Skip to main content

Composability System

Summary

This page describes the composability system that helps bring together content from a multitude of plugins into one Backstage application.

The core principle of the composability system is that plugins should have clear boundaries and connections. It should isolate crashes within a plugin, but allow navigation between them. It should allow for plugins to be loaded only when needed, and enable plugins to provide extension points for other plugins to build upon. The composability system is also built with an app-first mindset, prioritizing simplicity and clarity in the app over that in the plugins and core APIs.

The composability system isn't a single API surface. It is a collection of patterns, primitives, and APIs. At the core is the concept of extensions, which are exported by plugins for use in the app. There is also a primitive called component data, which helps keep the structure of the app more declarative. There are also RouteRefs that help route between pages in a flexible way, which is especially important when bringing together different open source plugins.

Concepts

This section is a brief look into all the concepts that help support the composability system.

Component Data

Component data is a composability primitive that is introduced as a way to provide a new data dimension for React components. Data is attached to React components using a key, and is then readable from any JSX elements created with those components, using the same key, as illustrated by the following example:

const MyComponent = () => <h1>This is my component</h1>;
attachComponentData(MyComponent, 'my.data', 5);

const element = <MyComponent />;
const myData = getComponentData(element, 'my.data');
// myData === 5

The purpose of component data is to provide a method for embedding data that can be inspected before rendering elements. Element inspection is a pattern that is quite common among React libraries, and used for example by react-router and material-ui to discover properties of the child elements before rendering. Although in those libraries only the element type and props are typically inspected, while our component data adds more structured access and simplifies evolution by allowing for multiple different versions of a piece of data to be used and interpreted at once.

One of the use-cases for component data is to support route and plugin discovery through elements in the app. Through this we allow for the React element tree in the app to be the source of truth, both for which plugins are used, as well as all top-level plugin routes in the app. The use of component data is not limited to these use-cases though, as it can be used as a primitive to create new abstractions as well.

Extensions

Extensions are what plugins export for use in an app. Most typically they are React components, but in practice they can be any kind of JavaScript value. They are created using create*Extension functions, and wrapped with plugin.provide() in order to create the actual exported extension.

The extension type is a simple one:

export type Extension<T> = {
expose(plugin: BackstagePlugin): T;
};

The power of extensions comes from the ability of various actors to hook into their usage. The creation and plugin wrapping is controlled by whoever owns the creation function, the Backstage core is able to hook into the process of exposing the extension outside the plugin, and in the end the app controls the usage of the extension.

The Backstage core API currently provides two different types of extension creators, createComponentExtension, and createRoutableExtension. Component extensions are plain React component with no particular requirements, for example a card for an entity overview page. The component will be exported more or less as is, but is wrapped to provide things like an error boundary, lazy loading, and a plugin context.

Routable extensions build on top of component extensions and are used for any component that should be rendered at a specific route path, such as top-level pages or entity page tab content. When creating a routable extension you need to supply a RouteRef as mountPoint. The mount point will be the handle of the component for the outside world, and is used by other components and plugins that wish to link to the routable component.

As of now there are only two extension creation functions in the core library, but more may be added in the future. There are also some plugins that provide ways to extend functionality through their own extensions, like createScaffolderFieldExtension from @backstage/plugin-scaffolder. Extensions are also not tied to React, and can both be used to model generic JavaScript concepts, as well as potentially bridge to rendering libraries and web frameworks other than React.

Extensions from a Plugin's Point of View

Extensions are one of the primary methods to traverse the plugin boundary, and the way that plugins provide concrete content for use within an app. They replace existing component export concepts such as Router or *Cards for display on entity overview pages.

