Welcome to agoda-devfeedback, the JavaScript/TypeScript package collection that's about to make your builds faster than a caffeinated squirrel on a sugar rush! We're here to collect metrics that relate to developers' experience, because who doesn't love a good statistic about how long they've been waiting for their build to finish?
What is the F5 Experience? Have a read here
Remember, we're all about that F5 Experience here at agoda-devfeedback. Our goal is to make your development process smoother than a JavaScript promise chain. Here's what that means for you:
- Setup Should Be a Breeze: You should be able to install these packages and get metrics faster than you can say "npm install".
- Fast Feedback Loop: We want your builds to be so fast, you'll forget what you were working on by the time they finish. (Okay, maybe not that fast, but you get the idea.)
This collection supports collecting build time (compilation time) metrics across multiple bundlers:
- Webpack (4.x or 5.x)
- Vite (4.x and up, including 6.x and Rolldown-based builds)
- Rspack/Rsbuild (1.x)
It's like a stopwatch for your builds, but cooler, and now with more bundlers! 🎮
The data is sent to the following default endpoints (customizable via environment variables):
| Bundler | Default | Environment Variable Override | Post Data Example |
|---|---|---|---|
| WebPack | "http://compilation-metrics/webpack" | WEBPACK_ENDPOINT | click here |
| Vite | "http://compilation-metrics/vite" | VITE_ENDPOINT | click here |
| Rspack | "http://compilation-metrics/rspack" | RSPACK_ENDPOINT | click here |
Lifecycle (type: "command") |
"http://compilation-metrics/command" | COMMAND_ENDPOINT | click here |
Heads up: Rspack and Rsbuild events used to be posted to the webpack endpoint despite the table above. They now go to
/rspackas documented. If your dashboards were reading them off the webpack endpoint, point them at/rspack(or setRSPACK_ENDPOINTback to the webpack URL during the transition).
First, let's get this party started. Install the package for your bundler of choice:
# For Webpack
npm install --save-dev agoda-devfeedback-webpack
# For Vite
npm install --save-dev agoda-devfeedback-vite2
# For Rspack/Rsbuild
npm install --save-dev agoda-devfeedback-rsbuildPro tip: When an error happens, the package will write the error message to devfeedback.log in your current working directory. You might want to add this to .gitignore, unless you want your repo to know about all your build failures. We won't judge.
If you're using Webpack, sprinkle this magic into your webpack.config.js:
const { WebpackBuildStatsPlugin } = require('agoda-devfeedback-webpack');
module.exports = {
// ... your other awesome config stuff ...
plugins: [
// ... your other cool plugins ...
new WebpackBuildStatsPlugin(),
],
};If Vite is your jam, add this to your vite.config.js:
import { viteBuildStatsPlugin } from 'agoda-devfeedback-vite2';
export default defineConfig({
// ... your brilliant config options ...
plugins: [
// ... your other fantastic plugins ...
viteBuildStatsPlugin(),
],
});If you're rocking with Rspack/Rsbuild, add this to your rsbuild.config.js:
import { RsbuildBuildStatsPlugin } from 'agoda-devfeedback-rsbuild';
export default {
// ... your awesome config options ...
plugins: [
// ... your other amazing plugins ...
RsbuildBuildStatsPlugin,
],
};Using raw Rspack (@rspack/cli, rspack serve) without Rsbuild? Add the plugin to your rspack.config.js instead:
const { RspackBuildStatsPlugin } = require('agoda-devfeedback-rsbuild');
module.exports = {
// ... your config ...
plugins: [new RspackBuildStatsPlugin()],
};All plugins accept a custom identifier for your builds. It's like a name tag for your builds! By default, they'll use your npm script name (like yarn dev or yarn build).
But if you want to be fancy:
// Webpack
new WebpackBuildStatsPlugin('production-build-deluxe');
// Vite
viteBuildStatsPlugin('vite-build-extraordinaire');
// Raw Rspack (rspack.config.js) — positional, or as an options object
new RspackBuildStatsPlugin('rspack-build-supreme');
new RspackBuildStatsPlugin({ customIdentifier: 'rspack-build-supreme' });The Rsbuild plugin always uses the npm script name.
