DevEX - reference for building teams, processes, and platforms
DOs and DONTs on the way to 10M users
1. DOs and DON'Ts on the way to over 10M users… IGT – Event July 2011
2. Wix Initial Architecture Server Public Sites Serving Sites Editing API User Authentication User Media upload, searches Public Media upload, management, searches Template Galleries management Database Wix Sites WixML Pages User Authentication Users Media Public Media Template Galleries
3. Wix Initial Architecture Tomcat, Hibernate, Custom web framework Everything generated from HBM files Built for fast development Statefull login (tomcat session), EHCache, File uploads Not considering performance, scalability, fast feature rollout, testing It reflected the fact that we didn’t really know what is our business Yoav A. - “it is great for a first stage start-up, but you will have to replace it within 2 years” Nadav A, after two years - “you were right, however, you failed to mention how hard it’s gonna be”
4. Wix Initial Architecture What we have learned Don’t worry about ‘building it right from the start’ – you won’t You are going to replace stuff you are building in the initial stages of a startup or any project Be ready to do it Get it up to customers as fast as you can. Get feedback. Evolve. Our mistake was not planning for gradual re-write Build for gradual re-write as you learn the problems and find the right solutions
5. Distributed Cache Next we added EHCache as Hibernate 2nd-level cache Why? Cause it is in the design How was it? Black Box cache How do we know what is the state of our system? How to invalidate the cache? When to invalidate it? How does operations manage the cache? Did we really need it? No We eventually dropped it
6. Distributed Cache What we have learned You don’t need a Cache Really, you don’t Cache is not part of an architecture. It is a means to make something more efficient Architect while ignoring the caching. Introduce caching only as needed to solve real performance problems. When introducing a cache, think about cache management – how do you know what is in the cache? How do you find invalid data in the cache? Invalidation – who invalidates the cache? When? Why? Cache Reset – can your architecture stand a cache restart?
7. Two years passed We learned what our business is – building websites We started selling premium websites
8. Two years passed Our architecture evolved We added a separate Billing segment We moved static file storage and serving to a separate instance But we started seeing problems Updates to our server imposed complete wix downtime Our static storage reached 500 GByte of small files, the limit of Bash scripts The server codebase became large and entangled. Feature rollout became harder over time, requiring longer and longer manual regression Strange full-table scans queries generated by Hibernate, which we still have no idea what code is responsible for… Statefull user sessions required a lot of memory and a statefull load balancer BillingDB Billing (Tomcat) Server (Tomcat) DB statics (Lighttpd) Media storage
9. Editor & Public Segments The Challenge - Updates to our Server imposed downtime for our customer’s websites Any Server or Database update has the potential of bringing down all Wix sites Is a symptom of a larger issue The Server served two different concerns Wix Users editing websites Viewing Wix Sites, the sites created by the Wix editor The two concerns require a different SLA Wix Sites should never ever have a downtime! Wix Sites should work as fast as possible, always! However, an editing system does not require this level of SLA. The two concerns evolve independently Releases of Editing feature should have no impact on existing Wix sites operations!