It is recommended to create the exported extensions either in the top-level plugin.ts file, or in a dedicated extensions.ts (or .tsx) file. That file should not contain the bulk of the implementation though, and in fact, if the extension is a React component it is recommended to lazy-load the actual component. Component extensions support lazy loading out of the box using the lazy component declaration, for example:

export const EntityFooCard = plugin.provide(
createComponentExtension({
component: {
lazy: () => import('./components/FooCard').then(m => m.FooCard),
},
}),
);

Routable extensions even enforce lazy loading, as it is the only way to provide a component:

export const FooPage = plugin.provide(
createRoutableExtension({
name: 'FooPage',
component: () => import('./components/FooPage').then(m => m.FooPage),
mountPoint: fooPageRouteRef,
}),
);

Using Extensions in an App

Right now all extensions are modelled as React components. The usage of these extension is like regular usage of any React components, with one important difference. Extensions must all be part of a single React element tree spanning from the root AppProvider.

For example, the following app code does NOT work:

const AppRoutes = () => (
<Routes>
<Route path="/foo" element={<FooPage />} />
<Route path="/bar" element={<BarPage />} />
</Routes>
);

const App = () => (
<AppProvider>
<AppRouter>
<Root>
<AppRoutes />
</Root>
</AppRouter>
</AppProvider>
);

But in this case it is simple to fix! Simply be sure to not create any intermediate components in the app, for example like this:

const appRoutes = (
<Routes>
<Route path="/foo" element={<FooPage />} />
<Route path="/bar" element={<BarPage />} />
</Routes>
);

const App = () => (
<AppProvider>
<AppRouter>
<Root>{appRoutes}</Root>
</AppRouter>
</AppProvider>
);

Naming Patterns

There are a couple of naming patterns to adhere to as you build plugins, which helps clarify the intent and usage of the exports.

DescriptionPatternExamples
Top-level Pages*PageCatalogIndexPage, SettingsPage, LighthousePage
Entity Tab ContentEntity*ContentEntityJenkinsContent, EntityKubernetesContent
Entity Overview CardEntity*CardEntitySentryCard, EntityPagerDutyCard
Entity Conditionalis*AvailableisPagerDutyAvailable, isJenkinsAvailable
Plugin Instance*PluginjenkinsPlugin, catalogPlugin
Utility API Reference*ApiRefconfigApiRef, catalogApiRef

Routing System

The routing system of Backstage relies heavily on the composability system. It uses RouteRefs to represent routing targets in the app, which at runtime will be bound to a concrete path, but provides a level of indirection to help mix together different plugins that otherwise wouldn't know how to route to each other.

The concrete path for each RouteRef is discovered based on the element tree in the app. Let's consider the following example:

const appRoutes = (
<Routes>
<Route path="/foo" element={<FooPage />} />
<Route path="/bar" element={<BarPage />} />
</Routes>
);

We'll assume that FooPage and BarPage are routable extensions, exported by the fooPlugin and barPlugin respectively. Since the FooPage is a routable extension it has a RouteRef assigned as its mount point, which we'll refer to as fooPageRouteRef.

Given the above example, the fooPageRouteRef will be associated with the '/foo' route. If we want to route to the FooPage, we can use the useRouteRef hook to create a concrete link to the page. The useRouteRef hook takes a single RouteRef as its only parameter, and returns a function that is called to create the URL. For example like this:

const MyComponent = () => {
const fooRoute = useRouteRef(fooPageRouteRef);
return <a href={fooRoute()}>Link to Foo</a>;
};

Now let's assume that we want to link from the BarPage to the FooPage. We don't want to reference the fooPageRouteRef directly from our barPlugin, since that would create an unnecessary dependency on the fooPlugin. It would also provided little flexibility in allowing the app to tie plugins together, with the links instead being dictated by the plugins themselves. To solve this, we use ExternalRouteRefs. Much like regular route references, they can be passed to useRouteRef to create concrete URLs, but they can not be used as mount points in routable component and instead have to be associated with a target route using route bindings in the app.

