angular/modules/ssr-benchmarks/README.md
Matthieu Riegler 84b6896956 refactor(platform-server): Add an ssr benchmark setup. (#57647)
In order to investigate the performances of SSR, this commit introduces a benchmark suite which will measure several step of the rendering.

PR Close #57647
2024-10-04 10:45:22 -07:00

60 lines
No EOL
4 KiB
Markdown

## Intro
This small benchmark suite is dedicated to mesure & describe how compute time is spent when rendering an application like in SSR.
## Struture
* `./main.ts` is the entry point to run the benchmark
* `./src` contains a sample app that exports a `render` function.
* This app render a table of variable size, which depends on data (`initData()`)
* This app is then rendered X numbers of times
* Individual function calls are measured with `startMeasuring()`/`stopMeasuring()` from the core package.
* If you add a new measure, make sure to add it also to the `levels` map for it to be represented correctly in the result
## Build & run
`yarn bazel run //modules/ssr-benchmarks:run`
### Running the benchmark in a browser environment
`yarn bazel run //modules/ssr-benchmarks:run_browser`
This bazel target will build the benchmark, start a http-server with a html that will load the benckmark script.
The benchmark script with this target will have DOM Emulation disabled.
The result will be visible in the devtools console.
Note: Due to the CLI adding some polyfills, @angular/build is patched to disable DOM emulation and running server code inside a browser:
1. removing an import from `node:module` in `polyfills.server.mjs` (with `tail ...`)
2. removing the import of `platform-server/init`.
To run create a usable flame chart, prepare a narrowed run (like `benchmarkRun(10000, 20);`).
Then in the performance tab of the devtools, trigger "Record & Reload" to generate a profile.
### Deopt Explorer
A target is dedicated to generate a v8 log that can be fed to the [Deopt Explorer extension](https://github.com/microsoft/deoptexplorer-vscode).
1. Run `yarn bazel run //modules/ssr-benchmarks:run_deopt`,
2. open the project generated at the path after `Successfully ran all commands in test directory:`,
3. open the logfile in the extension
## Result example
=== table with 10000 rows, with 1000 renders ===
┌─────────┬──────────────────────────────────────┬──────────┬──────────┬────────────┬───────────┐
│ (index) │ name │ min │ average │ percentage │ max │
├─────────┼──────────────────────────────────────┼──────────┼──────────┼────────────┼───────────┤
│ 0 │ ' renderApplication ' │ '77.0ms' │ '86.4ms' │ '100.0%' │ '259.2ms' │
│ 1 │ ' └ createServerPlatform ' │ '0.0ms' │ '0.1ms' │ '0.1%' │ '3.7ms' │
│ 2 │ ' └ bootstrap ' │ '35.9ms' │ '42.6ms' │ '49.3%' │ '138.4ms' │
│ 3 │ ' └ _render ' │ '39.7ms' │ '43.8ms' │ '50.7%' │ '124.9ms' │
│ 4 │ ' └ whenStable ' │ '0.0ms' │ '0.0ms' │ '0.0%' │ '0.0ms' │
│ 5 │ ' └ prepareForHydration ' │ '13.1ms' │ '14.8ms' │ '17.1%' │ '53.4ms' │
│ 6 │ ' └ insertEventRecordScript ' │ '0.0ms' │ '0.0ms' │ '0.0%' │ '0.0ms' │
│ 7 │ ' └ serializeTransferStateFactory' │ '0.0ms' │ '0.0ms' │ '0.0%' │ '0.1ms' │
│ 8 │ ' └ renderToString ' │ '7.3ms' │ '8.9ms' │ '10.3%' │ '41.8ms' │
└─────────┴──────────────────────────────────────┴──────────┴──────────┴────────────┴───────────┘
Note: The max measure is often an outlier of the first few measures, probably before the JIT optimisation happens