vitest/docs/guide/profiling-test-performance.md

215 lines
8.2 KiB
Markdown

# Profiling Test Performance
When you run Vitest it reports multiple time metrics of your tests:
> ```bash
> RUN v2.1.1 /x/vitest/examples/profiling
>
> ✓ test/prime-number.test.ts (1) 4517ms
> ✓ generate prime number 4517ms
>
> Test Files 1 passed (1)
> Tests 1 passed (1)
> Start at 09:32:53
> Duration 4.80s (transform 44ms, setup 0ms, collect 35ms, tests 4.52s, environment 0ms, prepare 81ms)
> # Time metrics ^^
> ```
- Transform: How much time was spent transforming the files. See [File Transform](#file-transform).
- Setup: Time spent for running the [`setupFiles`](/config/#setupfiles) files.
- Collect: Time spent for collecting all tests in the test files. This includes the time it took to import all file dependencies.
- Tests: Time spent for actually running the test cases.
- Environment: Time spent for setting up the test [`environment`](/config/#environment), for example JSDOM.
- Prepare: Time Vitest uses to prepare the test runner. When running tests in Node, this is the time to import and execute all internal utilities inside the worker. When running tests in the browser, this also includes the time to initiate the iframe.
## Test runner
In cases where your test execution time is high, you can generate a profile of the test runner. See NodeJS documentation for following options:
- [`--cpu-prof`](https://nodejs.org/api/cli.html#--cpu-prof)
- [`--heap-prof`](https://nodejs.org/api/cli.html#--heap-prof)
- [`--prof`](https://nodejs.org/api/cli.html#--prof)
:::warning
The `--prof` option does not work with `pool: 'threads'` due to `node:worker_threads` limitations.
:::
To pass these options to Vitest's test runner, define `poolOptions.<pool>.execArgv` in your Vitest configuration:
::: code-group
```ts [Forks]
import { defineConfig } from 'vitest/config'
export default defineConfig({
test: {
pool: 'forks',
poolOptions: {
forks: {
execArgv: [
'--cpu-prof',
'--cpu-prof-dir=test-runner-profile',
'--heap-prof',
'--heap-prof-dir=test-runner-profile'
],
// To generate a single profile
singleFork: true,
},
},
},
})
```
```ts [Threads]
import { defineConfig } from 'vitest/config'
export default defineConfig({
test: {
pool: 'threads',
poolOptions: {
threads: {
execArgv: [
'--cpu-prof',
'--cpu-prof-dir=test-runner-profile',
'--heap-prof',
'--heap-prof-dir=test-runner-profile'
],
// To generate a single profile
singleThread: true,
},
},
},
})
```
:::
After the tests have run there should be a `test-runner-profile/*.cpuprofile` and `test-runner-profile/*.heapprofile` files generated. See [Inspecting profiling records](#inspecting-profiling-records) for instructions how to analyze these files.
See [Profiling | Examples](https://github.com/vitest-dev/vitest/tree/main/examples/profiling) for example.
## Main thread
Profiling main thread is useful for debugging Vitest's Vite usage and [`globalSetup`](/config/#globalsetup) files.
This is also where your Vite plugins are running.
:::tip
See [Performance | Vite](https://vitejs.dev/guide/performance.html) for more tips about Vite specific profiling.
We recommend [`vite-plugin-inspect`](https://github.com/antfu-collective/vite-plugin-inspect) for profiling your Vite plugin performance.
:::
To do this you'll need to pass arguments to the Node process that runs Vitest.
```bash
$ node --cpu-prof --cpu-prof-dir=main-profile ./node_modules/vitest/vitest.mjs --run
# ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ^^^^^
# NodeJS arguments Vitest arguments
```
After the tests have run there should be a `main-profile/*.cpuprofile` file generated. See [Inspecting profiling records](#inspecting-profiling-records) for instructions how to analyze these files.
## File transform
In cases where your test transform and collection time is high, you can use `DEBUG=vite-node:*` environment variable to see which files are being transformed and executed by `vite-node`.
