Skip to content

intro.md

Latest commit

 

History

History
97 lines (67 loc) · 7.96 KB

intro.md

File metadata and controls

97 lines (67 loc) · 7.96 KB
sidebar_position 1

WasmEdge Plug-in System Introduction

While the WasmEdge language SDKs allow registering host functions from a host (wrapping) application, developers should implement the host functions before compilation. However, for a more flexible and dynamic extension of the host functions, WasmEdge provides a plug-in architecture to load the plug-in shared library.

A WasmEdge plug-in is a software component that extends the functionality of the WasmEdge runtime. Currently, developers can follow the guides to implement the plug-ins in C API (recommended), C++ and Rust. With the help of the WasmEdge SDKs in the supported languages, developers can load and register the host functions from the plug-in shared libraries, allowing them to seamlessly integrate the plug-ins into the WasmEdge runtime as if they were part of the core runtime.

graph LR
    A((Host Application)) -- Loads --> B((Plug-in Shared Library))
    B -- Registers --> C((Wasmedge Runtime))
Loading

Benefits of Using WasmEdge Plug-in

WasmEdge plug-ins are designed to extend the functionality of the WasmEdge runtime and can be helpful for developers and end-users in several ways:

  • Customization: WasmEdge plug-ins can be customized to suit the specific needs of a project. Developers can create plug-ins that integrate with other systems or tools or provide unique functionality unavailable in the core WasmEdge runtime.

  • Performance: WasmEdge plug-ins are designed to work seamlessly with the core runtime, minimizing overhead and maximizing performance, which means they can provide additional functionality without sacrificing performance.

  • Ease of use: WasmEdge plug-ins are easy to use and integrate with the WasmEdge runtime. Developers can load the plug-in into the runtime and use its functions as part of the core runtime.

  • Scalability: Developers can compile their compute-intensive functions into host functions and package them into a plug-in to provide the better performance as running in native code.

WasmEdge plug-ins can provide developers and users with a versatile, customizable, high-performance, and secure way to extend the functionality of the WasmEdge runtime. WasmEdge plug-ins can also improve scalability and ease of use, making it easier to build and deploy complex applications on edge devices.

Loadable Plug-in

Loadable plug-ins are standalone shared libraries (.so/.dylib/.dll files) that the WasmEdge runtime environment can load at runtime. These plug-ins can provide additional functionality to the WasmEdge runtime environment, such as new modules that can be imported by WebAssembly modules.

Creating Loadable Plug-in

To create a loadable plug-in for WasmEdge, developers can use the WasmEdge Plug-in SDK, which provides a set of Rust, C, and C++ APIs for creating and registering plug-ins. The SDK also includes example code that demonstrates creating a simple plug-in that returns a string. By following the provided examples and leveraging the SDK's APIs, developers can quickly build custom plug-ins tailored to their specific needs.

Loading plug-in from paths

To use the loadable plug-ins, developers need to load them from specific paths into the WasmEdge runtime environment. The loading process involves the following steps:

  • Loadable plug-ins can be loaded from default paths by calling the WasmEdge_PluginLoadWithDefaultPaths() API. The default paths include:

    • The path specified in the environment variable WASMEDGE_PLUGIN_PATH.
    • The ./plugin/ directory relative to the WasmEdge installation path.
    • The ./wasmedge/ directory is located under the library path if WasmEdge is installed in a system directory such as /usr and /usr/local.

  • If the plug-ins are located in a specific path or directory, developers can use the WasmEdge_PluginLoadFromPath("PATH_TO_PLUGIN/plug-in.so") API to load the plug-ins from that particular location.

The WasmEdge runtime environment will search for the loadable plug-ins in the specified paths and load them if found.

The following flowchart shows the process of loading loadable plug-ins into the WasmEdge runtime environment from specific paths:

graph LR
    A((Start)) --> B(Loadable Plug-ins)
    B --> C{Load Plug-ins}
    C --> D[Load from Default Paths]
    C --> E[Load from Specific Path]
    C --> F[Load from Specific Directory]
    D --> G{Is Plug-in Found?}
    E --> G
    F --> G
    G -- Yes --> H(Load Plug-in)
    H --> I(End)
    G -- No --> I
Loading

The flowchart shows loading loadable plug-ins into the WasmEdge runtime environment. The process involves searching for plug-ins in default paths, a specific path, or a specific directory. If a plug-in is found in any of these locations, it is loaded into the runtime environment. The flowchart enables developers to quickly load plug-ins and extend the capabilities of the WasmEdge runtime.

By following this flowchart, developers can effectively load loadable plug-ins into the WasmEdge runtime environment from specific paths, expanding the runtime's functionality according to their requirements.

WasmEdge Currently Released Plug-ins

There are several plug-in releases with the WasmEdge official releases. Please check the following table to check the release status and how to build from the source with the plug-ins.

Plug-in Rust Crate Released Platforms Build Steps
WasmEdge-Process wasmedge_process_interface manylinux2014 x86_64, manylinux2014 aarch64, and ubuntu 20.04 x86_64 (since 0.10.0) Build Wtih WasmEdge-Process
WASI-Crypto wasi-crypto manylinux2014 x86_64, manylinux2014 aarch64, and ubuntu 20.04 x86_64 (since 0.10.1) Build With WASI-Crypto
WASI-NN with OpenVINO backend wasi-nn ubuntu 20.04 x86_64 (since 0.10.1) Build With WASI-NN
WASI-NN with PyTorch backend wasi-nn ubuntu 20.04 x86_64 (since 0.11.1) Build With WASI-NN
WASI-NN with TensorFlow-Lite backend wasi-nn manylinux2014 x86_64, manylinux2014 aarch64, and ubuntu 20.04 x86_64 (since 0.11.2) Build With WASI-NN
WasmEdge-Image wasmedge_tensorflow_interface manylinux2014 x86_64, manylinux2014 aarch64, ubuntu 20.04 x86_64, darwin x86_64, and darwin arm64 (since 0.13.0) Build With WasmEdge-Image
WasmEdge-Tensorflow wasmedge_tensorflow_interface manylinux2014 x86_64, manylinux2014 aarch64, ubuntu 20.04 x86_64, darwin x86_64, and darwin arm64 (since 0.13.0) Build With WasmEdge-Tensorflow
WasmEdge-TensorflowLite wasmedge_tensorflow_interface manylinux2014 x86_64, manylinux2014 aarch64, ubuntu 20.04 x86_64, darwin x86_64, and darwin arm64 (since 0.13.0) Build With WasmEdge-TensorflowLite

:::note Due to the OpenVINO dependency, we only release the WASI-NN plug-in for the OpenVINO backend on Ubuntu 20.04 x86_64 now. We'll work with manylinux2014 versions in the future. :::