SlideShare a Scribd company logo
MEASURING THE
PERFORMANCE
OF SINGLE PAGE
APPLICATIONS
| | |nic jansma SOASTA nicj.net @nicj
| | |philip tellis SOASTA bluesmoon.info @bluesmoon
SLIDES
slideshare.net/nicjansma/
WHO ARE WE?
Nic Jansma
SOASTA
Philip Tellis
SOASTA
DEFINITIONS
RUM
Real User Monitoring
Gathering performance metrics from real user
experiences
Versus Synthetic Monitoring, with emulated users in a
controlled environment
RUM: HOW IT'S DONE
JavaScript measures the browser's events and
performance interfaces
Listen for readyStatechanges and the onloadevent
Measure DNS, TCP, SSL, Request and Response times
from NavigationTiming and user measurements from
UserTiming (if available)
Gather User Agent characteristics (Version, Screen Size,
etc)
Beacon this data back to the cloud for analytics
NAVIGATIONTIMING
NAVIGATIONTIMING
RESOURCETIMING
RESOURCETIMING
RESOURCETIMING
BOOMERANG
Created by Philip Tellis @ Yahoo
Gathers performance metrics and characteristics of page
load and beacons data to your server (aka RUM)
Open-source project (with contributions from SOASTA)
https://github.com/lognormal/boomerang/
SPAS
SINGLE PAGE APPS
Run on a single page, dynamically bringing in content as
necessary
Built with frameworks like AngularJS, Ember.js,
Backbone.js, React, etc.
SPAS
HARD VS. SOFT NAVIGATIONS
Hard Navigation: The first page load, which will include
all static HTML, JavaScript, CSS, the SPA framework itself
(e.g. angular.js), plus showing the initial route
Soft Navigation: Any subsequent route (address bar)
change
Any URL might be loaded via either hard or soft navigation
3 CHALLENGES
OF MEASURING THE
PERFORMANCE OF SPAS
CHALLENGE #1
THE ONLOAD EVENT NO LONGER
MATTERS
Traditional websites:
On navigation, the browser begins downloading all of the
JavaScript, CSS, images and other static resources
Once all static resources are fetched, the body's onload
event will fire
This is traditionally what websites consider as page load
complete
This is traditionally what RUM measures
TRADITIONAL WEBSITE
WATERFALL
CHALLENGE #1
THE ONLOAD EVENT NO LONGER
MATTERS
Single Page Apps:
Load all static content like a traditional website
The frameworks' code will also be fetched (e.g.
angular.js)
(the onload event fires here)
Once the SPA framework is loaded, it starts looking at
routes, fetching views and data
All of this content is fetched after the onloadevent
SPA WATERFALL
SPA WATERFALL
Browser fires onloadat 1.225
seconds
All visual resources (.jpgs) aren't
complete until after 1.7 seconds
Filmstrip confirms nothing is
shown until around 1.7 seconds
onloadfired 0.5 seconds too
early!
CHALLENGE #1
THE ONLOAD EVENT NO LONGER
MATTERS
Single Page Apps:
Core problem is that most of the interesting stuff (e.g.
fetching images, JavaScript, CSS and XHRs for the route)
happens after the onload
The browser doesn't fire any "fully loaded"-style events
after onload
CHALLENGE #2
SOFT NAVIGATIONS ARE NOT REAL
NAVIGATIONS
Each route change, user interaction, or visual update is
dynamically fetched from the server
There are APIs to change the URL (and detect changes) in
the address bar without actually navigating
New content is dynamically swapped in over the old
content
The browser is no longer doing a traditional navigation,
where it's tearing down the old page
CHALLENGE #2
SOFT NAVIGATIONS ARE NOT REAL
NAVIGATIONS
This is great for performance
The browser is no longer re-rendering the same header,
footer or common components
The browser is no longer re-parsing the same HTML,
JavaScript and CSS
CHALLENGE #2
SOFT NAVIGATIONS ARE NOT REAL
NAVIGATIONS
Bad for traditional RUM tools:
Stop caring after the measuring the "one" navigation
Won't run again until the next time it loads on a full
navigation
Browser events (readyState, onload) and metrics
(NavigationTiming) are all geared toward a single load
event
CHALLENGE #3
THE BROWSER WON’T TELL YOU
WHEN ALL RESOURCES HAVE BEEN
DOWNLOADED
The browser fires onloadonly once
The onloadevent helps us know when all static content
was fetched
In a soft navigation scenario, the browser does not fire the
onloadevent again, so we don't know when its content
was fetched
CHALLENGE #3
THE BROWSER WON’T TELL YOU
WHEN ALL RESOURCES HAVE BEEN
DOWNLOADED
SPA soft navigations may fetch:
Templates
Images
CSS
JavaScript
XHRs
Videos
CHALLENGE #3
THE BROWSER WON’T TELL YOU
WHEN ALL RESOURCES HAVE BEEN
DOWNLOADED
SPA frameworks often fire events around navigations.
AngularJS events:
$routeChangeStart: When a new route is being
navigated to
$viewContentLoaded: Emitted every time the ngView
content is reloaded
But neither of these events have any knowledge of the work
they trigger, fetching new IMGs, CSS, JavaScript, etc!
