Throttle back on your scroll events and watch out for the IntersectionObserver!
On the web, we’re used to change. We embrace change. It’s part of what makes the web unique: an ever-evolving, ever-moving ecosystem of growing users, each with their own ideas, challenges, and goals to bring to the web.
Part of embracing that change means, as builders for the web, we have to be light on our toes and ready to adapt to factors within and without our control. It’s how you prepare and react to the changes that really matters.
Life moves pretty fast. If you don't stop and look around once in a while, you could miss it.Ferris Bueller
If you’ve ever built (or used) a lazy-loader or implemented infinite-scrolling on a website, you might be familiar with the history of these techniques. Extremely popular within native phone apps for their benefit to loading times and lower bandwidth overhead, lazy-loading and infinite-scrolling are part of a methodology for building leaner apps and websites, specifically to do with speed and performance of loading times.
only when an image is within the viewport should it begin loading.
Similarly, an infinite-scrolling technique typically involves a long list of many items—too many to display on the page at once (to keep the page weight low). When the user reaches the bottom of the list—that is to say that the bottom of the list is visible within the viewport—the next set of
n items are loaded and placed in the DOM at the bottom of the list. This gives the user the impression that the list is never-ending, as it is technically as limitless as the dataset being represented.
But be aware that these techniques come with their own set of UX and/or performance considerations.
Over the many years of employing lazy-loading, infinite-scrolling, and similar techniques on the web, developers have come up with some clever solutions.
At it’s simplest, we hook into the window’s
But we run a performance risk with this implementation. And in order to understand the risk, it’s important to also understand a bit about how your browser displays a webpage, and what your browser has to do when you scroll or resize a webpage.
Given that the refresh rate of most devices today is 60Hz—the display refreshes its screen 60 times per second—we can determine that we have a budget of approximately 16 milliseconds to do what we need to do for each frame/display refresh.
In reality, however, the browser has housekeeping work to do, so all of your work needs to be completed inside 10ms.Google Web Fundamentals
This might seem like an impossible amount of time to do anything with, but for a computer, this isn’t so much of a stretch. That’s not to say that it’s easy for the browser, nor that the browser is even doing it in the most efficient manner, in the most efficient circumstances, so we should do what we can to offer the browser any help!
This short, 16ms portion of our performance budget is a critical one. It can mean the difference between a smooth website, promoting a positive user experience, and a janky website, a wholly negative user experience.
action we need to perform involves checking the visibility of an element as we move around the page, we need to make sure that any
checks we’re making happen in under 16ms, or we’ll fail to build each
frame of the 60 FPS animation in time. Failing to do so results in frames being painted late or not being painted at all, and the delaying or removal of any frames will naturally result in a lower FPS. Too low and your users will start to notice.
As the user scrolls quickly around the website, their browser is silently firing thousands of scroll events, 100% of which we’re trying to gate-check and before their consequences are delivered to the user.
Like a gatling gun, these events will fire at a rate nearly, if not completely, imperceptible to the human eye. In order to get an idea of the volume of these events, browse to your favourite news site and scroll from the top of the page to the bottom. Try to count the height of the page in pixels. Bonus points (and bonus events) if you do this on a small screen!
Alternatively, check out how much you can resize a page on a pointer-based device. I’ve tried this on a desktop monitor, and was able to get my browser window as large as
1665px × 902px and as small as
400px × 198px. That gives me
1265 × 704 = 890,560 possible ways to resize my browser. While I doubt any of your users navigate your site by slinky-ing their browser around the screen, we must be aware of such a situation and do the
web development dance of anticipating and preparing for outlier circumstances.
There are a handful of approaches to prevent our code from thrashing our users’ browsers. I won’t highlight all of them here, but two of the most common techniques involve limiting how often we perform our checks.
In this example, we set up a lock to prevent the check from firing if we’re already scrolling or resizing. Each time an event fires, we set our
stuff we want, the lock will be up, and our function will not run completely again until the first pass is complete, upon which time the lock will be lifted and we’ll be free to do more
While we’ve stopped our code from executing every scroll or resize event, we’ve also limited the opportunities in time for our actions to be carried out. Instead of our actions firing at the exact moment they should, we may defer the action to the next frame if too many events occur within our 16ms budget. The complexity of the code has also increased, and tracking our lock feels a little wonky.
