Embracing the Network – ColdFront, September 2015

The network is intrinsically unreliable. More so, the network is out of your control as a developer. Therefore, we must design systems which embrace the unpredictability of the network and defend against it all costs. How can you prioritise the delivery of your core content? What best-practices can you use to optimise your assets? How can we design interfaces which adapt and respond to changing network conditions? And finally, how are new APIs such as ServiceWorker changing the way we think about the network?

During this talk Patrick will share his experiences delivering high-performance websites to millions of users over the past 3 years at The Guardian and Financial Times. Which – most importantly – are resilient to the network.

Transcript

1. Embracing the network Patrick Hamann — ColdFront Conference, September 2015
 @patrickhamann Modern techniques for building resilient front ends !
4. Why?
5. https://en.wikipedia.org/wiki/Fallacies_of_distributed_computing <a href=“http://next.ft.com”>FT.com</a>
9. https://en.wikipedia.org/wiki/Fallacies_of_distributed_computing 1. The network is reliable. 2. Latency is zero. 3. Bandwidth is infinite. 4. The network is secure. 5. Topology doesn't change. 6. There is one administrator. 7. Transport cost is zero. 8. The network is homogeneous. 
 The fallacies of distributed computing 
 PETER DEUTSCH — SUN MICROSYSTEMS, 1994 https://en.wikipedia.org/wiki/Fallacies_of_distributed_computing
10. Multiple points of failure Dmytrii Shchadei — http://www.slideshare.net/metrofun/reduce-mobile-latency " # # # $ Internal latency Firewalls, Load Balancers, Servers % Internet routing latency CDNs, ISPs, Caches, Proxies Control plane latency ~600ms on average UK 3G connection ~200ms
11. The radio state machine http://developer.android.com/training/efficient-downloads/efficient-network-access.html Radio standby Radio low power Radio full power 2s latency 1.5s latency Radio idle for 5 seconds Radio idle for 12 seconds
12. Title Text Request Map: http://requestmap.webperf.tools/
13. Failure is inevitable • 1,000,000 page views a day • 3 years product life-time • (1000000*360)*3 = 1,080,000,000 • 1 billion opportunities for something to go wrong.
14. Baking in the assumption that everything can and will fail leads you to think differently about how you solve problems. 
 SAM NEWMAN — BUILDING MICROSERVICES, O’REILLY 2015 http://shop.oreilly.com/product/0636920033158.do
15. 1. Testing and monitoring failure 2. Designing for failure 3. Embracing failure 4. The future
16. Testing and monitoring failure
17. A single point of failure (SPOF) is a part of a system that, if it fails, will stop the entire system from working. SPOFs are undesirable in any system with a goal of high availability or reliability, be it a business practice, software application, or other industrial system. https://en.wikipedia.org/wiki/Single_point_of_failure
18. Front end SPOFs http://www.stevesouders.com/blog/2010/06/01/frontend-spof/ Chrome Firefox IE Opera Safari External script Blank below Blank below Blank below Blank below Blank below Stylesheet Flash Flash Blank below Flash Blank below inlined 
 @font-face Delayed Flash Flash Flash Delayed Stylesheet @font-face Delayed Flash Totally blank Flash Delayed Script then
 @font-face Delayed Flash Totally blank Flash Delayed
19. http://uk.businessinsider.com/doubleclick-for-publishers-down-2014-11
20. http://uk.businessinsider.com/doubleclick-for-publishers-down-2014-11
21. http://uk.businessinsider.com/doubleclick-for-publishers-down-2014-11
23. https://twitter.com/scottjehl/status/636303029533282304
24. https://en.wikipedia.org/wiki/Fallacies_of_distributed_computing
25. Jeremy Keith — https://flic.kr/p/o9thWy
26. Resource timing API http://www.w3.org/TR/resource-timing/ var resourceList = window.performance.getEntriesByType(‘resource'); for (i = 0; i < resourceList.length; i++) { if (resourceList[i].initiatorType == 'link' || 'script') { navigator.sendBeacon('https://stats.ft.com/', { resource: resourceList[i].name, timing: (resourceList[i].responseEnd - resourceList[i].startTime) }); } }
29. • Test 1st and 3rd party dependencies for SPOFs • Use network shaping to simulate mobile conditions • Monitor all response times, timeouts and errors • Alert if metrics pass their thresholds • Consider creating chaos in production