ANGULAR TIMELINE
ANGULARJS EVENT WATERFALL
HOW CAN WE MEASURE
SPA NAVIGATIONS?
We need to figure out at what point the navigation started
(the start event), through when we consider the navigation
complete (the end event).
THE START EVENT
For hard navigations:
The start event is when the browser starts the process of
loading the next page
This is the same time as with traditional web app
navigations
We can use NavigationTiming's navigationStartif
available, to know when the browser navigation began
If NavigationTiming isn't available, and the user is
navigating between pages on the same site, you can use
cookies to measure when the navigation began (see
Boomerang for an implementation)
THE START EVENT
Challenge #2: Soft navigations are not real navigations
We need to figure out when the user's view is going to
significantly change
The browser history is changing
SPA framework routing events can give us an indicator
that the view will be changing
Other important events that might indicate a view change
are a user click, or an XHR that triggers DOM changes
THE START EVENT:
HISTORY STATE
The window.historyobject can tell us when the URL is
changing:
When pushStateor replaceStateare being called, the
app is possibly updating its view
When the user hits Back or Forward, the
window.popstateevent is fired, and the app will
possibly update the view
(future events will give us more info)
THE START EVENT:
ROUTING
SPA frameworks fire routing events when the view is
changing:
AngularJS: $rootScope.$on("$routeChangeStart")
Ember.js: beforeModelor willTransition
Backbone.js: router.on("route")
THE START EVENT: CLICKS
When the user has clicks something, they might be doing
simple interactions (e.g. a drop-down menu)
Or, they might be triggering a UI update
(future events will give us more info)
THE START EVENT: XHRS
An XMLHttpRequest(network activity) might indicate
that the page's view is being updated
Or, it could be a periodic poller (e.g. a scoreboard update)
Or, it could be in reaction to a user interaction (e.g.
autocomplete)
(future events will give us more info)
THE START EVENT
To determine if a user click or XHR is really triggering a
navigation, we can listen to what happens next
If there was a lot of subsequent network activity, we can
keep on listening for more events
If history (address bar) changed, we can consider the
event the start of a navigation
If the DOM was updated significantly, we can consider the
event the start of a navigation
If nothing else happened, it was probably just an
insignificant interaction
SPA NAVIGATIONS
THE END EVENT
When do we consider the SPA navigation complete?
There are many definitions of complete:
When all networking activity has completed
When the UI is visually complete (above-the-fold)
When the user can interact with the page
THE END EVENT
Traditional RUM measures up to the onloadevent:
This is when all resources have been fetched
The page isn't fully loaded until at least then
The UI might have been above-the-fold visually complete
already
It's traditionally when the user can fully interact with the
page
SINGLE POINTS OF
FAILURE (SPOFS)
Which resources could affect visual completion of the page?
External JavaScript files
External CSS files
Media (images, video)
THE END EVENT
For hard navigations, the onloadevent no longer matters
(Challenge #1)
The onloadevent only measures up to when all static
resources were fetched
The SPA framework will be dynamically loading its UI only
after the static JavaScript has been loaded
We want to mark the end of the hard navigation only after
all of the resources were fetched and the UI is complete
THE END EVENT
For soft navigations, the browser won’t tell you when all
resources have been downloaded (Challenge #3)
The onloadonly fires once on a page
APIs like ResourceTiming can give you details about
network resources after they've been fetched
But to know when to stop, we need to know if there are
any outstanding resources
So let's monitor all network activity!
THE END EVENT
Let's make our own SPA onloadevent:
Similar to the body onloadevent, let's wait for all network
activity to complete
This means we will have to intercept both implicit
resource fetches (e.g. from new DOM elements) as well as
programmatic (e.g. XHR) resource fetches
MONITORING XHRS
XMLHttpRequests play an important role in SPA
frameworks
XHRs are used to fetch HTML, templates, JSON, XML, data
and other assets
We should monitor to see if any XHRs are occuring
The XMLHttpRequestobject can be proxied
Intercept the .open()and .send()methods to know
when an XHR is starting
MONITORING XHRS
Simplified code ahead!
Full code at
github.com/lognormal/boomerang/blob/master/plugins/auto_xhr
MONITORING XHRS
varorig_XHR=window.XMLHttpRequest;
window.XMLHttpRequest=function(){
varreq=neworig_XHR();
orig_open=req.open;
orig_send=req.send;
req.open=function(method,url,async){
//saveURLdetails,listenforstatechanges
req.addEventListener("load",function(){...});
req.addEventListener("timeout",function(){...});
req.addEventListener("error",function(){...});
req.addEventListener("abort",function(){...});
orig_open.apply(req,arguments);
};
req.send=function(){
//savestarttime
orig_send.apply(req,arguments);
}
}
MONITORING XHRS
By proxying the XHR code, you can:
Know which URLs are being fetched
Know when a XHR has started
Know when a XHR has completed, timed out, error or
aborted
Measure XHR states even on browsers that don't support
ResourceTiming
Most importantly, know if there are any outstanding
XHRs
MONITORING XHRS
Downsides:
Need additional code to support XDomainRequest
Timing not as accurate when browser is busy (rendering,
etc) as callbacks will be delayed
You can fix-up timing via ResourceTiming (if available)
OTHER RESOURCES
XHR is the main way to fetch resources via JavaScript
What about Images, JavaScript, CSS and other HTML
elements that trigger resource fetches?
