Paul Gschwendtner 53c4c8de8f test: ensure global options for benchmark tests can be set in bazel (#34753)
Previously, when the benchmark tests ran outside of Bazel, developers
had the posibility to control how the tests run through command line
options. e.g. `--dryrun`. This no longer works reliable in Bazel where
command line arguments are not passed to the text executable.

To make the global options still usable (as they could be still useful
in some cases), we just pass them through the Bazel `--test_env`. This
reduces the code we need to read the command line, but still preserves
the flexibility in a Bazel idiomatic way.

PR Close #34753
2020-01-29 09:22:27 -08:00

45 lines
1.6 KiB
Markdown

# How to run the benchmarks locally
## Run in the browser
```bash
yarn bazel run modules/benchmarks/src/tree/{name}:devserver
# e.g. "ng2" tree benchmark:
yarn bazel run modules/benchmarks/src/tree/ng2:devserver
```
## Run e2e tests
```
# Run e2e tests of individual applications:
yarn bazel test modules/benchmarks/src/tree/ng2/...
# Run all e2e tests:
yarn bazel test modules/benchmarks/...
```
## Use of *_aot.ts files
The `*_aot.ts` files are used as entry-points within Google to run the benchmark
tests. These are still built as part of the corresponding `ng_module` rule.
## Specifying benchmark options
There are options that can be specified in order to control how a given benchmark target
runs. The following options can be set through [test environment variables](https://docs.bazel.build/versions/master/command-line-reference.html#flag--test_env):
* `PERF_SAMPLE_SIZE`: Benchpress performs measurements until `scriptTime` predictively no longer
decreases. It does this by using a simple linear regression with the amount of samples specified.
Defaults to `20` samples.
* `PERF_FORCE_GC`: If set to `true`, `@angular/benchpress` will run run the garbage collector
before and after performing measurements. Benchpress will measure and report the garbage
collection time.
* `PERF_DRYRUN`: If set to `true`, no results are printed and stored in a `json` file. Also
benchpress only performs a single measurement (unlike with the simple linear regression).
Here is an example command that sets the `PERF_DRYRUN` option:
```bash
yarn bazel test modules/benchmarks/src/tree/baseline:perf --test_env=PERF_DRYRUN=true
```