last updated: November 08, 2022

7 minute read

Server Side Rendering vs React Server Components

With the release of Next.js version 13, React server components are finally an official, stable feature of the React ecosystem. But for those not as familiar with the ins and outs of server rendering and hydration, the impact of this release may not be so clear.

Part of the reason for this confusion is that much of the material written/discussed on the subject tries to avoid directly comparing server components with server side rendering, claming instead that the two compliment one another. While this is technically true, comparisons are often much easier to digest, and it's not too hard to divise a situation where a comparison is fairly reasonable.

Say we have an app made of the following two components:

export default function Parent() {
return (
<div>
<p>this is some static markup</p>
<Child />
</div>
);
}
function Child() {
const [counter, setCounter] = useState(0);
return (
<div>
<p>{counter}</p>
<button onClick={() => setCounter((prevCounter) => prevCounter + 1)}>
Increase
</button>
</div>
);
}

The key thing to notice is that Parent only contains static markup, while Child contains state and event handlers - the building blocks of interactivity.

Let's consider what the request lifecycle would look like if you went to the page corresponding to Parent.

  1. At build time (or request time, if you use getServerSideProps), Next.js runs the default export for the page, i.e. Parent, to generate its markup. This includes running Child to determine its own markup for the first render. In our case, this would generate something that roughly looks like the following:
<div>
<p>this is some static markup</p>
<div>
<p>0</p>
<button>Increase</button>
</div>
</div>
  1. At request time, Next.js sends the html generated in step 1 to the browser, along with a script tag that references a javascript bundle which lives on the server.

This javascript bundle contains the React code for our two components, Parent and Child

  1. The browser paints the html for the user to see, and fetches the javascript bundle from the server
  2. The javascript bundle is fetched and executed, booting up the Next.js runtime on the browser
  3. Next.js (on the browser) reruns the code of the two components, but instead of only generating the html for the first render like it did in step 1, it also creates state (with useState) and event listeners (with onClick). The state and event listeners are attached to the DOM, and your app is now interactive. This process of rerunning components on the browser to make server-generated html interactive is known as hydration.

Let's consider server components.

In a nutshell, server components are React components that render to html on the server - client components (i.e. all current React components) render to html on the client. This'll become more clear when we go through the request lifecycle.

A few interesting tradeoffs to consider:

  1. Server components can import client components, but client components cannot import server components
  • However, a client component can render a server component as a child if it's passed down as a prop!
  1. Server components can't have any kind of interactivity or state
  2. Any code or libraries imported into a server component isn't shipped to the client - only the html that code generates

if you're already familiar with server components, you may know they're not actually rendered to html, but rather a new JSON-like format. This is important for maintaining client-side state when a client-component's server-component parent is refetched - but that's not important for our purposes.

With that in mind, let's take a look at a version of Parent and Child which uses server component architecture (SCA).

// parent.server.js
export default function Parent() {
return (
<div>
<p>this is some static markup</p>
<Child />
</div>
);
}
// child.client.js
export default function Child() {
const [counter, setCounter] = useState(0);
return (
<div>
<p>{counter}</p>
<button onClick={() => setCounter((prevCounter) => prevCounter + 1)}>
Increase
</button>
</div>
);
}

This should look similar to our SSR code, but now our Parent and Child are in two different files. Notice that Child has a .client.js file extension, while Parent has a .server.js extension - this is to indicate which is a server component, and which is a client component. Since Parent only renders static markup, it's a great candidate for a server component, while Child needs to be a client component since it uses state.

Let's go through the request lifecycle for our new code and compare it to the Next.js lifecycle.

  1. At request time, React runs the server components relevant to the request (in our case Page) to generate its html. However, unlike in SSR, this does not include running Child to determine its markup for the first render. Instead, Child is left as an empty div to be rendered on the client. This means we generate markup that looks something like the following:
<div>
<p>this is some static markup</p>
<div />
</div>
  1. React sends the html generated in step 1 to the browser, along with a script tag that references a javascript bundle which lives on the server.

Unlike SSR, whose bundle includes the React code for Parent and Child, the bundle when using SCA only includes the code for Child!

  1. The browser paints the html for the user to see, and fetches the javascript bundle from the server
  2. The javascript bundle is fetched and executed, booting up the React runtime on the browser
  3. React (on the browser) runs the code necessary to render Child. Unlike SSR, there's no hydration - the only components which would need to be hydrated (i.e. stateful client components) were never rendered to dehydrated html in the first place!

Let's really hammer in the differences between these two lifecycles.

  1. With SSR, the React code for Parent is sent to the browser in the javascript bundle, while with SCA, Parent is not. In other words, server components do not increase the bundle size! Imagine if Parent imported a heavy date formatting library - with SSR, the whole library would be sent to the browser, while with server components, it's not.
  2. The reason the React code for Parent isn't sent in the javascript bundle with SCA is that server components don't need to hydrate! Remember, hydration is the process of adding state and event listeners to dehydrated html - but server components, by definition, don't have any interactivity. Meanwhile, client components are only rendered on the client - they're never rendered to dehydrated html on the server.

Let's switch gears and talk about some of the side-effects of only rendering client components on the browser: with our example, the user is left with a blank <div /> until React renders the component in step 5. Not the best user experience. In fact, doesn't Next.js use SSR to server-render components that would otherwise be rendered on the client precisely to avoid this situation - only today, the entirety of every React app is composed of client components.

What if - and this is where things get interesting - we could server-render our client components, hydrate them on the client, but leave our server components alone? By using server-rendered client components, we would only need to ship the code for Child to the browser, while leaving Parent out of the bundle altogether. In other words, server-rendered child components take the reduced bundle of SCA with the great ux of having initial markup generated from SSR.

In essence, this is Next.js version 13 - SCA with server-rendered client components. It took a little while to get here, but I hope I could give you a better understanding of this new release - it's really cool stuff, and I'm excited to see how it plays out in practice.

you might also like:

How I Built My Blog

May 05, 2023

The stack I chose, or how I learned to stop worrying and love Next.js, mdx, prisma, and tailwind

software engdevopsreactnextjs

2