We can't proxy the Imageobject as that only works if you
create a new Image()in JavaScript
If only we could listen for DOM changes...
MUTATION OBSERVER
:
http://developer.mozilla.org/en-
US/docs/Web/API/MutationObserver
MutationObserver provides developers a way to react to
changes in a DOM
Usage:
observe()for specific events
Get a callback when mutations for those events occur
MUTATIONOBSERVER
Simplified code ahead!
Full code at
github.com/lognormal/boomerang/blob/master/plugins/auto_xhr
varobserver=newMutationObserver(observeCallback);
observer.observe(document,{
childList:true,
attributes:true,
subtree:true,
attributeFilter:["src","href"]
});
functionobserveCallback(mutations){
varinteresting=false;
if(mutations&&mutations.length){
mutations.forEach(function(mutation){
if(mutation.type==="attributes"){
interesting|=isInteresting(mutation.target);
}elseif(mutation.type==="childList"){
for(vari=0;i<mutation.addedNodes.length;i++){
interesting|=isInteresting(mutation.addedNodes[i]);
}
}
});
}
if(!interesting){
//completetheeventafterNmillisecondsifnothingelsehappens
}
});
MUTATIONOBSERVER
Simplified workflow:
Start listening when an XHR, click, route change or other interesting navigation-like
event starts
Use MutationObserverto listen for DOM mutations
Attach loadand errorevent handlers and set timeouts on any IMG, SCRIPT, LINK
or FRAME
If an interesting element starts fetching keep the navigation "open" until it completes
After the last element's resource has been fetched, wait a few milliseconds to see if it
kicked off anything else
If not, the navigation completed when the last element's resource was fetched
MUTATIONOBSERVER
What's interesting to observe?
Internal and cached resources may not fetch anything, so
you have to inspect elements first
IMGelements that haven't already been fetched
(naturalWidth==0), have external URLs (e.g. not data-
uri:) and that we haven't seen before.
SCRIPTelements that have a srcset
IFRAMEselements that don't have javascript:or
about:protocols
LINKelements that have a hrefset
MUTATIONOBSERVER
Downsides:
Not 100% supported in today's market
Can't be used to monitor all resources (e.g. fonts from
CSS)
MUTATIONOBSERVER
Polyfills (with performance implications):
github.com/webcomponents/webcomponentsjs
github.com/megawac/MutationObserver.js
WHY NOT
RESOURCETIMING?
Doesn't ResourceTiminghave all of the data we need?
ResourceTiming events are only added to the buffer after
they complete
In order to extend the SPA navigation end time, we have
to know if any resource fetches are outstanding
MUTATIONOBSERVER
Polyfill ResourceTiming via MutationObserver
For extra credit, you could use the data you gathered with
Mutation Observer to create a Waterfall for browsers that
don't support ResourceTiming but do support
MutationObserver (e.g. iOS).
MUTATIONOBSERVER
Polyfill ResourceTiming via MutationObserver
FRONT-END VS. BACK-
END
In a traditional page load:
FRONT-END VS. BACK-
END
Traditional websites:
Back-End: HTML fetch start to HTML response start
Front-End: Total Time - Back-End
FRONT-END VS. BACK-
END
Single Page Apps:
Depends on your application's patterns, but...
Back-End: Any timeslice with an XHR outstanding
Front-End: Total Time - Back-End
MONITORING PAGE
COMPONENTS
It's not just about navigations
What about components, widgets and ads?
You can apply the previous techniques to page components
For measuring performance, you need a start time and an end time
The start time is probably driven by your code (e.g. a XHR fetch) or a user
interaction (e.g. a click)
The end time can be measured via XHR interception, MutationObservers,
or callbacks from your resource fetches
MONITORING PAGE
COMPONENTS
How do you measure visual completion?
Challenges:
When an IMGhas been fetched, that's not when it's
displayed to the visitor (it has to decode, etc.)
When you put HTML into the DOM, it's not immediately on
the screen
MONITORING PAGE
COMPONENTS
Use setTimeout(..., 0)or setImmediateto get a
callback after the browser has finished parsing some DOM
updates
varxhr=newXMLHttpRequest();
xhr.open("GET","/fetchstuff");
xhr.addEventListener("load",function(){
$(document.body).html(xhr.responseText);
setTimeout(function(){
varendTime=Date.now();
varduration=endTime-startTime;
},0);
});
varstartTime=Date.now();
xhr.send();
MONITORING PAGE
COMPONENTS
This isn't perfect:
The browser may be doing layout, rendering or drawing
async or on another thread
But it's better than ignoring all the work the browser has
to do to render DOM changes
LIFECYCLE
What happens over time?
How well does your app behave?
LIFECYCLE
It's not just about measuring interactions or how long
components take to load
Tracking metrics over time can highlight performance,
reliability and resource issues
LIFECYCLE
You could measure:
Memory usage: window.performance.memory
(Chrome)
DOM Length:
document.documentElement.innerHTML.length
DOM Nodes:
document.getElementsByTagName("*").length
JavaScript errors: window.onerror
Bytes fetched: ResourceTiming2or XHRs
Frame rate: requestAnimationFrame
THE FUTURE!
