Best practices

Performance

Out of the box, SvelteKit does a lot of work to make your applications as performant as possible:

Code-splitting, so that only the code you need for the current page is loaded
Asset preloading, so that 'waterfalls' (of files requesting other files) are prevented
File hashing, so that your assets can be cached forever
Request coalescing, so that data fetched from separate server load functions is grouped into a single HTTP request
Parallel loading, so that separate universal load functions fetch data simultaneously
Data inlining, so that requests made with fetch during server rendering can be replayed in the browser without issuing a new request
Conservative invalidation, so that load functions are only re-run when necessary
Prerendering (configurable on a per-route basis, if necessary) so that pages without dynamic data can be served instantaneously
Link preloading, so that data and code requirements for a client-side navigation are eagerly anticipated

Nevertheless, we can't (yet) eliminate all sources of slowness. To eke out maximum performance, you should be mindful of the following tips.

Diagnosing issuespermalink

Google's PageSpeed Insights and (for more advanced analysis) WebPageTest are excellent ways to understand the performance characteristics of a site that is already deployed to the internet.

Your browser also includes useful developer tools for analysing your site, whether deployed or running locally:

Chrome - Lighthouse, Network, and Performance devtools
Edge - Lighthouse, Network, and Performance devtools
Firefox - Network and Performance devtools
Safari - enhancing the performance of your webpage

Note that your site running locally in dev mode will exhibit slightly different behaviour to your production app, which you can preview after building.

Instrumentingpermalink

If you see in the network tab of your browser that an API call is taking a long time and you'd like to understand why, you may consider instrumenting your backend with a tool like OpenTelemetry or Server-Timing headers.

Optimizing assetspermalink

Imagespermalink

Images are often one of the most impactful areas of opportunity for optimization. Svelte provides an offically-supported image optimization package. Please see the images page for more details. Additionally, Lighthouse is great for identifying which images on your site are the most problematic / in need of optimizations.

Videospermalink

Video files can be very large, so extra care should be taken to ensure that they're optimized:

Compress videos with tools such as Handbrake. Consider converting the videos to HTML5 video formats such as WebM or MP4.
Lazy load videos located below the fold.
Strip the audio track out of muted videos using a tool like FFmpeg.

Fontspermalink

When possible, preload fonts by calling resolve with the appropriate preload option in your handle hook, and ensure you've set the font-display option in your CSS. To reduce the size of font files, utilize subsetting.

Reducing code sizepermalink

Svelte versionpermalink

We recommend running the latest version of Svelte. Svelte 4 is smaller and faster than Svelte 3. (The Svelte 5 preview is much smaller and faster still, but we don't recommend that you upgrade to this version until it's production ready.)

Packagespermalink

rollup-plugin-visualizer can be helpful for identifying which packages are contributing the most to the size of your site. You may also find areas of opportunity manually inspecting the build output with build: { minify: false }, or via the network tab of your browser's devtools.

External scriptspermalink

Try to minimize the number of third-party scripts running in the browser. For example, instead of using JavaScript-based analytics you may wish to use server-side implementations. Many hosting providers with SvelteKit adapters offer such functionality such as Cloudflare, Netlify, and Vercel.

You also may consider running third-party scripts in a web worker with Partytown's SvelteKit integration.

Selective loadingpermalink

Code imported with static import declarations will be automatically bundled with the rest of your page. If there is a piece of code you need only when some condition is met, use the dynamic import(...) form instead.

You can fetch data and code before a page is actually loaded when the user hovers over (or begins to click) a link with the appropriate link options. This is configured by default on the <body> element when you create a new SvelteKit app.

For slow-loading data that isn't needed immediately, the object returned from your load function can contain a promise rather than the data itself. For server load functions, this will cause the data to stream in after the navigation (or initial page load).

One of the biggest performance killers is what is referred to as a waterfall, which is a series of requests that is made sequentially. This can happen on the server or in the browser.

Asset waterfalls can occur in the browser when your HTML requests JS which requests CSS which requests a background image and web font. SvelteKit will largely solve this class of problems for you by adding modulepreload tags or headers, but you should view the network tab in your devtools to check whether additional resources need to be preloaded. Pay special attention to this if you use web fonts since they need to be handled manually.
If a universal load function makes an API call to fetch the current user, then uses the details from that response to fetch a list of saved items, and then uses that response to fetch the details for each item, the browser will end up making multiple sequential requests. This is deadly for performance, especially for users that are physically located far from your backend. Avoid this issue by using server load functions where possible.
Server load functions are also not immune to waterfalls (though they are much less costly since they rarely involve roundtrips with high latency), for example if you query a database to get the current user, then use that data to make a second query for a list of saved items, then use that result to get the details of each item. It will typically be more performant to issue a single query with a database join.

Hostingpermalink

Your frontend should be located in the same data center as your backend to minimize latency. For sites with no central backend, many SvelteKit adapters support deploying to the edge, which means handling each user's requests from a nearby server. This can reduce load times significantly. Some adapters even support configuring deployment on a per-page basis. You should also consider serving images from a CDN (which are typically edge networks) — the hosts for many adapters offered by SvelteKit will do this automatically.

Ensure your host uses HTTP/2 or newer. Vite's code splitting creates numerous small files for improved cacheability, which results in excellent performance, but this does assume that your files can be loaded in parallel with HTTP/2.