picostitch
crafting (and) JavaScript

Resource Timing (part 2) - Loading Dependencies

This is post #2 about Resource Timing, with a focus on understanding loading dependencies. In part 1 "Resource Timing - The API" I covered the attributes responseEnd and startTime. Now I will try to show how the API can be used to see which resources block each other, which is the step to understand before optimizing site speed, a very essential step to know what is the right thing to optimize.

The Resource Timing API provides information about resources loaded on a site, it is one of the Performance APIs, accessible via JavaScript using window.performance. Here I will explore how much details can be gathered about each resource that is loaded (JS files, CSS files, fetches, etc.). Especially about their timing and dependencies.

The Context

Let's shortly sum up how we gather those data. In all modern browsers, there is a built-in functionality that you see in action, right here:

window.performance.getEntriesByType('resource')

This returns all performance resource timing entries for all resources. As an array of instances where each item implements the PerformanceResourceTiming interface, a set of defined properties.

[{
    // ... shortened
    name: "https://picostitch.com/_global.css",
    initiatorType: "link",
    startTime: 18.41499999864027,
    // ... shortened
}, {...}]

The startTime Attribute

The spec says that the attribute startTime is "the time immediately before the [browser] starts to queue the resource for fetching". This means, for example if the browser is about to load the HTML page the startTime is recorded. This time is relative from "the time when the browsing context is first created". In other words when the browser starts loading a page. So this is a very browser-internal thing, that we would not be able to get access to without extending the browser itself. That proves that this API can give us insights which we could not be able to obtain otherwise, especially not on any website itself. These numbers are also way more explicit in regards to what the user expriences.

A startTime of 0 is the moment when the browsing context was created, when the browser started loading this page. This is for example a website reload, or navigating to a new site, opening a tab with a certain URL and alike. That means a startTime=20 means 20 milliseconds after the browser context was created. Having a timestamp relative to the browser context creation allows to get insights on the resource timings without any timestamp offset calculations, as needed when using Date.now() (more in the previous post).

Let's look at some real life numbers:

The "startTime" for all resources requested by this page

Hint: The chart above shows the real numbers of the page you are looking at, if you reload, the numbers above may change (changes can be seen best on the x-axis scale).

In the chart one should see that some requests start at the same time. These requests have most probably been triggered by the same resource. For example the "blog.css", the logo and the "plausible.js" files start to get loaded around the same time, as the first resources. This is because the HTML file that directly references them, triggers the loading of these files at the same time. The browser reads the HTML file, interprets it and finds <link> and <script> tags that reference those resources, so they start to get loaded, hence they have the lowest startTime.

In the chart one can also see that the "_global.css" start loading way later. This is because it is referenced inside the "blog.css" file, which needs to be downloaded and read by the browser first to know that the "_global.css" needs to get loaded. This way one can understand better how the browser handles the resources, in which order and also using which constraints to load them.

The initiatorType Attribute

This attribute is the type that triggered this resource to be loaded. In the chart above one can see strings like initiatorType: "navigation", initiatorType: "link", initiatorType: "script" and others. There are basically two types of attributes, the initiatorType is either:

  1. The name of the HTML tag (like <script> or <link>), the DOM node (or as the spec text says the "value as the localName of that element [DOM], if the request is a result of processing the element"),
  • this refers to attributes like src in a <script src="resource.js"> tag,
  • the href attribute in a <link href="resource.css"> tag or
  • the src atttribute of an <img> tag among others.
  1. Or it is one of "css", "navigation", "xmlhttprequest", "fetch", "beacon" or "other" when the loading was not directly triggered by a DOM node,
  • these are things like loading a new page ("navigation"),
  • loading a resource from within a CSS file ("css") or
  • the request done via the native JavaScript window.fetch() method ("fetch").

This attribute gives a bit more context to the loading times and makes the loading dependencies talked about in the paragraph before easier to understand. A CSS file needs to be loaded by the browser, before it can know what other files are referenced inside of it and need to be loaded. This might also surface the question how much should be inlined, or if preload should be used to speed up website loading.

Focus on the Dependencies

I hope this post allows to gather better insights and hopefully also decouple the dependencies among resources where accurate and therefore speed up the website. I tried to focus on just looking at the two attributes startTime and initiatorType so the loading dependencies become very obvious and in case it is what one wants to focus on, they might be a good start.

This site is quite simple and therefore makes the dependencies easy to understand. If you try this out on a more complex page loading dependencies will also be a bit harder to identify. Have fun untangling your loading dependencies. Feel free to share findings and insights with me.

Soon: You want to know more? Read [part 3 of this series about "Waterfall Chart"].

Main photo by Ben Harritt on Unsplash
The first version of this post I wrote and published for the HolidayCheck TechBlog, June 2019.