OK, that sounded like a lot of work-arounds to measure
Single Page Apps.
Yep.
Why can't the browser just tell give us performance data for
SPAs in a better, more performant way?
LISTENING FOR
RESOURCE FETCHES
Instead of instrumenting XMLHttpRequestand using
MutationObserverto find new elements that will fetch:
W3C Fetch standard
A Fetch Observer
( ) that notifies
us when a resource fetch starts/stops
Less overhead than MutationObserver
Tracks all resources rather than just DOM elements from
MutationObserver
https://fetch.spec.whatwg.org/
https://github.com/whatwg/fetch/issues/65
THANKS!
- -slideshare.net/nicjansma/ @nicj @bluesmoon
soasta.io/SPAperfbook

More Related Content

PDF
DAST in CI/CD pipelines using Selenium & OWASP ZAP
PPTX
Web Application Penetration Testing Introduction
DOCX
Snapchat Usability Testing
PDF
Scouter와 influx db – grafana 연동 가이드
ODP
Gatling - Stress test tool
PPTX
How to Improve Performance Testing Using InfluxDB and Apache JMeter
PPTX
Communication in android
PDF
Top 50 Performance Testing Interview Questions | Edureka
DAST in CI/CD pipelines using Selenium & OWASP ZAP
Web Application Penetration Testing Introduction
Snapchat Usability Testing
Scouter와 influx db – grafana 연동 가이드
Gatling - Stress test tool
How to Improve Performance Testing Using InfluxDB and Apache JMeter
Communication in android
Top 50 Performance Testing Interview Questions | Edureka

What's hot (20)

PDF
Introduction to Force.com Canvas Apps
PDF
e2e testing with cypress
PPTX
Building MAUI UI in C#.pptx
PPTX
다음카카오 Pc&Mobile 웹 접근성 개선 사례
PDF
What is shodan
PDF
A Threat Hunter Himself
PPTX
PPTX
Android application analyzer
PPT
Automated Testing with Agile
PDF
Digital Marketing Automation with Salesforce Marketing Cloud
PDF
클라우드 기반 Unity 게임 서버 구축, 60분이면 충분하다
PPTX
GUI Testing
PPTX
Platform Events by Tim Taylor
PPTX
Salesforce integration best practices columbus meetup
PDF
LWC Episode 3- Component Communication and Aura Interoperability
DOC
38475471 qa-and-software-testing-interview-questions-and-answers
PPTX
Build a Mobile App with Google Forms and AppSheet
PDF
Marcel Hild - Spryker (e)commerce framework als Alternative zu traditioneller...
PDF
KrishnaToolComparisionPPT.pdf
PDF
Zabbix - Gerenciando relatórios personalizados com Jasper Reports
Introduction to Force.com Canvas Apps
e2e testing with cypress
Building MAUI UI in C#.pptx
다음카카오 Pc&Mobile 웹 접근성 개선 사례
What is shodan
A Threat Hunter Himself
Android application analyzer
Automated Testing with Agile
Digital Marketing Automation with Salesforce Marketing Cloud
클라우드 기반 Unity 게임 서버 구축, 60분이면 충분하다
GUI Testing
Platform Events by Tim Taylor
Salesforce integration best practices columbus meetup
LWC Episode 3- Component Communication and Aura Interoperability
38475471 qa-and-software-testing-interview-questions-and-answers
Build a Mobile App with Google Forms and AppSheet
Marcel Hild - Spryker (e)commerce framework als Alternative zu traditioneller...
KrishnaToolComparisionPPT.pdf
Zabbix - Gerenciando relatórios personalizados com Jasper Reports
Ad

Viewers also liked (20)

PPTX
Testing your Single Page Application
PDF
Measuring Real User Performance in the Browser
PPTX
Testing Single Page Webapp
PDF
Accelerated Mobile Pages
PDF
Measuring Continuity
PDF
AMP Accelerated Mobile Pages - Getting Started & Analytics
PPTX
Single Page Applications with AngularJS 2.0
PPTX
Get AMP'ed for Accelerated Mobile Pages - SEO Grail Philadelphia 1/20/16
PDF
The Dark Side of Single Page Applications
PDF
AngularJS with RequireJS
PDF
Understanding User Experience Design_V2
PPT
Intro to IA/IxD/UXD in the agency world
PPT
System Informacyjny A System Informatyczny Prezentacja
PDF
Using Phing for Fun and Profit
PPTX
Using Modern Browser APIs to Improve the Performance of Your Web Applications
PDF
The Happy Path: Migration Strategies for Node.js
PPTX
HTTP Application Performance Analysis
PPTX
Appcelerator Titanium Intro
PDF
Appcelerator Titanium Intro (2014)
PDF
Sails.js Intro
Testing your Single Page Application
Measuring Real User Performance in the Browser
Testing Single Page Webapp
Accelerated Mobile Pages
Measuring Continuity
AMP Accelerated Mobile Pages - Getting Started & Analytics
Single Page Applications with AngularJS 2.0
Get AMP'ed for Accelerated Mobile Pages - SEO Grail Philadelphia 1/20/16
The Dark Side of Single Page Applications
AngularJS with RequireJS
Understanding User Experience Design_V2
Intro to IA/IxD/UXD in the agency world
System Informacyjny A System Informatyczny Prezentacja
Using Phing for Fun and Profit
Using Modern Browser APIs to Improve the Performance of Your Web Applications
The Happy Path: Migration Strategies for Node.js
HTTP Application Performance Analysis
Appcelerator Titanium Intro
Appcelerator Titanium Intro (2014)
Sails.js Intro
Ad

Similar to Measuring the Performance of Single Page Applications (20)

PPTX
Single page applications the basics
PDF
The Characteristics of a Successful SPA
PDF
AppSphere 15 - How Your Monitoring Strategy Needs to Evolve for Single Page Apps
PDF
Can Single Page Applications Deliver a World-Class Web UX?
