HashiConf 2024 Now streaming live from Boston! Attend for free
Packer
Packer Home

Debugging Packer Builds

Using packer build -on-error=ask allows you to inspect failures and try out solutions before restarting the build.

For remote builds with cloud providers like Amazon Web Services AMIs, debugging a Packer build can be eased greatly with packer build -debug. This disables parallelization and enables debug mode.

Debug mode informs the builders that they should output debugging information. The exact behavior of debug mode is left to the builder. In general, builders usually will stop between each step, waiting for keyboard input before continuing. This will allow you to inspect state and so on.

In debug mode once the remote instance is instantiated, Packer will emit to the current directory an ephemeral private SSH key as a .pem file. Using that you can ssh -i <key.pem> into the remote build instance and see what is going on for debugging. The key will only be emitted for cloud-based builders. The ephemeral key will be deleted at the end of the Packer run during cleanup.

For a local builder, the SSH session initiated will be visible in the detail provided when PACKER_LOG=1 environment variable is set prior to a build, and you can connect to the local machine using the userid and password defined in the kickstart or preseed associated with initializing the local VM.

It should be noted that one of the options -on-error is to retry, the retry of the step in question has limitations:

  • the template Packer is building is not reloaded from file.
  • the resources specified from builders are reloaded from file.

Check the specifics on your builder to confirm their behavior.

Windows

As of Packer 0.8.1 the default WinRM communicator will emit the password for a Remote Desktop Connection into your instance. This happens following the several minute pause as the instance is booted. Note a .pem key is still created for securely transmitting the password. Packer automatically decrypts the password for you in debug mode.

Debugging Packer

Issues occasionally arise where certain things may not work entirely correctly, or may not appear to work correctly. In these cases, it is sometimes helpful to see more details about what Packer is actually doing.

Packer has detailed logs which can be enabled by setting the PACKER_LOG environmental variable to any value but "" (empty string) and "0" like this PACKER_LOG=1 packer build <config.json>. This will cause detailed logs to appear on stderr. The logs contain log messages from Packer as well as any plugins that are being used. Log messages from plugins are prefixed by their application name.

Note that because Packer is highly parallelized, log messages sometimes appear out of order, especially with respect to plugins. In this case, it is important to pay attention to the timestamp of the log messages to determine order.

In addition to simply enabling the log, you can set PACKER_LOG_PATH in order to force the log to always go to a specific file when logging is enabled. Note that even when PACKER_LOG_PATH is set, PACKER_LOG must be set in order for any logging to be enabled.

Debugging Plugins

Each packer plugin runs in a separate process and communicates with RPC over a socket therefore using a debugger will not work (be complicated at least).

But most of the Packer code is really simple and easy to follow with PACKER_LOG turned on. If that doesn't work adding some extra debug print outs when you have homed in on the problem is usually enough.

Debugging Packer in Powershell/Windows

In Windows you can set the detailed logs environmental variable PACKER_LOG or the log variable PACKER_LOG_PATH using PowerShell environment variables. For example:

$env:PACKER_LOG=1
$env:PACKER_LOG_PATH="packerlog.txt"

If you find a bug with Packer, please include the detailed log by using a service such as gist.

Issues Installing Ubuntu Packages

Issues may arise using and building Ubuntu AMIs where common packages that should be installed from Ubuntu's Main repository are not found during a provisioner step:

amazon-ebs: No candidate version found for build-essential
amazon-ebs: No candidate version found for build-essential

This, obviously can cause problems where a build is unable to finish successfully as the proper packages cannot be provisioned correctly. The problem arises when cloud-init has not finished fully running on the source AMI by the time that packer starts any provisioning steps.

Adding the following provisioner to the Packer template, allows for the cloud-init process to fully finish before packer starts provisioning the source AMI.

{
  "type": "shell",
  "inline": [
    "cloud-init status --wait"
  ]
}

Issues when using numerous Builders/Provisioners/Post-Processors

Packer uses a separate process for each builder, provisioner, post-processor, and plugin. In certain cases, if you have too many of these, you can run out of file descriptors. This results in an error that might look like

error initializing provisioner 'powershell': fork/exec /files/go/bin/packer:
too many open files

On Unix systems, you can check what your file descriptor limit is with ulimit -Sn. You should check with your OS vendor on how to raise this limit.

Issues when using long temp directory

Packer uses Unix sockets internally, which are created inside the default directory for temporary files. Some operating systems place a limit on the length of the socket name, usually between 80 and 110 characters. If you get an error like this (for any builder, not just Docker):

Failed to initialize build 'docker': error initializing builder 'docker': plugin exited before we could connect

you should try setting your temp directory to something shorter. This can be done through the TMPDIR environment variable.

Edit this page on GitHub