I’ve been spending time with Packer and cloud-init in the week, trying to create a re-built AMI to help enquicken the deployment process of a otherwise ansible process from fresh AMI procedure.

A few of the things I’ve noted about the stacks are here. I’ve gotten some of these merged as doc clarifications, but blogging about them is faster than any upstream contribution :)


Packer templates are not reloaded on retry

Found when I was trying to do a -debug -on-retry=ask iterative loop, and wondering why my edits weren’t helping.

Now in the documentation!


#cloud-init is your user-data on AWS

If you’re running aws ec2 run-instances (or the boto3 equivalent), your --user-data can be a file:// containing cloud-init pragma. Like a #!/bin/env python file can be interpreted as a Python file, no matter the extention, a file starting with #cloud-init will be run as cloud-init data by AWS on boot of your new instance.

Debugging cloud-init on AWS

/var/lib/cloud/ has lots of useful things, including:

  • /var/lib/cloud/instance/ has the input data and processed cloud-config.txt that was run for your instance
  • (technically it’s /var/lib/cloud/instances/i-INSTANCE_UID/, but symlinks, yo.

Don’t have more than one runcmd block

(I can’t find documentation to back this up.)

Comparing /var/lib/cloud/instance/user-data.txt and /var/lib/cloud/instance/cloud-config.txt I can see that my multiple runcmd blocks were ignored, and only my last declared block was run.

Which is great when debugging I was trying to work out why the write_content functionaity was working, but not (most) of the runcmd.

Update: turns out this is a limitation in yaml itself. You can’t declare multiple keys of the same name. This also explains why I had ‘valid’ yaml but was declaring write_files instead of write_content. Thanks Noah for the clarification!