PPT
single page application
PPTX
Single page applications
PPTX
Single page applications - TernopilJS #2
PPTX
Single Page Applications: Your Browser is the OS!
PDF
Seo for single page applications
PPTX
Making Single Page Applications (SPA) faster
PPTX
Velocity spa faster_092116
PPTX
DIGIT Noe 2016 - Overview of front end development today
PDF
WebApp / SPA @ AllFacebook Developer Conference
PDF
Single Page Application (SPA): A Comprehensive Guide for Beginners
PDF
27.1.2014, Tampere. Perinteinen mobiilimaailma murroksessa. Petri Niemi: Sing...
PDF
Building for, perceiving and measuring performance for mobile web
PPTX
To SPA or not to SPA
PPTX
Single page App
PPTX
WebNetConf 2012 - Single Page Apps
PDF
Offline-First Progressive Web Apps
Single page applications the basics
The Characteristics of a Successful SPA
AppSphere 15 - How Your Monitoring Strategy Needs to Evolve for Single Page Apps
Can Single Page Applications Deliver a World-Class Web UX?
single page application
Single page applications
Single page applications - TernopilJS #2
Single Page Applications: Your Browser is the OS!
Seo for single page applications
Making Single Page Applications (SPA) faster
Velocity spa faster_092116
DIGIT Noe 2016 - Overview of front end development today
WebApp / SPA @ AllFacebook Developer Conference
Single Page Application (SPA): A Comprehensive Guide for Beginners
27.1.2014, Tampere. Perinteinen mobiilimaailma murroksessa. Petri Niemi: Sing...
Building for, perceiving and measuring performance for mobile web
To SPA or not to SPA
Single page App
WebNetConf 2012 - Single Page Apps
Offline-First Progressive Web Apps

More from Nicholas Jansma (8)

PPTX
Modern Metrics (2022)
PPTX
Check Yourself Before You Wreck Yourself: Auditing and Improving the Performa...
PDF
When Third Parties Stop Being Polite... and Start Getting Real
PDF
Reliably Measuring Responsiveness
PDF
Forensic Tools for In-Depth Performance Investigations
PDF
Javascript Module Patterns
PDF
Make It Fast - Using Modern Browser Performance APIs to Monitor and Improve t...
PPTX
Debugging IE Performance Issues with xperf, ETW and NavigationTiming
Modern Metrics (2022)
Check Yourself Before You Wreck Yourself: Auditing and Improving the Performa...
When Third Parties Stop Being Polite... and Start Getting Real
Reliably Measuring Responsiveness
Forensic Tools for In-Depth Performance Investigations
Javascript Module Patterns
Make It Fast - Using Modern Browser Performance APIs to Monitor and Improve t...
Debugging IE Performance Issues with xperf, ETW and NavigationTiming

Recently uploaded (20)

PDF
Build a system with the filesystem maintained by OSTree @ COSCUP 2025
PDF
Encapsulation theory and applications.pdf
PDF
7 ChatGPT Prompts to Help You Define Your Ideal Customer Profile.pdf
PDF
Getting Started with Data Integration: FME Form 101
PDF
Diabetes mellitus diagnosis method based random forest with bat algorithm
PPT
Teaching material agriculture food technology
PPTX
KOM of Painting work and Equipment Insulation REV00 update 25-dec.pptx
PDF
MIND Revenue Release Quarter 2 2025 Press Release
PDF
Reach Out and Touch Someone: Haptics and Empathic Computing
PPTX
TechTalks-8-2019-Service-Management-ITIL-Refresh-ITIL-4-Framework-Supports-Ou...
PPTX
cloud_computing_Infrastucture_as_cloud_p
PPTX
Spectroscopy.pptx food analysis technology
PDF
Network Security Unit 5.pdf for BCA BBA.
PDF
Video forgery: An extensive analysis of inter-and intra-frame manipulation al...
PDF
Univ-Connecticut-ChatGPT-Presentaion.pdf
PPTX
A Presentation on Artificial Intelligence
PDF
Blue Purple Modern Animated Computer Science Presentation.pdf.pdf
PPTX
Digital-Transformation-Roadmap-for-Companies.pptx
PDF
Empathic Computing: Creating Shared Understanding
PDF
Heart disease approach using modified random forest and particle swarm optimi...
