|
| 1 | +--- |
| 2 | +jupyter: |
| 3 | + jupytext: |
| 4 | + text_representation: |
| 5 | + extension: .md |
| 6 | + format_name: markdown |
| 7 | + format_version: '1.1' |
| 8 | + jupytext_version: 1.1.0 |
| 9 | + kernelspec: |
| 10 | + display_name: Python 3 |
| 11 | + language: python |
| 12 | + name: python3 |
| 13 | +--- |
| 14 | + |
| 15 | +# Custom Components |
| 16 | + |
| 17 | +This is a guide on how to build a simple app and custom component spec |
| 18 | +and launch it via two different schedulers. |
| 19 | + |
| 20 | +See the [Quickstart Guide](quickstart.md) for installation and basic usage. |
| 21 | + |
| 22 | +## Hello World |
| 23 | + |
| 24 | +Lets start off with writing a simple "Hello World" python app. This is just a |
| 25 | +normal python program and can contain anything you'd like. |
| 26 | + |
| 27 | +<div class="admonition note"> |
| 28 | +<div class="admonition-title">Note</div> |
| 29 | +This example uses Jupyter Notebook `%%writefile` to create local files for |
| 30 | +example purposes. Under normal usage you would have these as standalone files. |
| 31 | +</div> |
| 32 | + |
| 33 | +```python |
| 34 | +%%writefile my_app.py |
| 35 | + |
| 36 | +import sys |
| 37 | +import argparse |
| 38 | + |
| 39 | +def main(user: str) -> None: |
| 40 | + print(f"Hello, {user}!") |
| 41 | + |
| 42 | +if __name__ == "__main__": |
| 43 | + parser = argparse.ArgumentParser( |
| 44 | + description="Hello world app" |
| 45 | + ) |
| 46 | + parser.add_argument( |
| 47 | + "--user", |
| 48 | + type=str, |
| 49 | + help="the person to greet", |
| 50 | + required=True, |
| 51 | + ) |
| 52 | + args = parser.parse_args(sys.argv[1:]) |
| 53 | + |
| 54 | + main(args.user) |
| 55 | +``` |
| 56 | + |
| 57 | +Now that we have an app we can write the component file for it. This |
| 58 | +function allows us to reuse and share our app in a user friendly way. |
| 59 | + |
| 60 | +We can use this component from the `torchx` cli or programmatically as part of a |
| 61 | +pipeline. |
| 62 | + |
| 63 | +```python |
| 64 | +%%writefile my_component.py |
| 65 | + |
| 66 | +import torchx.specs as specs |
| 67 | + |
| 68 | +def greet(user: str, image: str = "my_app:latest") -> specs.AppDef: |
| 69 | + return specs.AppDef( |
| 70 | + name="hello_world", |
| 71 | + roles=[ |
| 72 | + specs.Role( |
| 73 | + name="greeter", |
| 74 | + image=image, |
| 75 | + entrypoint="python", |
| 76 | + args=[ |
| 77 | + "-m", "my_app", |
| 78 | + "--user", user, |
| 79 | + ], |
| 80 | + ) |
| 81 | + ], |
| 82 | + ) |
| 83 | +``` |
| 84 | + |
| 85 | +We can execute our component via `torchx run`. The |
| 86 | +`local_cwd` scheduler executes the component relative to the current directory. |
| 87 | + |
| 88 | +```sh |
| 89 | +torchx run --scheduler local_cwd my_component.py:greet --user "your name" |
| 90 | +``` |
| 91 | + |
| 92 | +If we want to run in other environments, we can build a Docker container so we |
| 93 | +can run our component in Docker enabled environments such as Kubernetes or via |
| 94 | +the local Docker scheduler. |
| 95 | + |
| 96 | +<div class="admonition note"> |
| 97 | +<div class="admonition-title">Note</div> |
| 98 | +This requires Docker installed and won't work in environments such as Google |
| 99 | +Colab. If you have not done so already follow the install instructions on: |
| 100 | +[https://docs.docker.com/get-docker/](https://docs.docker.com/get-docker/)</a> |
| 101 | +</div> |
| 102 | + |
| 103 | +```python |
| 104 | +%%writefile Dockerfile.custom |
| 105 | + |
| 106 | +FROM ghcr.io/pytorch/torchx:0.1.0rc1 |
| 107 | + |
| 108 | +ADD my_app.py . |
| 109 | +``` |
| 110 | + |
| 111 | +Once we have the Dockerfile created we can create our docker image. |
| 112 | + |
| 113 | +```sh |
| 114 | +docker build -t my_app:latest -f Dockerfile.custom . |
| 115 | +``` |
| 116 | + |
| 117 | +We can then launch it on the local scheduler. |
| 118 | + |
| 119 | +```sh |
| 120 | +torchx run --scheduler local_docker my_component.py:greet --image "my_app:latest" --user "your name" |
| 121 | +``` |
| 122 | + |
| 123 | +If you have a Kubernetes cluster you can use the [Kubernetes scheduler](schedulers/kubernetes.rst) to launch |
| 124 | +this on the cluster instead. |
| 125 | + |
| 126 | + |
| 127 | +<!-- #md --> |
| 128 | +```sh |
| 129 | +$ docker push my_app:latest |
| 130 | +$ torchx run --scheduler kubernetes my_component.py:greet --image "my_app:latest" --user "your name" |
| 131 | +``` |
| 132 | +<!-- #endmd --> |
| 133 | + |
| 134 | + |
| 135 | +## Builtins |
| 136 | + |
| 137 | +TorchX also provides a number of builtin components with premade images. You can discover |
| 138 | +them via: |
| 139 | + |
| 140 | +```sh |
| 141 | +torchx builtins |
| 142 | +``` |
| 143 | + |
| 144 | +You can use these either from the CLI, from a pipeline or programmatically like |
| 145 | +you would any other component. |
| 146 | + |
| 147 | +```sh |
| 148 | +torchx run utils.echo --msg "Hello :)" |
| 149 | +``` |
0 commit comments