Want to track bootstrap chunk sizes? We've got you covered! Pass a size limit (in KB) and production builds report bundleStats for the chunk named bootstrap (Vite, Rspack and Rsbuild):
// Vite: the second parameter
viteBuildStatsPlugin('vite-build-extraordinaire', 1000); // 1 mega byte
// Raw Rspack: an option
new RspackBuildStatsPlugin({ bootstrapBundleSizeLimitKb: 1000 });
// Rsbuild: use the factory instead of the plain plugin
import { rsbuildBuildStatsPlugin } from 'agoda-devfeedback-rsbuild';
plugins: [rsbuildBuildStatsPlugin({ bootstrapBundleSizeLimitKb: 1000 })];bootstrapChunkSizeBytes is the size of the chunk's emitted JavaScript, and bootstrapChunkSizeLimitBytes is the limit you passed, in bytes. The size is reported without a limit too, and is left out when there is no bootstrap chunk. Rspack and Rsbuild only report it for production builds, since dev chunk sizes say nothing about the budget. See examples/rspack.json.
A compile time is one number out of the several a developer actually waits through. Between git pull and a working app there is an install, a dev server start, and a browser that has to finish booting. Vite and Rspack/Rsbuild now report all of them.
These arrive as a new event type, type: "command", on the COMMAND_ENDPOINT:
phase |
What it measures | Emitted by |
|---|---|---|
install |
The package manager run, plus cold/warm and whether the lockfile changed | install hooks (npm, yarn, pnpm) |
devserver |
Time until the dev server is listening — the number that was previously never measured, because closeBundle does not fire in dev |
Vite, Rsbuild, Rspack watch |
clientready |
Time until the app is usable in the browser, with DOMContentLoaded and first contentful paint | Vite, Rsbuild, Rspack watch |
codegen, typecheck, lint, test, build, other |
A package script or task-runner task, see Package scripts and task runners | pnpm script-shell wrapper, Nx plugin |
Aborted runs count too: Ctrl-C on a dev server produces a devserver event with success: false and signal: "SIGINT". A developer who gave up waiting is the most interesting data point on the chart.
RspackBuildStatsPlugin only sees compiler hooks, not the dev server, so its spans are anchored on the compiler. All of them carry command: "rspack watch", which is how a dashboard tells them apart from the Vite and Rsbuild numbers.
devserverruns from the firstwatchRunto the first completed compile. Unlike Vite and Rsbuild, which stop the clock when the server is listening, it includes the first compile: underrspack servethe server is not usable until that compile is done anyway, but don't put the two on one axis without a filter.clientreadyruns from the same firstwatchRununtil the browser has loaded the page and painted, withdomContentLoadedMsandfirstContentfulPaintMsas detail. Once per process: reloads and extra tabs are not new dev server starts.rspackhmr(a separate event type onRSPACK_ENDPOINT, like Vite'svitehmr): edit → hot update applied in the browser, timed from the rebuild'swatchRun. One per rebuild, however many tabs are open, carrying that rebuild'scompilationHash. Sent without printing anything, likevitehmr. Example.
To report back, the plugin adds a small runtime module to the bundle, in watch mode only: production, one-off and child compilations are untouched. It connects to ws://<page hostname>:<random port>, so it also works when the dev server runs on a remote machine, as long as that port is reachable. No HMR (hot: false) means no rspackhmr events; clientready still works.
Every event — including the existing webpack, vite, vitehmr, rspack and rsbuild payloads — now carries a sessionId, so install → dev server ready → first HMR stitch into one timeline. It is purely additive; nothing that existed changed shape.
The session id is resolved with zero setup: a small state file under node_modules/.cache/devfeedback (or a tmpdir, before node_modules exists) that rolls over after four idle hours. If you want exact session boundaries, set one yourself and it wins:
export DEVFEEDBACK_SESSION_ID=$(uuidgen)Two tiers, and the first one needs nothing from you.
Default — no repo change. agoda-devfeedback-common runs its own postinstall hook, which infers the install duration from the package manager's process start time. The span ends when our package is linked rather than when the whole install finishes, so it undercounts a little.
Exact — two lines in the consuming repo. Add both hooks and you get the true install span, plus a trustworthy cold/warm flag on a fresh clone:
{
"scripts": {
"preinstall": "node -e \"try{require('agoda-devfeedback-common/hooks/preinstall')}catch(e){}\"",
"postinstall": "node -e \"try{require('agoda-devfeedback-common/hooks/postinstall')}catch(e){}\""
},
"devDependencies": {
"agoda-devfeedback-common": "^2.0.0"
}
}Add agoda-devfeedback-common as a direct devDependency for this tier — under pnpm a transitive dependency is not resolvable from the repo root. When these hooks are present the bundled one stands down, so you get one event, not two. On a genuinely cold clone the preinstall file does not exist yet, the try/catch swallows it, and the bundled hook falls back to process start time — cold clones are still measured, just less precisely.
