Deco performance guide

Version: 1.0.0

Three rules for high performance

Understand the framework and system architecture

Using a tool efficiently implies understanding its components and capabilities. This allows for a healthy codebase from the start.

Example: deco offers components like Image that help accelerate the efficient delivery of images, but users can still use the img element if they prefer.

Learn to analyze performance

Learn to analyze the performance of your pages. Navigating a page should not be slow, not even for you. If it's slow for the developer, it will be slow for the user. Understand the available tools and how they work.

Monitoring performance is an ongoing task

Once the system stops performing, it is common to stop caring about performance until it becomes much more difficult to make corrections. Detecting performance issues is an ongoing task, but it should not be an end in itself.

Important: the most important thing (the golden rule) is the user experience. Sometimes the system may not deliver the best performance, as long as it is a conscious decision to deliver the best experience to the user.

Analyzing page performance

Testing performance involves understanding a series of tools and possible metrics that seek to understand what needs to be improved (and how). There are tools that help in this process. Understand how and when to use them.

Local testing

Example of using browser tools for debugging

Always test your page, section, loader, or action locally. It doesn't always need to be a structured test. If it's already slow for you, it will be slow for the user.

Run the deployed test versions. Then use the tools available in the browser itself to debug performance issues. With them, you can monitor the performance of the page in detail and simulate adverse conditions (such as a lower quality network or CPU). More information in the guide below:

Pagespeed test

Example of pagespeed test

Pagespeed is a testing tool developed by Google to evaluate the main performance aspects of a page. It explores metrics such as how long it takes for some content to be displayed, how long it takes for the system to become interactive, etc.

The main metrics are described below:

Metric Meaning example of good value
FCP First Contentful Paint up to 1.8s
LCP Largest Contentful Paint up to 2.5s
TBT Total Blocking Time 200ms
Speed Index Performance index of the page's most popular content up to 3.4
CLS Cumulative Layout Shift up to 0.1

source (adapted): https://web.dev/articles/fcp

Google aggregates these metrics into an index between 0 and 100, generating the pagespeed score.

Since it is a test executed in a production environment, it is subject to variations. However, a sudden drop in the pagespeed score implies looking at the page's performance as soon as possible.

Deco metrics test

Examples of deco metrics

By default, deco offers a set of metrics that are evaluated on a page. These metrics are specialized especially for systems like deco (SSR with hydration and with a lot of external data loading).

Metric Meaning
Config LCP Loading configuration of the largest content of the page
Page HTML Page size in bytes
Page Islands Number of islands on the page
Islands Props Size in bytes of the island's properties
Loaders latencies Response time of the page's loaders

Each metric is decomposed into its components, allowing for better identification of a possible source of performance loss. There is no ideal value for each metric, but it is important to understand that a large HTML impacts response time for the user and metrics like pagespeed.

A user who downloads at 100 kb/s will take 5s to download a 500 kb page. This is especially impactful for mobile users operating on low-bandwidth networks or situations.

Core Web Vitals test

Example of Core Web Vitals test

So far, we have only explored synthetic tests, that is, tests that are performed artificially and are only an approximation of the real world. These tests are relevant because a performance drop in a synthetic test usually reflects a performance drop in the real world.

Core Web Vitals represents a collection of metrics extracted from real user experience. It includes pagespeed metrics as well as other metrics specific to real user interaction.

Access through the Google CrUX website or directly from the CrUX App on your deco site.

Since it is a collected metric, it only has aggregated meaning. Google categorizes the values typically on a month-to-month basis, so it is more of a monitoring metric to diagnose any behavior issues that have gone unnoticed over time or that reflect a change in the page's audience.

Debugging performance issues

If, however, none of the tools help in the process of identifying a performance issue, perform some manual tests that can identify the cause of the problem:

  • If you don't know which change caused a performance loss:

    • Test older versions of the system to see which version impacted performance.
  • If it is not clear, within a version, what makes it slow, eliminate parts of the system until you identify the cause.

    • For a slow page, eliminate some sections until the page returns to good performance. The most recently deleted section (or its loaders or islands) may be the cause of the problem.
    • Eliminate added dependencies. Check if the system improves with this.
  • Update your dependencies.

    • Performance improvements are constantly added to the system and can fix issues that lead to slowness.

Performance improvements

After identifying a problem, whether through a negative experience or some tests indicating an inadequate metric value, action must be taken.

Each test above indicates, for each metric, possible culprits and where to look. If you have identified the culprit, follow one of the guides below to implement related improvements.

Note that the situations in which you should take action are just examples of some cases.

๐Ÿ–ผ๏ธ Images (jpg, png, gifs, ...)

When to act...

  • High LCP value (largest image takes a long time to download / view)
  • Pages "jumping" on the screen
  • Large size of downloaded images

Guide: Optimizing images

Tip: Use deco image components, such as <Image> and <Picture>, and configure them correctly, including width and height.

๐Ÿ“ˆ Images (SVG)

When to act...

  • Page size is large and the pages contain embedded and repeated SVGs
  • Issues with Speed Index

Guide: Optimizing SVGs

๐Ÿ–น Fonts

When to act...

  • Text appears to "change size" suddenly
  • Font file takes a long time to load
  • High FCP value (indicating the font as the problem)

Guide: Optimizing fonts

Tip: Use standard fonts offered by Google. If necessary, use smaller fonts (preferably woff/woff2).

๐Ÿ“œ Third-party scripts

When to act...

  • The system takes a long time to load (high TBT)
  • The screen "jumps" due to a component inserted by a third-party script (high CLS)
  • A third-party script has a large size

Guide: Optimizing scripts

Tip: Try to only use what is strictly necessary or replace third-party scripts with lighter versions. If not possible, postpone the execution of the script until after the page has already loaded and the user is interacting with it.

๐Ÿ”„ Data loading efficiency (loaders)

When to act...

  • The system indicates high latency for a loader
  • The page takes a long time to load initially
  • The size of props in islands is large

Guide: Optimizing loaders

Tips:

  • Use inline loaders to transform data to be sent to a section and/or island
  • Consider deferring the display (Deferred) of sections with expensive loaders
  • Change the loader's props to reduce the amount of data loaded in the loaders
  • Save loaders that are reused on different pages/sections

๐Ÿ๏ธ Islands

When to act...

  • The system indicates a large number of islands
  • The page takes a long time to load initially
  • The size of props in islands is large

Guide: Optimizing islands

Tips: Prefer pure CSS to avoid islands. Use children to pass JSX into an island. Reduce the scope of the island as much as possible (e.g., prefer a button as an island rather than an entire form).

Relevant sources