Continuous Full System Deployment
From CitconWiki
Jump to navigationJump to searchThe printable version is no longer supported and may have rendering errors. Please update your browser bookmarks and please use the default browser print function instead.
Tips
- Your snapshots should be as bare as possible
- Use Puppet or Chef to configure it (use the exact same thing that you used to configure production)
- Do basic tests (ping, ssh and the web server is available)
Ensure QA and Production are the same
- Deployable unit was a system image - it has to be in a Debian package
- Issue: If you do an apt-get on QA and pull down packages, that's fine. But 2 days later when deploying again to production, that package may have changed and this can introduce issues with your application. Solution: Have your own repo and cache that version. Rather than
Packages with no dependencies
Live Updating Thousands of Servers
- Image takes too long
Pain Points
- Getting different systems deployed is no problem. How do we
- Clone image/install OS/ get manifests and run them, install app, wait for system to start, run tests. 10 minute delay if you need to test something immediately