We create a new ExternalRouteRef inside the barPlugin, using a neutral name that describes its role in the plugin rather than a specific plugin page that it might be linking to, allowing the app to decide the final target. If the BarPage for example wants to link to an external page in the header, it might declare an ExternalRouteRef similar to this:

const headerLinkRouteRef = createExternalRouteRef({ id: 'header-link' });

Binding External Routes in the App

The association of external routes is controlled by the app. Each ExternalRouteRef of a plugin should be bound to an actual RouteRef, usually from another plugin. The binding process happens once at app startup, and is then used through the lifetime of the app to help resolve concrete route paths.

Using the above example of the BarPage linking to the FooPage, we might do something like this in the app:

createApp({
bindRoutes({ bind }) {
bind(barPlugin.externalRoutes, {
headerLink: fooPlugin.routes.root,
});
},
});

Given the above binding, using useRouteRef(headerLinkRouteRef) within the barPlugin will let us create a link to whatever path the FooPage is mounted at.

Note that we are not importing and using the RouteRefs directly in the app, and instead rely on the plugin instance to access routes of the plugins. This is a new convention that was introduced to provide better namespacing and discoverability of routes, as well as reduce the number of separate exports from each plugin package. The route references would be supplied to createPlugin like this:

// In foo-plugin
export const fooPlugin = createPlugin({
routes: {
root: fooPageRouteRef,
},
...
})

// In bar-plugin
export const barPlugin = createPlugin({
externalRoutes: {
headerLink: headerLinkRouteRef,
},
...
})

Also note that you almost always want to create the route references themselves in a different file than the one that creates the plugin instance, for example a top-level routes.ts. This is to avoid circular imports when you use the route references from other parts of the same plugin.

Another thing to note is that this indirection in the routing is particularly useful for open source plugins that need to leave flexibility in how they are integrated. For plugins that you build internally for your own Backstage application, you can choose to go the route of direct imports or even use concrete routes directly. Although there can be some benefits to using the full routing system even in internal plugins. It can help you structure your routes, and as you will see further down it also helps you manage route parameters.

Optional External Routes

When creating an ExternalRouteRef it is possible to mark it as optional:

const headerLinkRouteRef = createExternalRouteRef({
id: 'header-link',
optional: true,
});

An external route that is marked as optional is not required to be bound in the app, allowing it to be used as a switch for whether a particular link should be displayed or action should be taken.

When calling useRouteRef with an optional external route, its return signature is changed to RouteFunc | undefined, allowing for logic like this:

const MyComponent = () => {
const headerLink = useRouteRef(headerLinkRouteRef);

return (
<header>
My Header
{headerLink && <a href={headerLink()}>External Link</a>}
</header>
);
};

Parameterized Routes

A feature of RouteRefs is the possibility of adding named and typed parameters. Parameters are declared at creation, and will enforce presence of the parameters in the path in the app, and require them as a parameter when using useRouteRef.

The following is an example of creation and usage of a parameterized route:

// Creation of a parameterized route
const myRouteRef = createRouteRef({
id: 'myroute',
params: ['name']
})

// In the app, where MyPage is a routable extension with myRouteRef set as mountPoint
<Route path='/my-page/:name' element={<MyPage />}/>

// Usage within a component
const myRoute = useRouteRef(myRouteRef)
return (
<div>
<a href={myRoute({name: 'a'})}>A</a>
<a href={myRoute({name: 'b'})}>B</a>
</div>
)

It is currently not possible to have parameterized ExternalRouteRefs, or to bind an external route to a parameterized route, although this may be added in the future if needed.

Subroutes

The last kind of route refs that can be created are SubRouteRefs, which can be used to create a route ref with a fixed path relative to an absolute RouteRef. They are useful if you have a page that internally is mounted at a sub route of a routable extension component, and you want other plugins to be able to route to that page.

