Change the `RetryFunc` from a plain `error` return type to a
specialized `RetryError` which must decide whether it is
retryable or not.
Add `RetryableError` / `NonRetryableError` factory functions that
callers are meant to use to build up these errors.
This makes it eminently clear whether or not a given error is
retryable from inside the client code.
Goal here is to _not_ change any behavior, simply reflect the
existing behavior with the new, clearer, API.
All of these RetryErrors were meant to fail right away, but instead
caused retry looping because the typecheck in the implementation of
`resource.Retry()` only catches the value type, and not the pointer
type.
Refs #5537
Pull CIDR block validation into a shared func ready to be used elsewhere
Example of new err message:
```
Errors:
* aws_vpc.foo: "cidr_block" must contain a valid network CIDR,
expected "10.0.0.0/16", got "10.0.1.0/16"
```
This landed in aws-sdk-go yesterday, breaking the AWS provider in many places:
3c259c9586
Here, with much sedding, grepping, and manual massaging, we attempt to
catch Terraform up to the new `awserr.Error` interface world.
The AWS API call ModifyVpcAttribute will allow only one attribute to be
modified at a time. Modifying both results in the error:
Fields for multiple attribute types specified: enableDnsHostnames, enableDnsSupport
Retructure the provider to honor this restriction.
Also, enable DNS support before attempting to enable DNS hostnames,
since the former is a prerequisite of the latter.
Additionally, fix what must have been a copy&paste error, setting
enable_dns_support to the value of enable_dns_hostnames.
If the state file contained a VPC or a route table which no longer
exists, Terraform would fail to create the correct plan, which is to
recreate them.
In the case of VPCs, this was due to incorrect error handling. The AWS
SDK returns a aws.APIError, not a *aws.APIError on error. When the VPC
no longer exists, upon attempting to refresh state Terraform would
simply exit with an error.
For route tables, the provider would recognize that the route table no
longer existed, but would not make the appropriate call to update the
state as such. Thus there'd be no crash, but also no plan to re-create
the route table.
Mitchell Hashimoto