30. 1. Testing and monitoring failure 2. Designing for failure 3. Embracing failure 4. The future
31. Designing for failure
33. There seems to be an excessive amount of waiting around for pages to refresh and load -it doesn't seem as quick as the previous version. Luke Wroblewski – http://www.lukew.com/ff/entry.asp?1797
38. Time and perception Delay User perception 0 - 100ms Instant 100 - 300ms Small perceptible delay 300 - 1000ms Machine is working 1000+ ms Likely mental context switch 10,000+ ms Task is abandoned
42. • Respond within a 1000ms • Avoid loading spinners • Consider using skeleton screens • Re-use familiar UI patterns within error messaging • Use human friendly copy, especially in error states • Visualise the state of the network
43. 1. Testing and monitoring failure 2. Designing for failure 3. Embracing failure 4. The future
44. Embracing failure
45. Pre-connect Ilya Grigorik — Eliminating roundtrips wth pre-connect: https://www.igvita.com/2015/08/17/eliminating-roundtrips-with-preconnect/ socket setup
46. Pre-connect <link href='https://media.ft.com/' rel=‘preconnect’> <link href='https://api.livefyre.com/' rel='preconnect' crossorigin> function preconnectTo(url) { var hint = document.createElement("link"); hint.rel = "preconnect"; hint.href = url; document.head.appendChild(hint); } Ilya Grigorik — Eliminating roundtrips wth pre-connect: https://www.igvita.com/2015/08/17/eliminating-roundtrips-with-preconnect/
48. Caching Libraries Utilities Application Rarely change Frequent change
49. Caching Libraries Utilities Application Rarely change Frequent change
50. Caching Libraries Utilities Application Rarely change Frequent change & core.js & app.js
51. HTTP/2 R.U.M data sampled over 1 month from https://next.ft.com/
52. Alternative protocols
53. Alternative protocols Method Communication Support AJAX request → response All major browsers Long poll request → wait → response All major browsers WebSockets client ↔ server Very good WebRTC peer ↔ peer Medium Server-sent events client ← server Good, except IE
55. Loading fonts https://tabatkins.github.io/specs/css-font-rendering/ Chrome Opera Safari Firefox IE Timeout 3s 3s no timeout 3s 0 Fallback Yes Yes n/a Yes Yes Swap Yes Yes n/a Yes Yes
56. CSS Font Rendering Controls Module https://tabatkins.github.io/specs/css-font-rendering/ body { font-display: swap; } .ft-promo { font-display: block; }
57. ' ServiceWorker & ( %
58. ' ServiceWorker & ( % 
59. ' ServiceWorker & ( %
60. Cache only self.addEventListener('install', function(event) { event.waitUntil( caches.open(cacheName).then(function(cache) { return cache.addAll([ 'http://fonts.ft.com/serif', 'http://fonts.ft.com/sans-serif' ]); }) ); }); https://jakearchibald.com/2014/offline-cookbook
61. Cache only https://jakearchibald.com/2014/offline-cookbook self.addEventListener('fetch', function(event) { // If this is a request for font file if (/fonts\.ft\.com/.test(event.request)) { // If a match isn't found in the cache, the response // will look like a connection error event.respondWith(caches.match(event.request)); } });
62. X Cache only ' & ( %  ( X ( ( &
63. Loading CSS https://developers.google.com/web/fundamentals/performance/critical-rendering-path/analyzing-crp?hl=en Request page Network Render GET html Build DOM response Build CSSOM Render page GET css GET js response response idle idle Render blocking Run JS Render blocking
64. Loading CSS Request page Network Render GET html Build DOM response Build CSSOM Render page GET css GET js response response idle idle Async Render blocking Run JS https://developers.google.com/web/fundamentals/performance/critical-rendering-path/analyzing-crp?hl=en
65. https://github.com/filamentgroup/loadCSS
66. Loading CSS Request page Network Render GET html Build DOM response Build CSSOM Render page GET css response idle idle Async https://developers.google.com/web/fundamentals/performance/critical-rendering-path/analyzing-crp?hl=en Reflow
67. Stale-while-revalidate self.addEventListener('fetch', function(event) { // Get stale from cache event.respondWith( caches.open(cacheName).then(function(cache) { return cache.match(event.request).then(function(response) { // If in cache relvalidate var network = fetch(event.request).then(function(response){ if(response.status < 400) { cache.put(event.request, response.clone()); } }); // Return stale or network if none return response || network; }); }) ); }); https://jakearchibald.com/2014/offline-cookbook
