4 September 2009 | Web Development

LIVE at DrupalCon Paris 2009: How I learned to stop worrying and love the cloud

Session: Dr Clonelove: How I learned to stop worrying and love the cloud
Speaker: Peter Brownell, Co-Founder of School Of Everything, a Drupal based start-up in London. He is also a Director of Code Positive, and has been involved in the London Drupal Community for many years.

Session: How I learned to stop worrying and love the cloud

Notes

MySQL proxy changes the game.

“Clustering involves very high documentation. We run three servers. The idea is to move to small instances of dataases. Amazon gives you two main cluster machines. Large instances are 4 times more expensive than the small one. “

“When we began, it was obvious nothing was persistent.”

Handling failures

If your machine cannot be reconfigured, it can’t be in a cloud. You need to know how your machine was set up. Do things well. Assume that any part of the cloud can die.

  • How do we go about reconfiguring that?
  • Be able to rebuild from backups.
  • Document and blog.
  • Fire drill – practice makes perfect.

Rebuild debian host. Rebuild Drupal site. Drupal’s easy to rebuild. In emergencies, put in on a single server and temporarily disable some admin stuff that you don’t need; and then put it in front of a proxy.

Cloud City

Never have faith in just one ally. Set up a base off the cloud. Keep a backup on the ground. Monitor from the outside. Mail smarthost to avoid auto greylisting.

Foreseeing problems

Monitor as much as possible. Don’t trust anything. Look for clues for where problems arise. Do capacity planning. “We used Nagios and Munin”.

Hardware

The physical world moves a little slower. Virtual machines are cheap but you need to plan. Configure one original and make many copies, in multiple zones. Only pay for the time you use. It’s not two of everything always, it’s a spare when you need it.

The pieces putting the cluster together

  • How safe is your data on the storage? “For the first month, we ran our MySQL database on EC2.
  • Write to S3, read from local DB.
  • Elastic Block Store – EBS.
  • Point-in-time snapshots.

Clone 2 Clone Filesystem

  • Avoid dependencies
  • ClusterFS sync files instantly
  • Primary webserver uses persistent storage
  • Deliver images off S3. Drupal cloudfront modules makes it easy. Upgrade to Amazon’s CDN for a bit more.
  • Round Robin DNS is a cheap solution for load balancing.

Summary

Cloud forces best practice. No time for slacker sysadmins!
Avoid dependencies. Work backwards
Document!

0 Comments

Welcome to my blog, where I write about art, tech and social change

Articles

Sponsored Links

Twitter Updates

    More »

    Sites I Read

    Rebels