Known gaps, so nobody is surprised:
- Install events are spooled, not sent. An install must never wait on the network, so events are written to a small local NDJSON file and delivered by the next dev server or build start.
spooledAttells you the delivery was deferred. Nobody watches an install dashboard in real time. --ignore-scriptsskips everything here.- pnpm blocks dependency lifecycle scripts by default. Allow
agoda-devfeedback-commonin one of two places:- In the repo (recommended for monorepos that already keep a build allowlist): add it to the allowlist in
pnpm-workspace.yaml, or in thepnpmfield of the rootpackage.json. Every clone then works without a bootstrap step.# pnpm-workspace.yaml, newer pnpm allowBuilds: agoda-devfeedback-common: true
# pnpm-workspace.yaml, pnpm 10 and earlier onlyBuiltDependencies: - agoda-devfeedback-common
- Per machine, once during dev machine bootstrap, in
~/.config/pnpm/config.yaml, so repos stay untouched:(pnpm 10 and earlier call thisallowBuilds: agoda-devfeedback-common: true
onlyBuiltDependencies.)
- In the repo (recommended for monorepos that already keep a build allowlist): add it to the allowlist in
- npm only: with
timing=truein.npmrc, npm's own per-phase timers are scraped from~/.npm/_logs/*-timing.jsonand attached asnpmTimers. That is where you find out a Playwright browser download or anode-gyprebuild is what actually costs you three minutes.
The bundler plugins only see the bundler. Typecheck, lint, test, codegen and every other package script run around it, so agoda-devfeedback-common ships two more emitters for those. Both produce command events that share the sessionId, the spool and the transport with everything else.
pnpm: every pnpm run <script> (pnpm ≥ 10.13, macOS and Linux). devfeedback-shell is a scriptShell that runs the real shell (/bin/sh) and records the script. Wire it up in .pnpmfile.cjs, with agoda-devfeedback-common as a root devDependency:
// .pnpmfile.cjs
try { module.exports = require('agoda-devfeedback-common/pnpmfile').pnpmfile(__dirname); } catch { module.exports = {}; }hooks.updateConfig runs on every pnpm command and sets scriptShell to the absolute path of the wrapper only once it is installed, so a cold clone's first pnpm install runs exactly as before. Don't put script-shell=./node_modules/.bin/devfeedback-shell in .npmrc: that path doesn't exist on a fresh clone and breaks the install itself. If your pnpmfile already has hooks, compose devfeedbackUpdateConfig(__dirname) from the same module into your own updateConfig.
Nx: every task, cache hits included (Nx ≥ 20.4). Register the plugin in nx.json; it uses postTasksExecution and needs no dependency on nx beyond the one you already have:
{ "plugins": ["agoda-devfeedback-common/nx"] }Scripts that Nx runs are recorded by the Nx plugin, not the wrapper, so nothing is counted twice. A pnpm build that runs nx run-many -t build gives one pnpm run build event for the whole run and one nx run build event per project, all with rootCommand: "pnpm run build".
What each event adds to the usual command fields:
| Field | Example | Meaning |
|---|---|---|
phase |
test |
From the script or target name, see below |
target |
test:unit |
Script or task-runner target name |
project |
@acme/web |
Workspace package or Nx project. projectName stays the repo |
runner |
pnpm, nx |
What ran it |
command |
pnpm run test:unit |
Built from the runner and the target |
rootCommand |
nx run-many -t build,test |
The top-level invocation, for nested pnpm run calls and Nx tasks |
outcome |
cache |
success, failure, aborted (signal or exit 130, so a script that exits 130 on purpose counts too), cache (Nx cache hit), skipped |
runId |
1481533057… |
Nx only: one per Nx invocation |
exitCode, signal |
130 and SIGINT for Ctrl-C |
Phases come from the name: typecheck/type-check/tsc/check-types → typecheck, lint/eslint/lint:* → lint, test/test:*/vitest/jest → test, build/build:* → build, codegen/generate/gen:* → codegen, anything else → other with the name in target. Override it per repo in the root package.json:
{
"devfeedback": {
"phases": { "e2e": "test", "openapi": "codegen" },
"skip": ["dev", "start", "serve", "watch", "storybook"]
}
}skip replaces the default list, which is dev, start, serve and watch: a long-running command's duration is the length of the session, not a measurement, and dev server time-to-ready already comes from the bundler plugins. A skipped script still passes DEVFEEDBACK_SESSION_ID on, so the dev server it starts joins the session. Nx continuous tasks are skipped for the same reason. Install lifecycle scripts (preinstall, postinstall, prepare, …) are part of the install span and are never recorded on their own, and neither are dependencies' own scripts.