10. Editor & Public Segments Our Solution Split the Server into two Segments – Public and Editor The Public segment targets serving websites for Wix Users Has mostly read-only usage pattern – only updated on when a site is published Simple publishing system Simple means it is easier to have higher SLA Simple and read-only means it is easier to make DRP Built using Spring MVC 2.5 and JDBC (without Hibernate) MySQL used as NoSQL – single large table with XML text fields The Editor segment Exposes the Wix Editing APIs, as well as user account and galleries management APIs. Has different release schedule compared to the Public segment Public (Tomcat) Public DB Editor (Tomcat) Editor DB
11. Editor & Public Segments What we have learned Architecture is inspired by aspects such as SLA Release Cycles – deployment flexibility Separate Segments for discrete concerns Editing (Editor Segment) Publishing (Public Segment) modularity – in terms of breaking the architecture to services / modules / sub-systems (or any other name) Enabler for gradual re-write Enabler for continues delivery Simplifies QA, Operations & Release Cycles Public (Tomcat) Public DB Editor (Tomcat) Editor DB
12. Editor & Public Segments What we have learned MySQL is a damn good NoSQLengine Our public DB is (mainly) one table with 10M entries Queries are by primary key Updates are by primary key Instead of relations, we use text/xml columns Immutable Data Architect your data to be immutable MySql sequential keys cause problems Locks on key generation Require a single instance to generate keys We use GUID keys Can be generated by any client No locks in key value generation Enabler for Master-Master replication Public (Tomcat) Public DB Editor (Tomcat) Editor DB
13. Authentication Segment The Challenge – Statefull user sessions Makes it harder to separate software to different segments Requires shared library or single sign-on Requires statefull load balancer Takes more memory Our solution Dedicated authentication segment – authentication is a specific concern Cookie based login – solves all the above problems Implement stronger security solution only where really required What we have learned Do not use statefull sessions – they don’t scale (easily) Use cookie based login – this is the standard on the web today
14. The Challenge – Our static storage reached over 500 GByte of small files The “upload to app server, copy to static server, serve by file server” pattern proved inefficient, slow and error prone Disk IO became slow and inefficient as the number of files increased We needed a solution we can grow with – HTTP connections number of files We needed control over caching and Http headers Our Solution Prospero You’ve heard about it in a previous presentation
15. What we have learned It was hard to make Prospero work right But it was the right choice for us Media files caching headers are critical Max-age, ETag, if-modified-since, etc. Think how to tune those parameters for media files, as per your specific needs We tried using Amazon S3 as secondary storage However, they proved unreliable The S3 service had connection reliability issues S3 have “lost” a 200M files, 30 Tbyte volume We found that using a CDN in front of Prospero is very affective
16. CDN What we have learned Use a CDN! CDN acts as a great connection manager We have CDN hit ratio’s of over 99.9% Use the “Cache Killer” pattern http://static.wix.com/client/css/viewer.css?v=327 Makes flushing files from the CDN redundant Enabler for longer caching periods There are many vendors We started with 1 CDN vendor We are now “playing” with multiple CDN vendors Different CDN vendors have advantages at different geo Tune HTTP Headers per CDN Vendor CDN Vendors interpret HTTP heads differently
17. Wix-Framework / TDD / DevOps The Challenge The server codebase became large and entangled Feature rollout became harder over time, requiring longer and longer manual regression The Solution The Wix Framework + TDD / CI / CD / DevOps TDD – Developer responsible for all automated tests With QA Developers CI / CD – automate build & deployment Solve the build / test / deployment / rollback issues first Get the project to production with minimal functionality DevOps - Developer is responsible from Productto 10,000 active users
18. Wix-Framework The Wix Framework Java, Spring, Jetty, MySQL, MongoDB Spring MVC based Adjustments for Flash Flash imposes some restrictions on HTTP which require special handling DevOps support Built-in support for monitoring dashboard, configuration, usage tracking, profiling and Self-Test in every app server Automated Deployment – using Chef and SouceChef TDD Support Unit-Testing helpers Multi-browser Javascript Unit-Testing support, integrated with IDE Integration Testing framework, allows running the same tests in embedded mode (for developer) and deployed in production like env Embedded MySQL & Embedded MongoDB
19. TDD / DevOps Our development process Developer writes code and Unit-Tests Developer and QA Developer write integration tests (IT) Build on developer machine On check-in Mark RC Deploy to QA env Mark GA Deploy to production Compile Unit Tests Embedded ITs Run ITs Compile Unit Tests Self Tests Deploy – IT Env RC QA Self Tests Deploy – QA Env Monitor GA Self Tests Deploy – Production Monitor
20. Wix-Framework / TDD / DevOps What we have learned TDD / CI / CD / DevOps works It enables us to release frequently We release 2-3 times a day With higher confidence in release quality With reduced risk It takes time and effort Requires management commitment Transforms the way R&D, Product and operations work Requires investment in supporting tools – the Wix Framework Infrastructure – Build Server, Artifactory, Maven (+Plugins), Chef, SourceChef, Testing environments (virtual boxes), Server Dashboard, New Relic. It is worth the effort! A Wix developer – “never again shell I work without automated testing”