provider/aws: Fix issue upgrading to Terraform v0.9+ with AWS OpsWorks Stacks

[catsby]

Fix an issue when upgrading from Terraform < 0.9 to 0.9+, when we added
support for the regional endpoints in OpsWorks Stacks. OpsWorks Stacks
can only be managed via the endpoint with which they were created, not
where the stack resides:

Stacks can only be accessed or managed within the endpoint in which they are created

• http://docs.aws.amazon.com/opsworks/latest/APIReference/Welcome.html

Prior to #12688, all OpsWorks stacks were created via us-east-1 endpoint, which was the only way to create them until very recent. Amazon has since added regional endpoints, and deprecated the approach of all-stacks-go-thru-us-east.

When we removed the default of us-east-1 in Terraform v0.9.0, any opsworks_stack resources created in a version of Terraform < 0.9 began querying the API with whatever region the Provider had been configured, not necessarily us-east-1, the region they were created with.

Any user in this situation would be presented with Terraform trying to create a new opsworks stack, because it was unable to find it due to the region diff. See #12842 .

Here, we add a stack_endpoint computed attribute, where we'll store the region in which the stack is found. On the next UPDATE or REFRESH action, Terraform will attempt to find the stack in the default Provider region. If not found, it will attempt to find it in us-east-1. If it's found in east, we'll save that as the stack_endpoint.

Future UPDATE, REFRESH, and DELETE operations will check this computed attribute and get the connection needed, if the stack_endpoint is found.

The acceptance test TestAccAWSOpsWorksStack_classic_endpoints should cover this. It first creates a stack in us-east-1, using us-east-1 region. In the second step, we switch to us-west-2 and preform a plan-only check of the configuration. Without the ability to check us-east-1 after checking the updated Provider region, this test step would fail to find the stack created and offer to create it, failing the step because it produces a non-empty plan.

[stack72]

LGTM! Thanks for the fix :)

% make testacc TEST=./builtin/providers/aws TESTARGS='-run=TestAccAWSOpsWorksStack_'                ✭
==> Checking that code complies with gofmt requirements...
go generate $(go list ./... | grep -v /terraform/vendor/)
2017/03/28 12:24:47 Generated command/internal_plugin_list.go
TF_ACC=1 go test ./builtin/providers/aws -v -run=TestAccAWSOpsWorksStack_ -timeout 120m
=== RUN   TestAccAWSOpsWorksStack_classic_endpoints
--- PASS: TestAccAWSOpsWorksStack_classic_endpoints (45.10s)
PASS
ok  	github.com/hashicorp/terraform/builtin/providers/aws	45.121s```

[GuiSim]

I'm facing this issue right now! Thanks for the fix.
How can I know which version of Terraform has this fix? I'm running on 0.9.2.

If this fix is in 0.9.2, I might have a closely related issue that affects module.phobos.aws_opsworks_custom_layer and aws_opsworks_instance. It appears that Terraform can't see layers and instances created with classic endpoint using a version of Terraform older than 0.9. It will attempt to recreate them and fail.

Let me know if you'd like me to fill a new issue or if this is supposed to be fixed by this @catsby !

Thanks!

[catsby]

This fix should apply to opsworks_stacks only. In my research that was the only resource that had this limitation, according to the docs I read:

Stacks can only be accessed or managed within the endpoint in which they are created

It's entirely possible things that belong to this stack also have that limitation? If you discover this is the case, please open a new GitHub issue.

Thanks!

[rafaelmagu]

For the record, this does not fix the problem with existing stacks. It continues to think my stack doesn't exist, but won't allow me to import saying that the stack already exists.

[GuiSim]

Judging by @rafaelmagu 's last comment, I think that #12842 is not resolved and should be reopened.

[nabeken]

@catsby I think Stacks is just used to clarify since they have another AWS OpsWorks for Chef Automate so Stacks does not mean a stack in OpsWorks Stacks.

Actually, All API requests should go to regional endpoint, which means all TF opsworks resources should contact to it.

[nabeken]

@catsby Filed an issue #13483

[ghost]

I'm going to lock this issue because it has been closed for 30 days ⏳. This helps our maintainers find and focus on the active issues.

If you have found a problem that seems similar to this, please open a new issue and complete the issue template so we can capture all the details necessary to investigate further.