One of the core features of LogRocket is the replay of console and Redux logs in production web apps. To do this, you add the LogRocket SDK to your app which sends logs to LogRocket. Then, when triaging a bug or user issue, you can replay the logs in LogRocket to see what went wrong.
When we first designed the log viewer, we went through a number of design iterations, but eventually settled on replicating the look and feel of the Chrome console. After all, developers are already used to working with this interface, so why reinvent the wheel?
It turns out that re-building the Chrome console for LogRocket was no simple task, since our use case involved a number of complexities not present in Chrome’s implementation. In this piece, I’ll discuss how we used
react-measure to build a performant and maintainable log viewer.
Initial Design Concerns
The high-level design concerns for our log viewer were as follows:
Smooth scroll performance
Building a long list that scrolls smoothly isn’t trivial, but it’s crucial for a pleasant user experience. Since a session could potentially have thousands of logs, we knew that we’d need to build a virtual list where DOM nodes are unmounted when they leave the viewport.
User-interactive JSON tree
Lazy-loaded object expansion
This is a departure from how the Chrome console works (where all data is already in memory), which meant that we would need a loading state and error handling for failed queries.
The state of the log viewer should be persisted when the component unmounts and re-mounts. Basically, the state can’t be kept at the component level and should be stored in Redux.
Building a virtual list with React-Virtualized
When rendering a very long list, it is often prohibitively expensive to keep all items in the DOM, since each node requires a fixed amount of memory. To solve this problem, you can build a virtual list, where each item is only rendered when it is actually visible.
There are a number of React libraries that facilitate this, but the most feature-rich and robust is
react-virtualized. It provides a host of utility components for building virtual lists, grids, and tables. It has an active community of contributors and a slack group that is very helpful for discussing issues.
How it works
react-virtualized bypasses the browser’s layout engine to determine where to arrange items. As you scroll through a list, it looks at the current scroll position, and determines which items are in the viewport. It then renders those items, and uses absolute positioning to place each row in the correct place. As such, it can freely mount and unmount rows without affecting the positioning of subsequent rows.
One caveat of this approach is that rows with dynamic height are a bit tricky to implement. In a standard flow-based layout, or with Flexbox, if an item in a list grows in height, the browser will push down the subsequent items to make room.
react-virtualized, however, needs to be notified whenever an element’s height changes so it can adjust the
absolute positioning of subsequent items.
Lets take a look at the basic props of the
<List /> component:
react-virtualized needs to know the width and height of the list viewport in order to do its calculations. If your list isn’t fixed in width or height, there are helper components for hydrating these values.
The number of rows in the list.
If all rows in the list have a fixed height, this value can be a number. If the height of each row is different, rowHeight can be a function which returns the height of a given row by index. More on this later.
This is a function that takes in
index (and some other non-essential props) and returns the row to render. A simple implementation might look like this:
react-virtualized doesn’t actually take in the list itself as a prop. Simply knowing the length of the list and having a
rowRenderer function that can render a given row is all it needs!
I’m not going to describe every detail of our console implementation since much of it is a standard application of
react-virtualized, but there are a few bits where we diverged that are interesting.
As I described earlier,
react-virtualized takes a prop
rowHeight which is a function that returns the height of a row at a given index.
In this screenshot of the LogRocket log viewer, notice that there are 2 states for each row: default, and expanded. When a row is in the default pre-expanded state, it’s height is fixed at
22px tall. However, when a row is expanded, its height varies as the user expands different subtrees of the object.
We needed a way to write a
rowHeight function that handles dynamic height rows- something like this:
In order to implement
getExpandedRowHeight in the above psuedo-code, there were two potential options.
Guarantee deterministic height of an expanded object
To achieve this, we would have needed to design the object tree view component from the ground up to make its height a pure function of the subtrees that are expanded. Put another way, we could write a function that takes in a list of the expanded subtrees in an object, and have it return the height.
In theory this is doable, but there are number of complications. It is difficult to account for text that overflows to the next line, as this increases the height of the object. Also, making this guarantee would make it difficult to iterate on the look and feel of the log viewer since changes to things like margins and padding would need to be adjusted for.
Instead, we opted to use a library called
react-measure which provides a helpful abstraction for writing components that are aware of their own height.
react-measure wraps a given component and takes a prop,
onResize which is a function that is called whenever the component’s size changes.
In our case, whenever the size of a given row changes, we dispatch a Redux action which stores the height of that row in Redux. Then in our
rowHeight function, we simply get the height of the row from Redux, and
react-virtualized can render it properly.
There is a small performance penalty to this approach, since
react-measure uses the DOM resize-observer API which isn’t implemented natively in all browsers, but in practice this is fairly minimal.
To handle data fetching, we use
apollo-client which is a GraphQL client that works nicely in React apps. When a user clicks on a log entry to see the full object, the log entry goes into a loading state which has a fixed height.
Apollo makes a request to the backend and then populates the data in the Redux store. This then triggers
react-virtualized to update and the row height changes when the data is filled in.
To let users explore logged objects, we looked at a few different off-the-shelf components, but eventually chose
react-inspector. This library includes a host of components for logging objects and DOM nodes. We did, however, end up forking the library in order to build a controlled version (so we could keep state in Redux).
LogRocket is like a DVR for web apps, recording literally everything that happens on your React app. Instead of guessing why problems happen, you can aggregate and report on what state your application was in when an issue occurred. LogRocket also monitors your app's performance, reporting with metrics like client CPU load, client memory usage, and more.
The LogRocket Redux middleware package adds an extra layer of visibility into your user sessions. LogRocket logs all actions and state from your Redux stores.
Modernize how you debug your React apps — start monitoring for free.