68. Stale-while-revalidate ' & ( %  (
69. Stale-if-error self.addEventListener('fetch', function(event) { // Always fetch response from the network event.respondWith( fetch(event.request).then(function(response) { return caches.open(cacheName).then(function(cache) { // If we received an error response if(!response.ok) { return cache.match(event.request) } else { // Response was healthy so update cached version cache.put(event.request, response.clone()); return response; } }); }) ); });
70. Timeouts ' & ( % X * X
71. Timeout race function timeout(delay) { return new Promise(function(resolve, reject) { setTimeout(function() { resolve(new Response('', { status: 408, statusText: 'Request timed out.' })); }, delay); }); } self.addEventListener('fetch', function(event) { // Attempt to fetch with timeout event.respondWith(Promise.race([timeout(2000), fetch(event.request)])); });
72. Circuit breakers https://en.wikipedia.org/wiki/Fallacies_of_distributed_computing
73. Circuit breakers ' & ( % ( X ( ( X X X X Circuit closed (
74. Circuit breakers ' & ( % ( X ( ( X X Circuit open, fail fast (
75. https://github.com/yammer/circuit-breaker-js
76. importScripts('js/circuit-breaker.js'); CircuitBreaker.prototype.fetch = function(request) { return new Promise(function(resolve, reject) { this.run(function(success, fail) { fetch(request).then(function(response) { if(response.ok) { success(); } else { fail(); } resolve(response); }) .catch(function(err) { fail(); reject(Response.error()); }); }, function() { resolve(Response.error()); }); }.bind(this)); };
77. Fetch via circuit breaker var circuitBreakers = {}; var options = { windowDuraion: 1000, timeoutDuration: 3000, errorThreshold: 50, volumeThrehold: 2 }; self.addEventListener('fetch', function(event) { var url = event.request.url; if(!circuitBreakers[url]) { circuitBreakers[url] = new CircuitBreaker(options); } event.respondWith(circuitBreakers[url].fetch(event.request)); });
78. Dead letter queue ' & ( % X + offline
79. Dead letter queue ' & ( % + Online
80. Dead letter queue var queue = {}; function queueFailedRequest(request) { queue[Date.now()] = request; } self.addEventListener('fetch', function(event) { if(/track\.ft\.com/.test(event.request.url) { event.respondWith( fetch(event.request).then(function(response){ if(response.status >= 500) { return Response.error(); } else { return response; } }).catch(function() { queueFailedRequest(event.request); }) ); } });
81. Dead letter queue function replayQueuedRequests() { Object.keys(queue).forEach(function(event) { fetch(queue[event]).then(function(){ if(response.status >= 500) { return Response.error(); } delete queue[error]; }).catch(function() { var timeDelta = Date.now() - event; if (timeDelta > expiration) { delete queue[error]; } }); }); }
82. • Pre-connect sockets • Pre-fetch and cache critical resources • Bundle resources by their frequency of change • Choose appropriate protocols • Queue and batch requests • Use timeouts, bulkheads and circuit breakers • Treat the network as an enhancement!
83. 1. Testing and monitoring failure 2. Designing for failure 3. Embracing failure 4. The future
84. The future
86. Push notifications
87. Offline Network unavaliable
88. Background sync navigator.serviceWorker.ready.then(function(registration) { regitration.periodicSync.register({ tag: 'get-latet-news', // default: '' minPeriod: 12 * 60 * 60 * 1000, // default: 0 powertState: 'avoid-draining', // default: 'auto' networkState: 'avoid-cellular' // default: 'online' }).then(function(periodicsyncReg) { // success }, function() { // failure }); });
89. Progressive apps: • Responsive design • Secure (via TLS) • Network agnostic (via ServiceWorker) • Homescreen access • Push notifications • Background Sync
90. 1. Testing and monitoring failure 2. Designing for failure 3. Embracing failure 4. The future
91. Conclusion
92. • Bake in the assumption that everything can and will fail • Guard against SPOFs at all costs • Always provide visual user feedback • Optimise for efficient networking • Treat the network as an enhancement • Embrace the future
93. Thanks @patrickhamann speakerdeck.com/patrickhamann
 github.com/Financial-Times ! , - Do you want to help build this stuff? Join in. jobs@labs.ft.com