Build a system with the filesystem maintained by OSTree @ COSCUP 2025
Encapsulation theory and applications.pdf
7 ChatGPT Prompts to Help You Define Your Ideal Customer Profile.pdf
Getting Started with Data Integration: FME Form 101
Diabetes mellitus diagnosis method based random forest with bat algorithm
Teaching material agriculture food technology
KOM of Painting work and Equipment Insulation REV00 update 25-dec.pptx
MIND Revenue Release Quarter 2 2025 Press Release
Reach Out and Touch Someone: Haptics and Empathic Computing
TechTalks-8-2019-Service-Management-ITIL-Refresh-ITIL-4-Framework-Supports-Ou...
cloud_computing_Infrastucture_as_cloud_p
Spectroscopy.pptx food analysis technology
Network Security Unit 5.pdf for BCA BBA.
Video forgery: An extensive analysis of inter-and intra-frame manipulation al...
Univ-Connecticut-ChatGPT-Presentaion.pdf
A Presentation on Artificial Intelligence
Blue Purple Modern Animated Computer Science Presentation.pdf.pdf
Digital-Transformation-Roadmap-for-Companies.pptx
Empathic Computing: Creating Shared Understanding
Heart disease approach using modified random forest and particle swarm optimi...

Measuring the Performance of Single Page Applications

  • 1. MEASURING THE PERFORMANCE OF SINGLE PAGE APPLICATIONS | | |nic jansma SOASTA nicj.net @nicj | | |philip tellis SOASTA bluesmoon.info @bluesmoon
  • 3. WHO ARE WE? Nic Jansma SOASTA Philip Tellis SOASTA
  • 5. RUM Real User Monitoring Gathering performance metrics from real user experiences Versus Synthetic Monitoring, with emulated users in a controlled environment
  • 6. RUM: HOW IT'S DONE JavaScript measures the browser's events and performance interfaces Listen for readyStatechanges and the onloadevent Measure DNS, TCP, SSL, Request and Response times from NavigationTiming and user measurements from UserTiming (if available) Gather User Agent characteristics (Version, Screen Size, etc) Beacon this data back to the cloud for analytics
  • 12. BOOMERANG Created by Philip Tellis @ Yahoo Gathers performance metrics and characteristics of page load and beacons data to your server (aka RUM) Open-source project (with contributions from SOASTA) https://github.com/lognormal/boomerang/
  • 13. SPAS SINGLE PAGE APPS Run on a single page, dynamically bringing in content as necessary Built with frameworks like AngularJS, Ember.js, Backbone.js, React, etc.
  • 14. SPAS HARD VS. SOFT NAVIGATIONS Hard Navigation: The first page load, which will include all static HTML, JavaScript, CSS, the SPA framework itself (e.g. angular.js), plus showing the initial route Soft Navigation: Any subsequent route (address bar) change Any URL might be loaded via either hard or soft navigation
  • 15. 3 CHALLENGES OF MEASURING THE PERFORMANCE OF SPAS
  • 16. CHALLENGE #1 THE ONLOAD EVENT NO LONGER MATTERS Traditional websites: On navigation, the browser begins downloading all of the JavaScript, CSS, images and other static resources Once all static resources are fetched, the body's onload event will fire This is traditionally what websites consider as page load complete This is traditionally what RUM measures
  • 18. CHALLENGE #1 THE ONLOAD EVENT NO LONGER MATTERS Single Page Apps: Load all static content like a traditional website The frameworks' code will also be fetched (e.g. angular.js) (the onload event fires here) Once the SPA framework is loaded, it starts looking at routes, fetching views and data All of this content is fetched after the onloadevent
  • 20. SPA WATERFALL Browser fires onloadat 1.225 seconds All visual resources (.jpgs) aren't complete until after 1.7 seconds Filmstrip confirms nothing is shown until around 1.7 seconds onloadfired 0.5 seconds too early!
  • 21. CHALLENGE #1 THE ONLOAD EVENT NO LONGER MATTERS Single Page Apps: Core problem is that most of the interesting stuff (e.g. fetching images, JavaScript, CSS and XHRs for the route) happens after the onload The browser doesn't fire any "fully loaded"-style events after onload
  • 22. CHALLENGE #2 SOFT NAVIGATIONS ARE NOT REAL NAVIGATIONS Each route change, user interaction, or visual update is dynamically fetched from the server There are APIs to change the URL (and detect changes) in the address bar without actually navigating New content is dynamically swapped in over the old content The browser is no longer doing a traditional navigation, where it's tearing down the old page
  • 23. CHALLENGE #2 SOFT NAVIGATIONS ARE NOT REAL NAVIGATIONS This is great for performance The browser is no longer re-rendering the same header, footer or common components The browser is no longer re-parsing the same HTML, JavaScript and CSS
  • 24. CHALLENGE #2 SOFT NAVIGATIONS ARE NOT REAL NAVIGATIONS Bad for traditional RUM tools: Stop caring after the measuring the "one" navigation Won't run again until the next time it loads on a full navigation Browser events (readyState, onload) and metrics (NavigationTiming) are all geared toward a single load event
  • 25. CHALLENGE #3 THE BROWSER WON’T TELL YOU WHEN ALL RESOURCES HAVE BEEN DOWNLOADED The browser fires onloadonly once The onloadevent helps us know when all static content was fetched In a soft navigation scenario, the browser does not fire the onloadevent again, so we don't know when its content was fetched
  • 26. CHALLENGE #3 THE BROWSER WON’T TELL YOU WHEN ALL RESOURCES HAVE BEEN DOWNLOADED SPA soft navigations may fetch: Templates Images CSS JavaScript XHRs Videos
  • 27. CHALLENGE #3 THE BROWSER WON’T TELL YOU WHEN ALL RESOURCES HAVE BEEN DOWNLOADED SPA frameworks often fire events around navigations. AngularJS events: $routeChangeStart: When a new route is being navigated to $viewContentLoaded: Emitted every time the ngView content is reloaded But neither of these events have any knowledge of the work they trigger, fetching new IMGs, CSS, JavaScript, etc!