In this example, we set up a timeout so that the check will only run every 250ms at the most. Each event triggers a timeout which calls the action function after 250ms. If another event occurs before this 250ms timeout, the existing timeout will be destroyed and a new 250ms timeout will be created.
This also stops our code from executing every scroll or resize event, we’re also limiting the opportunities for our actions to be carried out. In this case, we could tinker with the
250 value until we find a millisecond value that has a less discernable delay, but the reality is that the ideal delay sits right at our performance budget:
And even then, we’re making a pretty broad assumption about the refresh rates of our users! We can be pretty sure, but not confident that our users’ device screens operate at 60Hz.
It seems the ideal solution would remove the guesswork in finding the best
middle-ground delay and the assumptions we have to make about refresh rates.
All of the techniques we’ve covered so far take an approach that, instead of reacting to each individual change as they happen, make proactive, repeated checks and set up reminders (in the form of
setTimeout) for the browser to do something with any changes that are found.
Imagine if answering phone calls involved picking up and putting down your phone, constantly, over-and-over, all day, in the hope that eventually someone will be on the other end to talk to you.
Might as well leave your oven on all day and all night, all year, because eventually you’ll bake bread, and you might even bake bread more than once that year.
What a waste.
This is like running back-and-forth across a crosswalk, with a stop sign in hand, hoping that eventually, one of the times that you cross, you’ll find some use by helping someone across road.
When we use these
setTimeout techniques, this is the kind of functionality we’re creating. We’re not coupling our observation and action steps in a simple way, and we’re creating arbitrary benchmarks based on our best guesses as developers. What we fail to account for is the unbelievably broad spectrum of devices we might be serving. And it is rightfully so that we should fail to account for this broad spectrum, as any guesses and benchmarks we make sit within a context and hold a bias that the browser/computer would not.
How often should the Crossing Guard cross the road?
Is it better to go back-and-forth often, taking only one person per crossing? Or is it better to delay before each back-and-forth in order to potentially take more than one person per crossing?
We have no way of guaranteeing an accurate
fits all guess.
Instead of doing so ourselves, let’s instead delegate these
guesses decisions to each individual user’s browser and device. Instead of the Crossing Guard running across the road constantly, they should wait until someone needs to cross, and then do so.
When we use this IntersectionObserver technique, we cease relying on arbitrary benchmarks, and we let the browser do the lifting that we had to burden ourselves with before. Rather than the Crossing Guard repeatedly crossing and hoping to be useful eventually, the Crossing Guard instead waits for any pedestrians before crossing with them.
On top of the performance benefits and relief of mental overhead you’ll gain from this technique, there are also a number of useful properties made available to you, the most useful of which I find to be
entry.target. These properties could very likely simplify your existing code, and could even offset the added verbosity you might incur from using IntersectionObserver over a previous technique.
Other benefits you’ll reap from IntersectionObserver include:
- clean, easily-repeated syntax to create new instances of an IntersectionObserver
- simple to begin and stop observing an Element at any point
- ability to observe multiple Elements in different, siloed ways
- simple to define what
observedmeans: visible at all, just visible, 50% visible, 100% visible, etc.
Further, take also careful note of IntersectionObserver’s browser support to see if you need the Polyfill or not.
An important caveat to note before diving too deep, at least at the time of writing, is that IntersectionObserver does not support observing pseudo-elements; rather, IntersectionObserver’s
observe() method expects a single, Element-type parameter. While this will make some implementations of IntersectionObserver more verbose than their equivalent older techniques, the mental overhead and performance tradeoffs that can be made are unquestionably beneficial.
IntersectionObserver could not be better suited as a technique for lazy-loading and infinite-scrolling. Because we aren’t defining any benchmarks by which the browser should operate, the limit to the performance with IntersectionObserver is defined by the user’s browser, not the developer building the website. This is the keystone of IntersectionObserver and other emerging techniques, such as ResizeObserver and Mutation Observer:
Let the developer figure out what to do. Let the browser figure out how to do it.