What is never sent:
- No argv, no script bodies, no output.
command,targetandrootCommandare built only from the script name, the Nx project and target, the runner and a known Nx verb.npm_lifecycle_scriptand NxterminalOutputare never read. There is no free-text error field. - Nothing from CI. Both emitters do nothing when
CIis set, unlessDEVFEEDBACK_CI=true. The bundler plugins are unchanged.
And what the wrapper does not change: stdio, cwd, exit codes and signals pass straight through, SIGINT/SIGTERM/SIGHUP are forwarded to the script, and on pnpm 11.28.1+ and 12.8.0+ the wrapper adds back the SIGINT trap pnpm wraps Bourne-shell scripts in (pnpm drops it for a custom scriptShell), so Ctrl-C behaves exactly as it would without it. A CI job checks this against the newest pnpm. Events are handed to a detached process that adds the metadata, spools and delivers them after the script has returned; the wrapper itself never touches the network. It does cost one Node.js process start per pnpm run, and loading a pnpmfile has a cost of its own on pnpm 12, about 100 ms per command, since pnpm starts a worker for it.
In CI nothing is recorded, but pnpm 12 still pays for the pnpmfile worker. To skip it, set PNPM_CONFIG_IGNORE_PNPMFILE=true on the CI steps that run scripts, after pnpm install. Don't set it for the install itself if your pnpmfile has other hooks.
Known gaps:
- Windows: not wrapped. pnpm only accepts a real
.exeasscriptShell, and a Node.js bin is a.cmdshim there, so the pnpmfile hook leaves Windows alone. The Nx plugin works everywhere. - A repo that already sets
scriptShellkeeps it and is not wrapped. To wrap a non-default shell, setDEVFEEDBACK_SHELL=/bin/bashin the developer's environment. - With
shell-emulator=truepnpm ignoresscriptShell, so nothing is recorded. - npm and yarn scripts are not covered yet.
| Tier | Repo change | What you get |
|---|---|---|
Shared preset (e.g. @agoda/vite-config re-exporting viteBuildStatsPlugin()) |
none, just a version bump | everything below except exact install spans |
| Direct install | one plugin line | same |
| Exact install timing | two scripts lines + a direct devDependency |
true install span and cold-clone accuracy |
| npm repos, opt-in | one .npmrc line (timing=true) |
per-phase and per-package install breakdown |
| pnpm scripts | a one-line .pnpmfile.cjs + a direct devDependency |
every pnpm run: typecheck, lint, test, codegen, … |
| Nx | one nx.json plugin entry |
every Nx task, cache hits included |
| Advanced | DEVFEEDBACK_SESSION_ID in your shell |
precise session boundaries |
Telemetry that slows people down gets deleted from configs, so:
- Every POST has a 1500 ms timeout. Off-VPN, nothing hangs.
- Git metadata is read once and cached until the repository actually changes, instead of spawning three
gitprocesses per HMR event. - The session id is resolved once per process; the HMR path performs no synchronous filesystem writes and no process spawns.
- Signal handlers write synchronously and then get out of the way, so Ctrl-C behaves exactly as it would without the plugin.
- The spool is capped at 256 KB and events older than a week are dropped rather than accumulated.
- Startup chatter is behind
DEVFEEDBACK_DEBUG=1. At the default log level the lifecycle events print nothing at all.
We welcome contributions! Whether you're fixing bugs, improving documentation, or adding support for the next big JavaScript build tool, we appreciate your help in making agoda-devfeedback even better. Check out our Contributing Guide for more details on how to get started.
Remember, in the world of agoda-devfeedback, there are no stupid questions, only builds that are taking too long!
Remember, in JavaScript development, there are only two types of projects: those that are measuring their build times, and those that are still waiting for their builds to finish. With agoda-devfeedback, you'll always know exactly how long you're waiting. (Spoiler alert: with our help, it won't be long!)
Happy coding, and may your builds be ever faster! 🚀