  • 30. HOW CAN WE MEASURE SPA NAVIGATIONS? We need to figure out at what point the navigation started (the start event), through when we consider the navigation complete (the end event).
  • 31. THE START EVENT For hard navigations: The start event is when the browser starts the process of loading the next page This is the same time as with traditional web app navigations We can use NavigationTiming's navigationStartif available, to know when the browser navigation began If NavigationTiming isn't available, and the user is navigating between pages on the same site, you can use cookies to measure when the navigation began (see Boomerang for an implementation)
  • 32. THE START EVENT Challenge #2: Soft navigations are not real navigations We need to figure out when the user's view is going to significantly change The browser history is changing SPA framework routing events can give us an indicator that the view will be changing Other important events that might indicate a view change are a user click, or an XHR that triggers DOM changes
  • 33. THE START EVENT: HISTORY STATE The window.historyobject can tell us when the URL is changing: When pushStateor replaceStateare being called, the app is possibly updating its view When the user hits Back or Forward, the window.popstateevent is fired, and the app will possibly update the view (future events will give us more info)
  • 34. THE START EVENT: ROUTING SPA frameworks fire routing events when the view is changing: AngularJS: $rootScope.$on("$routeChangeStart") Ember.js: beforeModelor willTransition Backbone.js: router.on("route")
  • 35. THE START EVENT: CLICKS When the user has clicks something, they might be doing simple interactions (e.g. a drop-down menu) Or, they might be triggering a UI update (future events will give us more info)
  • 36. THE START EVENT: XHRS An XMLHttpRequest(network activity) might indicate that the page's view is being updated Or, it could be a periodic poller (e.g. a scoreboard update) Or, it could be in reaction to a user interaction (e.g. autocomplete) (future events will give us more info)
  • 37. THE START EVENT To determine if a user click or XHR is really triggering a navigation, we can listen to what happens next If there was a lot of subsequent network activity, we can keep on listening for more events If history (address bar) changed, we can consider the event the start of a navigation If the DOM was updated significantly, we can consider the event the start of a navigation If nothing else happened, it was probably just an insignificant interaction
  • 39. THE END EVENT When do we consider the SPA navigation complete? There are many definitions of complete: When all networking activity has completed When the UI is visually complete (above-the-fold) When the user can interact with the page
  • 40. THE END EVENT Traditional RUM measures up to the onloadevent: This is when all resources have been fetched The page isn't fully loaded until at least then The UI might have been above-the-fold visually complete already It's traditionally when the user can fully interact with the page
  • 41. SINGLE POINTS OF FAILURE (SPOFS) Which resources could affect visual completion of the page? External JavaScript files External CSS files Media (images, video)
  • 42. THE END EVENT For hard navigations, the onloadevent no longer matters (Challenge #1) The onloadevent only measures up to when all static resources were fetched The SPA framework will be dynamically loading its UI only after the static JavaScript has been loaded We want to mark the end of the hard navigation only after all of the resources were fetched and the UI is complete
  • 43. THE END EVENT For soft navigations, the browser won’t tell you when all resources have been downloaded (Challenge #3) The onloadonly fires once on a page APIs like ResourceTiming can give you details about network resources after they've been fetched But to know when to stop, we need to know if there are any outstanding resources So let's monitor all network activity!
  • 44. THE END EVENT Let's make our own SPA onloadevent: Similar to the body onloadevent, let's wait for all network activity to complete This means we will have to intercept both implicit resource fetches (e.g. from new DOM elements) as well as programmatic (e.g. XHR) resource fetches
  • 45. MONITORING XHRS XMLHttpRequests play an important role in SPA frameworks XHRs are used to fetch HTML, templates, JSON, XML, data and other assets We should monitor to see if any XHRs are occuring The XMLHttpRequestobject can be proxied Intercept the .open()and .send()methods to know when an XHR is starting
  • 46. MONITORING XHRS Simplified code ahead! Full code at github.com/lognormal/boomerang/blob/master/plugins/auto_xhr
  • 48. MONITORING XHRS By proxying the XHR code, you can: Know which URLs are being fetched Know when a XHR has started Know when a XHR has completed, timed out, error or aborted Measure XHR states even on browsers that don't support ResourceTiming Most importantly, know if there are any outstanding XHRs
  • 49. MONITORING XHRS Downsides: Need additional code to support XDomainRequest Timing not as accurate when browser is busy (rendering, etc) as callbacks will be delayed You can fix-up timing via ResourceTiming (if available)
  • 50. OTHER RESOURCES XHR is the main way to fetch resources via JavaScript What about Images, JavaScript, CSS and other HTML elements that trigger resource fetches? We can't proxy the Imageobject as that only works if you create a new Image()in JavaScript If only we could listen for DOM changes...
