Post-processors in Packer operate on an artifact from the build phase of a packer run. The artifact may be a local virtual machine image, an identifier from a remote API, or a loose collection of files from artifice.
Today the shell-local post-processor operates on files, not artifacts. The result of this design is that for builds that produce remote artifacts, like AWS, the shell-local post-processor doesn't work. Also, if the artifact has multiple files this results in multiple invocations of the script.
Overall this behavior is very confusing and most people expect the shell-local post-processor to operate more like a lifecycle hook that runs at the end of the build.
There is more context in #3669 but I have had this conversation with 5 different people this week alone, so clearly this is broken and we need to fix it!
So what you want to change?
Just stumbled over this while surfing and now I finally realize why it seems like shell-local
provisioners are simply not run (no local artifact). I doubt I'm alone when I'm thinking that the shell-local
provisioner should be run once per build, regardless of what artifacts have been produced.
@cbednarski do you know if any work has been started on this? I'd be happy to get a PR going if only for discussion 😁
I doubt I'm alone when I'm thinking that the shell-local provisioner should be run once per build, regardless of what artifacts have been produced.
I'm having to create fake artifacts for using with shell-local. Here's how it looks:
{
"provisioners": [
<skipped main provisioners>
{
"type": "shell",
"inline": ["date > /tmp/dummy.txt"]
},
{
"type": "file",
"source": "/tmp/dummy.txt",
"destination": "dummy.txt",
"direction": "download"
}
],
"post-processors": [
<skipped main post-processors>
[
{
"type": "artifice",
"files": [
"dummy.txt"
]
},
{
"type": "shell-local",
"script": "script/last_post_processor_running_only_once.sh"
}
]
]
}
This workaround was suggested by Hashicorp support and works well so far.
Would definitely love to see it working without fake artifacts.
@iroller samesies, which is how I landed here 😺 https://github.com/travis-ci/packer-templates/blob/6969956e44f9d86724cdfe3b455b0cd74941f665/ci-connie.yml#L103-L122
The fake artifacts are one part. Keeping secret env vars secret is another. Overall, I'd be happy if post-processor/shell-local
behaved _mostly_ like provisioner/shell-local
, including having same/similar config fields.
I want to take this, but need more info about design, what i need to pass to the script? @Cbeck527 @rickard-von-essen ?
@cbednarski agreed that it would be more universally useful to execute once rather than once per file (or artifact). It's also unfortunate that it's focused specifically on files. If we're talking about useful changes to the behavior, how about supplying a JSON array of all structured artifact data to the script via stdin?
@jeremy-asher I think that creates a significant limitation since most applications you might want to run afterward will not expect that. However, you can accomplish something very similar using the manifest post-processor (see master for a reference).
Sorry about the radio silence in general. I will try to update with a spec soon so contributors can take a stab at this.
Yeah, bit of an extreme idea. The manifest post-process definitely does the trick! Nice!
jotting some notes
@mwhooker @cbednarski what args we need to pass to shell script in shell-local post processor ?
In previous version we pass artifact file, but now as i understand we no need to pass anything? Or i'm wrong?
right, no arguments.
2016-11-21 21:33 GMT+03:00 Matthew Hooker [email protected]:
right, no arguments.
So if you have no objections, please check my pr about shell-local
post-processor
Vasiliy Tolstov,
e-mail: v.[email protected]
closed via #4189
Awesome!
On Dec 8, 2016 21:41, "Matthew Hooker" notifications@github.com wrote:
Closed #3671 https://github.com/mitchellh/packer/issues/3671.
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
https://github.com/mitchellh/packer/issues/3671#event-887461962, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AAiCgyfyKODIS1WlCTGN1lRuyDh31qKcks5rGGt1gaJpZM4JDYqX
.
Most helpful comment
Just stumbled over this while surfing and now I finally realize why it seems like
shell-local
provisioners are simply not run (no local artifact). I doubt I'm alone when I'm thinking that theshell-local
provisioner should be run once per build, regardless of what artifacts have been produced.