Remove binary_component (#175)

Summary:
Pull Request resolved: #175

The diff replaces the usage of `binary_component` with directly instantiating `AppDef`

Reviewed By: kiukchung

Differential Revision: D30915931

fbshipit-source-id: 882dc8a4eebec7e7c40000ec6c5e5b05d654c57b

Author: aivanou

docs/source/components/base.rst

@@ -4,10 +4,6 @@ Base
 .. automodule:: torchx.components.base
 .. currentmodule:: torchx.components.base
 
-.. automodule:: torchx.components.base.binary_component
-.. currentmodule:: torchx.components.base.binary_component
-.. autofunction:: binary_component
-
 .. automodule:: torchx.components.base.roles
 .. currentmodule:: torchx.components.base.roles
 .. autofunction:: create_torch_dist_role

examples/apps/datapreproc/__init__.py

@@ -15,18 +15,15 @@
 from typing import Dict, Optional
 
 import torchx.specs as specs
-from torchx.components.base.binary_component import binary_component
 
 
 def data_preproc(
     image: str,
-    entrypoint: str,
     output_path: str,
+    entrypoint: str = "examples/apps/datapreproc/datapreproc.py",
     input_path: str = "http://cs231n.stanford.edu/tiny-imagenet-200.zip",
-    input_md5: str = "90528d7ca1a48142e341f4ef8d21d0de",
     env: Optional[Dict[str, str]] = None,
     resource: Optional[str] = None,
-    name: str = "datapreproc",
 ) -> specs.AppDef:
     """Data PreProc app.
 
@@ -37,10 +34,8 @@ def data_preproc(
         entrypoint: User script to launch
         output_path: Url-like path to save the processes compressed images
         input_path: Url-like path to fetch the imagenet dataset
-        input_md5: Hash of the input imagenet dataset file
         env: Env variables to transfer to the user script
         resource: String representation of the resource
-        name: Name of the worker
 
     Returns:
         specs.AppDef: Torchx AppDef
@@ -50,21 +45,23 @@ def data_preproc(
     args = [
         "--input_path",
         input_path,
-        "--input_md5",
-        input_md5,
         "--output_path",
         output_path,
     ]
     if resource:
         resource_def = specs.named_resources[resource]
     else:
         resource_def = specs.Resource(cpu=1, gpu=0, memMB=1024)
-
-    return binary_component(
-        name="datapreproc_role",
-        entrypoint=entrypoint,
-        args=args,
-        env=env,
-        image=image,
-        resource=resource_def,
+    return specs.AppDef(
+        name="datapreproc",
+        roles=[
+            specs.Role(
+                name="worker",
+                image=image,
+                entrypoint=entrypoint,
+                args=args,
+                env=env,
+                resource=resource_def,
+            )
+        ],
     )

examples/apps/datapreproc/component.py

@@ -14,7 +14,6 @@
 from typing import Optional, Dict
 
 import torchx.specs.api as torchx
-from torchx.components.base.binary_component import binary_component
 from torchx.specs import named_resources
 
 
@@ -59,15 +58,20 @@ def trainer(
     ]
     if skip_export:
         args.append("--skip_export")
-    return binary_component(
+    return torchx.AppDef(
         name="examples-lightning_classy_vision-trainer",
-        entrypoint=entrypoint,
-        args=args,
-        env=env,
-        image=image,
-        resource=named_resources[resource]
-        if resource
-        else torchx.Resource(cpu=1, gpu=0, memMB=1024),
+        roles=[
+            torchx.Role(
+                name="worker",
+                entrypoint=entrypoint,
+                args=args,
+                env=env,
+                image=image,
+                resource=named_resources[resource]
+                if resource
+                else torchx.Resource(cpu=1, gpu=0, memMB=1024),
+            )
+        ],
     )
 
 
@@ -88,19 +92,24 @@ def interpret(
         output_path: output path for model checkpoints (e.g. file:///foo/bar)
         resource: the resources to use
     """
-    return binary_component(
+    return torchx.AppDef(
         name="examples-lightning_classy_vision-interpret",
-        entrypoint="examples/apps/lightning_classy_vision/interpret.py",
-        args=[
-            "--load_path",
-            load_path,
-            "--data_path",
-            data_path,
-            "--output_path",
-            output_path,
+        roles=[
+            torchx.Role(
+                name="worker",
+                entrypoint="examples/apps/lightning_classy_vision/interpret.py",
+                args=[
+                    "--load_path",
+                    load_path,
+                    "--data_path",
+                    data_path,
+                    "--output_path",
+                    output_path,
+                ],
+                image=image,
+                resource=named_resources[resource]
+                if resource
+                else torchx.Resource(cpu=1, gpu=0, memMB=1024),
+            )
         ],
-        image=image,
-        resource=named_resources[resource]
-        if resource
-        else torchx.Resource(cpu=1, gpu=0, memMB=1024),
     )

examples/apps/dist_cifar/__init__.py

@@ -123,14 +123,12 @@
 # datapreproc outputs the data to a specified fsspec path. These paths are all
 # specified ahead of time so we have a fully static pipeline.
 
-from torchx.components.base.binary_component import binary_component
+
+from examples.apps.datapreproc.component import data_preproc
 
 processed_data_path: str = os.path.join(args.output_path, "processed")
-datapreproc_app: specs.AppDef = binary_component(
-    name="examples-datapreproc",
-    entrypoint="examples/apps/datapreproc/datapreproc.py",
-    args=["--input_path", data_path, "--output_path", processed_data_path],
-    image=args.image,
+datapreproc_app: specs.AppDef = data_preproc(
+    image=args.image, output_path=processed_data_path, input_path=data_path
 )
 
 # %%

examples/apps/lightning_classy_vision/__init__.py

@@ -28,20 +28,23 @@
 
 import kfp
 from torchx import specs
-from torchx.components.base.binary_component import binary_component
 from torchx.pipelines.kfp.adapter import container_from_app
 
 
 def pipeline() -> None:
-    # First we define our AppDef for the component. We use the binary_component
-    # helper for making single node components. AppDef is a core part of TorchX
+    # First we define our AppDef for the component. AppDef is a core part of TorchX
     # and can be used to describe complex distributed multi container apps or
     # just a single node component like here.
-    echo_app: specs.AppDef = binary_component(
+    echo_app: specs.AppDef = specs.AppDef(
         name="examples-intro",
-        entrypoint="/bin/echo",
-        args=["Hello TorchX!"],
-        image="alpine",
+        roles=[
+            specs.Role(
+                name="worker",
+                entrypoint="/bin/echo",
+                args=["Hello TorchX!"],
+                image="alpine",
+            )
+        ],
     )
 
     # To convert the TorchX AppDef into a KFP container we use