In the previous drinking champagne blog, I mentioned we have quite a few application and services in our internal production cloud (of course powered by Eucalyptus) and planet is one of them. What we did with planet, is to make it very simple to deploy and customize it: the original work was done by Mark Atwood. In order to take advantage of the cloud we heavily relied on the meta-data service.
The meta-data service allows for instances started in the cloud (private and public clouds which follows the AWS API) to learn data pertinent to the instance itself (hence it is called meta-data). Public IP, ssh keys, storage information, instance IDs (EMIs, kernel and ramdisk) are examples of what can be retrieved. All these data are accessible at http://169.254.169.254/latest/meta-data, and are easily accessible from within the instance with a browser or most likely with wget or curl.
Amongst these data, the user is allowed to pass few kB of data to the instance. To do so one could use the euca2ools and in particular the euca-run-instances with the -f or -d option. The instance can then access this data at http://169.254.169.254/latest/user-data: cloud-init uses it to run scripts at boot time or otherwise customize the instance and the official Amazon Linux AMI uses a port of cloud-init. Cloud-init is not yet available to all distros and earlier version of Eucalyptus suffered a bug which prevented cloud-init to work properly (instance would get delayed at boot time), so we decided to use a much simpler rc.local script to allow for a subset of its functionality. You can find information on the rc.local script we use and of other information about images on projects.eucalyptus.com. With these starter images in our production cloud, we set to host planet in an instance.
Our first attempt to run a service in the cloud, emulated the boot from EBS capability (see our issue tracker in the cloud), but this time we changed completely tack. We pushed all the planet's configuration into a Walrus bucket, then we created a script to be used when starting an instance.
You can inspect the script on projects.eucalyptus.com under the Cloud Application Architect area. The script gets all the css, ini and others needed file from a walrus bucket, installs nginx and other needed packages, and set up a cron-job to re-read the configuration files at set intervals, thus allowing for the dynamic configuration of planet. When we changed the logo, it was a matter of uploading the new css and png files, and presto! Planet got a new skin.
While this setup seems complicated at first glance, it is fairly easy: consider that it's all done with a few line of a shell script. It also allows for a easy failure recovery since restarting planet is a matter of 2 euca2ools command. If the instance were to fail we can issue:
euca-run-instances -f planet.sh -k my-ssh-key emi-F3DF1488
euca-associate-address -i i-xxxxx 126.96.36.199
and since there is no persistent data on the instance, that's all we need to do. And to apply a security update, we simply spin up a new instance with the same script (the script we use upgrades to all latest packages at start up), disassociate the public IP from the old instance and associate it to the new instance. And of course we can terminate the old planet once the elastic IP has been moved.
To customize the planet we upload the new version of the specific file to the Walrus bucket, and to do so we use a version of s3curl modified to allow for different endpoint then S3. For example to add a new feed we first get the current planet.ini:
modify it, to add the new feed, and we upload it back into the bucket
s3curl --id graziano --acl public-read --put planet.ini -- http://188.8.131.52:8773/services/Walrus/planet/planet.ini
and wait for the cron-job to execute. Easy, isn't it?
We have few more scripts which we use for our production services on projects.eucalyptus.com and if you have a similar script you want to share, let us know and we'll add it, or do a git-hub merge request. The recipes are ready to go, just add Eucalyptus to it.