```bash
$ DEBUG=vite-node:* vitest --run
RUN v2.1.1 /x/vitest/examples/profiling
vite-node:server:request /x/vitest/examples/profiling/global-setup.ts +0ms
vite-node:client:execute /x/vitest/examples/profiling/global-setup.ts +0ms
vite-node:server:request /x/vitest/examples/profiling/test/prime-number.test.ts +45ms
vite-node:client:execute /x/vitest/examples/profiling/test/prime-number.test.ts +26ms
vite-node:server:request /src/prime-number.ts +9ms
vite-node:client:execute /x/vitest/examples/profiling/src/prime-number.ts +9ms
vite-node:server:request /src/unnecessary-file.ts +6ms
vite-node:client:execute /x/vitest/examples/profiling/src/unnecessary-file.ts +4ms
...
```
This profiling strategy is a good way to identify unnecessary transforms caused by [barrel files](https://vitejs.dev/guide/performance.html#avoid-barrel-files).
If these logs contain files that should not be loaded when your test is run, you might have barrel files that are importing files unnecessarily.
You can also use [Vitest UI](/guide/ui) to debug slowness caused by barrel file.
The example below shows how importing files without barrel file reduces amount of transformed files by ~85%.
::: code-group
``` [File tree]
├── src
│ └── utils
│ ├── currency.ts
│ ├── formatters.ts <-- File to test
│ ├── index.ts
│ ├── location.ts
│ ├── math.ts
│ ├── time.ts
│ └── users.ts
├── test
│ └── formatters.test.ts
└── vitest.config.ts
```
```ts [example.test.ts]
import { expect, test } from 'vitest'
import { formatter } from '../src/utils' // [!code --]
import { formatter } from '../src/utils/formatters' // [!code ++]
test('formatter works', () => {
expect(formatter).not.toThrow()
})
```
:::
<img src="/module-graph-barrel-file.png" alt="Vitest UI demonstrating barrel file issues" />
To see how files are transformed, you can use `VITE_NODE_DEBUG_DUMP` environment variable to write transformed files in the file system:
```bash
$ VITE_NODE_DEBUG_DUMP=true vitest --run
[vite-node] [debug] dump modules to /x/examples/profiling/.vite-node/dump
RUN v2.1.1 /x/vitest/examples/profiling
...
$ ls .vite-node/dump/
_x_examples_profiling_global-setup_ts-1292904907.js
_x_examples_profiling_test_prime-number_test_ts-1413378098.js
_src_prime-number_ts-525172412.js
```
## Code coverage
If code coverage generation is slow on your project you can use `DEBUG=vitest:coverage` environment variable to enable performance logging.
```bash
$ DEBUG=vitest:coverage vitest --run --coverage
RUN v3.1.1 /x/vitest-example
vitest:coverage Reading coverage results 2/2
vitest:coverage Converting 1/2
vitest:coverage 4 ms /x/src/multiply.ts
vitest:coverage Converting 2/2
vitest:coverage 552 ms /x/src/add.ts
vitest:coverage Uncovered files 1/2
vitest:coverage File "/x/src/large-file.ts" is taking longer than 3s # [!code error]
vitest:coverage 3027 ms /x/src/large-file.ts
vitest:coverage Uncovered files 2/2
vitest:coverage 4 ms /x/src/untested-file.ts
vitest:coverage Generate coverage total time 3521 ms
```
This profiling approach is great for detecting large files that are accidentally picked by coverage providers.
For example if your configuration is accidentally including large built minified Javascript files in code coverage, they should appear in logs.
In these cases you might want to adjust your [`coverage.include`](/config/#coverage-include) and [`coverage.exclude`](/config/#coverage-exclude) options.
## Inspecting profiling records
You can inspect the contents of `*.cpuprofile` and `*.heapprofile` with various tools. See list below for examples.
- [Speedscope](https://www.speedscope.app/)
- [Performance Profiling JavaScript in Visual Studio Code](https://code.visualstudio.com/docs/nodejs/profiling#_analyzing-a-profile)
- [Profile Node.js performance with the Performance panel | developer.chrome.com](https://developer.chrome.com/docs/devtools/performance/nodejs#analyze)
- [Memory panel overview | developer.chrome.com](https://developer.chrome.com/docs/devtools/memory-problems/heap-snapshots#view_snapshots)