Modern Web development and operations practices

Modern Web development and
operations practices
Grig Gheorghiu
VP Tech Operations
Nasty Gal Inc.
@griggheo
Invited lecture for CSE 491 “Web Development” class at
MSU, courtesy of C. Titus Brown

Modern Web stack
l  Aim for horizontal scalability!
l  Ruby/Python front-end servers (Sinatra/Padrino,
Flask) with heavy-duty JavaScript libraries (jQuery,
bootstrap)
l  Intermediate RESTful API layer (Jetty/Groovy)
l  Back-end services (e-commerce platform, payment
gateway)
l  Indexing/search services: Solr, ElasticSearch
l  Database layer (relational such as MySQL/
PostgreSQL, noSQL such as Riak/Cassandra)

Operations stack
l  Hypervisor: KVM
l  Private/Hybrid cloud: OpenStack, CloudStack,
Eucalyptus
l  Public cloud: Amazon, GCE, Rackspace, Joyent
l  OS: Ubuntu 12.04 LTS
l  HTTP proxy/SSL terminator: nginx
l  Load balancer: haproxy
l  CDN: Akamai, EdgeCast, Fastly, Insight Logic
l  DDOS protection: CloudFlare

Development and deployment
pipeline
l  Github or private Git install
l  Deployment done with Jenkins and Capistrano/Fabric
l  Devs commit and push code which triggers Jenkins
run in development environment (Continuous
Integration)
l  Use Vagrant if you can so dev and prod are on the same OS
l  Unit tests get run
l  QA team deploys to staging environment
l  Integration tests get run including Selenium tests
l  Ops team deploys to production environment

Monitoring, graphing, logging
l  If it's not monitored, it's not in production
l  Monitoring is for Ops what testing is for Dev
l  External monitoring: Pingdom, Keynote
l  Internal monitoring tools: old (Nagios) and new
(Sensu)
l  Graphing tools: Graphite, Cacti, Ganglia, Munin
l  Logging tools: Logstash, ElasticSearch, Kibana (or go
commercial: Splunk)
l  Alerting tools: PagerDuty

Configuration management
l  Indispensable at scale
l  Infrastructure-as-code
l  Chef or Puppet, take your pick
l  We use Chef
l  Wide choice of community 'cookbooks' for major packages
(Apache, nginx, haproxy, mySQL etc)
l  Decent development pipeline (cookbooks pinned to versions
in different environments)
l  Decent testing tools: Foodcritic (lint), Chef Spec (unit testing)
Test Kitchen (integration testing using Vagrant)

What is 'devops'?
l  Open lines of communication and strong collaboration
between Dev, QA and Ops
l  JIRA: ticketing mechanism keeping everybody in the loop
l  Change Management Requests (CMR) for production
changes
l  Jabber for communication ('operations' chat room)
l  Devs need Ops skills; Ops need scripting skills at a
minimum; both need testing and DBA skills (‘full stack’)
l  Automation! It's the cloud era
l  One of the goals of a devops team is to deploy to
production N times a day with no downtime

To cloud or not to cloud
l  Public cloud
l  Advantages: fast provisioning, 'infinite' elasticity
l  Disadvantage: poor reliability, poor performance in some
cases ('noisy neighbors')
l  Private cloud
l  Advantages: tunable performance, increased control,
potentially better security
l  Disadvantage: decreased elasticity unless you are Facebook
l  Hybrid cloud: best of both worlds
l  Start with private cloud but be ready to expand horizontally
into public cloud

Do you need Big Data?
l  Only if the volume of data exceeds a big bare-metal
box running MySQL or PostgreSQL
l  Most people don't need Big Data
l  If you do, try Hadoop/Hive/Pig
l  Export data to S3 in csv.gz format
l  Read it in Hive tables
l  Run Hive/Pig queries, export to MySQL
l  Build dashboards on top of MySQL

Architecture is important
l  Horizontal layers that can be scaled independently
l  Both software and hardware should scale horizontally
l  Take advantage of The Cloud
l  DB layer (hard to scale horizontally if relational, easier if
NoSQL)
l  RESTful API layer
l  Front-end layer – responsive design, no difference between
mobile and desktop
l  Software load balancers in High Availability mode (haproxy +
keepalived)
l  No single points of failure!

Troubleshooting skills are important
•  Interview question: talk about all the different layers
(both software and networking) that are involved in a
user typing www.google.com in a browser and
getting the Google home page back
•  What can go wrong at each layer?
•  Need to understand the major Internet protocols
•  TCP/IP (tcpdump)
•  DNS (dig, nslookup)
•  HTTP (curl, Firebug, twill, Selenium)
•  SMTP

Modern Web development and operations practices

More Related Content

What's hot (20)

Viewers also liked (7)

Similar to Modern Web development and operations practices (20)

Recently uploaded (20)

Modern Web development and operations practices