For example:

// routes.ts
const rootRouteRef = createRouteRef({ id: 'root' });
const detailsRouteRef = createSubRouteRef({
id: 'root-sub',
parent: rootRouteRef,
path: '/details',
});

// plugin.ts
export const myPlugin = createPlugin({
routes: {
root: rootRouteRef,
details: detailsRouteRef,
},
});

export const MyPage = myPlugin.provide(
createRoutableExtension({
name: 'MyPage',
component: () => import('./components/MyPage').then(m => m.MyPage),
mountPoint: rootRouteRef,
}),
);

// components/MyPage.tsx
const MyPage = () => (
<Routes>
{/* myPlugin.routes.root will take the user to this page */}
<Route path="/" element={<IndexPage />} />

{/* myPlugin.routes.details will take the user to this page */}
<Route path="/details" element={<DetailsPage />} />
</Routes>
);

Catalog Components

To help structure the catalog entity pages in your app and choose what content to render in different scenarios, the @backstage/catalog plugin provides an EntitySwitch component. It works by selecting at most one element to render using a list of EntitySwitch.Case children.

For example, if you want all entities of kind "Template" to be rendered with a MyTemplate component, and all other entities to be rendered with a MyOther component, you would do the following:

<EntitySwitch>
<EntitySwitch.Case if={isKind('template')}>
<MyTemplate />
</EntitySwitch.Case>

<EntitySwitch.Case>
<MyOther />
</EntitySwitch.Case>
</EntitySwitch>

// Shorter form if desired:
<EntitySwitch>
<EntitySwitch.Case if={isKind('template')} children={<MyTemplate />}/>
<EntitySwitch.Case children={<MyOther />}/>
</EntitySwitch>

The EntitySwitch component will render the children of the first EntitySwitch.Case that returns true when the selected entity is passed to the function of the if prop. If none of the cases match, no children will be rendered, and if a case doesn't specify an if filter function, it will always match. The if property is simply a function of the type (entity: Entity) => boolean, for example, isKind can be implemented like this:

function isKind(kind: string) {
return (entity: Entity) => entity.kind.toLowerCase() === kind.toLowerCase();
}

The @backstage/catalog plugin provides a couple of built-in conditions, isKind, isComponentType, isResourceType, isEntityWith, and isNamespace.

In addition to the EntitySwitch component, the catalog plugin also exports a new EntityLayout component. It is a tweaked version and replacement for the EntityPageLayout component, and is introduced more in depth in the app migration section below.

NOTE: The rest of this documentation covers how to migrate older applications to the new composability system described above.

Porting Existing Plugins

There are a couple of high-level steps to porting an existing plugin to the new composability system:

  • Remove usage of router.addRoute or router.registerRoute within createPlugin, and export the page components as routable extensions instead.
  • Switch any Router export to instead be a routable extension.
  • Change any plain component exports, such as catalog overview cards, to be component extensions.
  • Stop exporting RouteRefs and instead pass them to createPlugin.
  • Stop accepting RouteRefs as props or importing them from other plugins, instead create an ExternalRouteRef as a replacement, and pass it to createPlugin.
  • Rename any other exported symbols according to the naming pattern table below.

Note that removing the existing exports and configuration is a breaking change in any plugin. If backwards compatibility is needed the existing code be deprecated while making the new additions, to then be removed at a later point.

Naming Patterns

Many export naming patterns have been changed to avoid import aliases and to clarify intent. Refer to the following table to formulate the new name:

DescriptionExisting PatternNew PatternExamples
Top-level PagesRouter\*PageCatalogIndexPage, SettingsPage, LighthousePage
Entity Tab ContentRouterEntity\*ContentEntityJenkinsContent, EntityKubernetesContent
Entity Overview Card\*CardEntity\*CardEntitySentryCard, EntityPagerDutyCard
Entity ConditionalisPluginApplicableToEntityis\*AvailableisPagerDutyAvailable, isJenkinsAvailable
Plugin Instanceplugin\*PluginjenkinsPlugin, catalogPlugin