Not defining "working_dir" triggers container replacement on every run about terraform-provider-docker

[mavogel]

• done @suzuki-shunsuke -> https://github.com/kreuzwerker/terraform-provider-docker/releases/tag/v2.9.0
• @dubo-dubon-duponey regarding #14 (comment): I added a simpler way: https://github.com/kreuzwerker/terraform-provider-docker/blob/master/CONTRIBUTING.md#test-against-current-terraform-iac-descriptions
• @quentinadam can you re-test :)

[suzuki-shunsuke]

#14 (comment)

Oh, sorry. I confirmed that this is fixed at the master.
I'll close #15 .

[suzuki-shunsuke]

Hmm..
I fixed the bug by hashicorp/terraform-provider-docker#272 ...

[suzuki-shunsuke]

Oh, Computed: true was removed??

terraform-provider-docker/docker/resource_docker_container.go

Lines 197 to 208 in 8476f9b

	"working_dir": {
	Type: schema.TypeString,
	Optional: true,
	ForceNew: true,
	DiffSuppressFunc: func(k, oldV, newV string, d *schema.ResourceData) bool {
	// treat "" as a no-op, which is Docker's default behavior
	if newV == "" {
	newV = oldV
	}
	return oldV == newV
	},
	},

[suzuki-shunsuke]

Computed: true was removed by this commit.
hashicorp/terraform-provider-docker@cdc0d91

[suzuki-shunsuke]

Will be fixed by #15

[dubo-dubon-duponey]

@suzuki-shunsuke that might be fixed already:

hashicorp/terraform-provider-docker#319

hashicorp/terraform-provider-docker#320

@quentinadam do you mind trying with master?

[quentinadam]

Happy to try that of course. I guess I should change something in the "versions.tf" file to use the latest master version instead of version 2.8.0 ? Would you mind pointing out what I should change in that file ? Here is the current version of the versions.tf file :

terraform {
  required_providers {
    digitalocean = {
      source = "digitalocean/digitalocean"
    }
    docker = {
      source = "kreuzwerker/docker"
    }
  }
  required_version = ">= 0.13"
}

[dubo-dubon-duponey]

@quentinadam

https://github.com/kreuzwerker/terraform-provider-docker/#developing-the-provider

These instructions suggest that you can just build it and use it by removing the version.

I don't remember if this ^ works or not though.

Here is what I've been doing (possibly because the above ^ is outdated?)

1. clone and build (make build) - you should now have a built plugin in your GOPATH/bin folder.

In my case:

dmp@macArena:~/Projects/Deploy/home$ which terraform-provider-docker
/Users/dmp/Applications/bin/gvm/pkgsets/go1.15/global/bin/terraform-provider-docker

2. Create ~/.terraformrc with the following:

provider_installation {
  filesystem_mirror {
    path    = "/Users/dmp/Projects/Deploy/home/built-plugins/"
    include = ["terraform-providers/docker"]
  }
}

^ adapt this path to whatever suits you.

3. Copy the plugin

dest=/Users/dmp/Projects/Deploy/home/built-plugins/registry.terraform.io/terraform-providers/docker/2.7.2/darwin_amd64
mkdir -p "$dest"

cp $(which terraform-provider-docker) "$dest"

Be sure to adapt the version and platform in the path to match your mileage.

Curious if there is a simpler / more straightforward way to achieve that.

[mavogel]

Then we'd trigger a release as soon as #12 is fixed

[dubo-dubon-duponey]

@mavogel @suzuki-shunsuke I've been thinking more about this patch.

It makes the situation better, for sure, but I now think it's wrong / incomplete.

Here is what should happen:

• if working_dir is explicitly set to value A:
  • if the existing container has value B (!= A) for working_dir, rebuild
  • if the existing container has value A, do not rebuild
• if working_dir is not set, or set to "":
  • if working_dir was previously NOT set (or set to to ""), do NOT rebuild (regardless of the value of working_dir in the running container)
  • if working_dir was previously set to value B (!= ""), and if the container "in dockerfile" working_dir was C (!= B), then rebuild

I have no idea how to do that... since this involve looking up both the past value of working_dir, and the value of working_dir if the container was started without explicitly overriding working_dir...

With what we have right now, this will be wrong in the following scenario:

• container was previously started with working_dir = A (where A is different from whatever the container had internally)
• we now ask for a reapply with working_dir = ""
• -> the container will not be rebuilt (while it should)

Again, I think what we have now is much better than before, but it will still be wrong in corner cases like that...

[suzuki-shunsuke]

#14 (comment)
@mavogel How about releasing the new version? #26 was merged.

[quentinadam]

Great ! I am confirming that it now works as expected (not defining "working_dir" does not trigger a container replacement on every run).

[mavogel]

Happy to hear that! Will close it.