  • 51. MUTATION OBSERVER : http://developer.mozilla.org/en- US/docs/Web/API/MutationObserver MutationObserver provides developers a way to react to changes in a DOM Usage: observe()for specific events Get a callback when mutations for those events occur
  • 52. MUTATIONOBSERVER Simplified code ahead! Full code at github.com/lognormal/boomerang/blob/master/plugins/auto_xhr
  • 55. MUTATIONOBSERVER Simplified workflow: Start listening when an XHR, click, route change or other interesting navigation-like event starts Use MutationObserverto listen for DOM mutations Attach loadand errorevent handlers and set timeouts on any IMG, SCRIPT, LINK or FRAME If an interesting element starts fetching keep the navigation "open" until it completes After the last element's resource has been fetched, wait a few milliseconds to see if it kicked off anything else If not, the navigation completed when the last element's resource was fetched
  • 56. MUTATIONOBSERVER What's interesting to observe? Internal and cached resources may not fetch anything, so you have to inspect elements first IMGelements that haven't already been fetched (naturalWidth==0), have external URLs (e.g. not data- uri:) and that we haven't seen before. SCRIPTelements that have a srcset IFRAMEselements that don't have javascript:or about:protocols LINKelements that have a hrefset
  • 57. MUTATIONOBSERVER Downsides: Not 100% supported in today's market Can't be used to monitor all resources (e.g. fonts from CSS)
  • 58. MUTATIONOBSERVER Polyfills (with performance implications): github.com/webcomponents/webcomponentsjs github.com/megawac/MutationObserver.js
  • 59. WHY NOT RESOURCETIMING? Doesn't ResourceTiminghave all of the data we need? ResourceTiming events are only added to the buffer after they complete In order to extend the SPA navigation end time, we have to know if any resource fetches are outstanding
  • 60. MUTATIONOBSERVER Polyfill ResourceTiming via MutationObserver For extra credit, you could use the data you gathered with Mutation Observer to create a Waterfall for browsers that don't support ResourceTiming but do support MutationObserver (e.g. iOS).
  • 62. FRONT-END VS. BACK- END In a traditional page load:
  • 63. FRONT-END VS. BACK- END Traditional websites: Back-End: HTML fetch start to HTML response start Front-End: Total Time - Back-End
  • 64. FRONT-END VS. BACK- END Single Page Apps: Depends on your application's patterns, but... Back-End: Any timeslice with an XHR outstanding Front-End: Total Time - Back-End
  • 65. MONITORING PAGE COMPONENTS It's not just about navigations What about components, widgets and ads? You can apply the previous techniques to page components For measuring performance, you need a start time and an end time The start time is probably driven by your code (e.g. a XHR fetch) or a user interaction (e.g. a click) The end time can be measured via XHR interception, MutationObservers, or callbacks from your resource fetches
  • 66. MONITORING PAGE COMPONENTS How do you measure visual completion? Challenges: When an IMGhas been fetched, that's not when it's displayed to the visitor (it has to decode, etc.) When you put HTML into the DOM, it's not immediately on the screen
  • 67. MONITORING PAGE COMPONENTS Use setTimeout(..., 0)or setImmediateto get a callback after the browser has finished parsing some DOM updates varxhr=newXMLHttpRequest(); xhr.open("GET","/fetchstuff"); xhr.addEventListener("load",function(){ $(document.body).html(xhr.responseText); setTimeout(function(){ varendTime=Date.now(); varduration=endTime-startTime; },0); }); varstartTime=Date.now(); xhr.send();
  • 68. MONITORING PAGE COMPONENTS This isn't perfect: The browser may be doing layout, rendering or drawing async or on another thread But it's better than ignoring all the work the browser has to do to render DOM changes
  • 69. LIFECYCLE What happens over time? How well does your app behave?
  • 70. LIFECYCLE It's not just about measuring interactions or how long components take to load Tracking metrics over time can highlight performance, reliability and resource issues
  • 71. LIFECYCLE You could measure: Memory usage: window.performance.memory (Chrome) DOM Length: document.documentElement.innerHTML.length DOM Nodes: document.getElementsByTagName("*").length JavaScript errors: window.onerror Bytes fetched: ResourceTiming2or XHRs Frame rate: requestAnimationFrame
  • 72. THE FUTURE! OK, that sounded like a lot of work-arounds to measure Single Page Apps. Yep. Why can't the browser just tell give us performance data for SPAs in a better, more performant way?
  • 73. LISTENING FOR RESOURCE FETCHES Instead of instrumenting XMLHttpRequestand using MutationObserverto find new elements that will fetch: W3C Fetch standard A Fetch Observer ( ) that notifies us when a resource fetch starts/stops Less overhead than MutationObserver Tracks all resources rather than just DOM elements from MutationObserver https://fetch.spec.whatwg.org/ https://github.com/whatwg/fetch/issues/65
  • 74. THANKS! - -slideshare.net/nicjansma/ @nicj @bluesmoon soasta.io/SPAperfbook