfiling fast at the tradeoff for constant CPU overhead. The behavior can |
| 276 | +be controlled via the `SENTRY_PROFILER_LOGGING_MODE` environment variable with values of `eager|lazy`. If you opt to use |
| 277 | +the lazy logging mode, calls to startProfiling may be slow (depending on environment and node version, it can be in the |
| 278 | +order of a few hundred ms). |
| 279 | + |
| 280 | +Example of starting a server with lazy logging mode. |
| 281 | + |
| 282 | +```javascript |
| 283 | +SENTRY_PROFILER_LOGGING_MODE=lazy node server.js |
| 284 | +``` |
| 285 | + |
| 286 | +## FAQ 💭 |
| 287 | + |
| 288 | +### Can the profiler leak PII to Sentry? |
| 289 | + |
| 290 | +The profiler does not collect function arguments so leaking any PII is unlikely. We only collect a subset of the values |
| 291 | +which may identify the device and os that the profiler is running on (if you are already using tracing, it is likely |
| 292 | +that these values are already being collected by the SDK). |
| 293 | + |
| 294 | +There is one way a profiler could leak pii information, but this is unlikely and would only happen for cases where you |
| 295 | +might be creating or naming functions which might contain pii information such as |
| 296 | + |
| 297 | +```js |
| 298 | +eval('function scriptFor${PII_DATA}....'); |
| 299 | +``` |
| 300 | + |
| 301 | +In that case it is possible that the function name may end up being reported to Sentry. |
| 302 | + |
| 303 | +### Are worker threads supported? |
| 304 | + |
| 305 | +No. All instances of the profiler are scoped per thread In practice, this means that starting a transaction on thread A |
| 306 | +and delegating work to thread B will only result in sample stacks being collected from thread A. That said, nothing |
| 307 | +should prevent you from starting a transaction on thread B concurrently which will result in two independent profiles |
| 308 | +being sent to the Sentry backend. We currently do not do any correlation between such transactions, but we would be open |
| 309 | +to exploring the possibilities. Please file an issue if you have suggestions or specific use-cases in mind. |
| 310 | + |
| 311 | +### How much overhead will this profiler add? |
| 312 | + |
| 313 | +The profiler uses the kEagerLogging option by default which trades off fast calls to startProfiling for a small amount |
| 314 | +of constant CPU overhead. If you are using kEagerLogging then the tradeoff is reversed and there will be a small CPU |
| 315 | +overhead while the profiler is not running, but calls to startProfiling could be slow (in our tests, this varies by |
| 316 | +environments and node versions, but could be in the order of a couple 100ms). |
0 commit comments