For a long time now, the diff logic has relied on the behavior of
`mapstructure.WeakDecode` to determine how various primitives are
converted into strings. The `schema.DiffString` function is used for
all primitive field types: TypeBool, TypeInt, TypeFloat, and TypeString.
The `mapstructure` library's string representation of booleans is "0"
and "1", which differs from `strconv.FormatBool`'s "false" and "true"
(which is used in writing out boolean fields to the state).
Because of this difference, diffs have long had the potential for
cosmetically odd but semantically neutral output like:
"true" => "1"
"false" => "0"
So long as `mapstructure.Decode` or `strconv.ParseBool` are used to
interpret these strings, there's no functional problem.
We had our first clear functional problem with #6005 and friends, where
users noticed diffs like the above showing up unexpectedly and causing
troubles when `ignore_changes` was in play.
This particular bug occurs down in Terraform core's EvalIgnoreChanges.
There, the diff is modified to account for ignored attributes, and
special logic attempts to handle properly the situation where the
ignored attribute was going to trigger a resource replacement. That
logic relies on the string representations of the Old and New fields in
the diff to be the same so that it filters properly.
So therefore, we now get a bug when a diff includes `Old: "0", New:
"false"` since the strings do not match, and `ignore_changes` is not
properly handled.
Here, we introduce `TypeBool`-specific normalizing into `finalizeDiff`.
I spiked out a full `diffBool` function, but figuring out which pieces
of `diffString` to duplicate there got hairy. This seemed like a simpler
and more direct solution.
Fixes#6005 (and potentially others!)
- Addresses the issue when local state file has logging_config populated and the user
disables the configuration via the UI (or in this case an
application of the TF config). This will now properly set the
logging_config during the read operation and identify the state as
diverging
Fixes#6390
* TF-6256 - SG Rule Retry
- Preferring slower but consistent runs when AWS API calls do not properly return the SG Rule in the list of ingress/egress rules.
- Testing has shown that several times that we had to exceed 20 attempts
before the SG was actually returned
* TF-6256 - Refactor of rule lookup
- Adjusting to use resource.Retry
- Extract lookup method for matching ipPermissions set
Here is an example that will setup the following:
+ An SSH key resource.
+ A virtual server resource that uses an existing SSH key.
+ A virtual server resource using an existing SSH key and a Terraform managed SSH key (created as "test_key_1" in the example below).
(create this as sl.tf and run terraform commands from this directory):
```hcl
provider "softlayer" {
username = ""
api_key = ""
}
resource "softlayer_ssh_key" "test_key_1" {
name = "test_key_1"
public_key = "${file(\"~/.ssh/id_rsa_test_key_1.pub\")}"
# Windows Example:
# public_key = "${file(\"C:\ssh\keys\path\id_rsa_test_key_1.pub\")}"
}
resource "softlayer_virtual_guest" "my_server_1" {
name = "my_server_1"
domain = "example.com"
ssh_keys = ["123456"]
image = "DEBIAN_7_64"
region = "ams01"
public_network_speed = 10
cpu = 1
ram = 1024
}
resource "softlayer_virtual_guest" "my_server_2" {
name = "my_server_2"
domain = "example.com"
ssh_keys = ["123456", "${softlayer_ssh_key.test_key_1.id}"]
image = "CENTOS_6_64"
region = "ams01"
public_network_speed = 10
cpu = 1
ram = 1024
}
```
You'll need to provide your SoftLayer username and API key,
so that Terraform can connect. If you don't want to put
credentials in your configuration file, you can leave them
out:
```
provider "softlayer" {}
```
...and instead set these environment variables:
- **SOFTLAYER_USERNAME**: Your SoftLayer username
- **SOFTLAYER_API_KEY**: Your API key
IPv6 support added.
We support 1 IPv6 address per interface. It seems like the vSphere SDK supports more than one, since it's provided as a list.
I can change it to support more than one address. I decided to stick with one for now since that's how the configuration parameters
had been set up by other developers.
The global gateway configuration option has been removed. Instead the user should specify a gateway on NIC level (ipv4_gateway and ipv6_gateway).
For now, the global gateway will be used as a fallback for every NICs ipv4_gateway.
The global gateway configuration option has been marked as deprecated.
* added update function with support for vcpu and memory
* waiting for vmware tools redundant with WaitForIP
* proper error handling of PowerOn task
* added test cases for update memory and vcpu
* reboot flag
this implements two new resource types:
* openstack_networking_secgroup_v2 - create a neutron security group
* openstack_networking_secgroup_rule_v2 - create a newutron security
group rule
Unlike their nova counterparts the neutron security groups allow a user
to specify the target tenant_id allowing a cloud admin to create per
tenant resources.
* Adding File Resource for vSphere provider
Allows for file upload to vSphere at specified location. This also
includes update for moving or renaming of file resources.
* Ensuring required parameters are provided
If any of the entries in `commands` on `docker_container` resources was
empty, the assertion to string panic'd. Since we can't use ValidateFunc
on list elements, we can only really check this at apply time. If any
value is nil (resolves to empty string during conversion), we fail with
an error prior to creating the container.
Fixes#6409.
We were passing in a disk path of `[datastore] `, which the createDisk
call would create a file named `.vmdk` on the datastore, which is not
the expected behavior. This make sure that if the user did not pass in
a vmdk path, that we call CreateDisk with an empty string like it
expects.
* provider/fastly: Add S3 Log Streaming to Fastly Service
Adds streaming logs to an S3 bucket to Fastly Service V1
* provider/fastly: Bump go-fastly version for domain support in S3 Logging
* provider/aws - CloudFront custom_error_response fixes for missing
- Omit custom_error_response response_* fields when not explicitly set via config for
SDK call
- Adding a test case to ensure that the response_error gets converted
to an empty string properly, versus "0". (Thanks @vancluever)
Fixes#6342
* - Fixing ACC test case resource names
When an SQS queue was deleted from the AWS Console, an error was thrown
to say that the Queue could not be found. This is now fixed to remove
the queue from the state on a specific not found exception
This commit should fix the following acceptance test failures:
=== RUN TestAccAzureDnsServerBasic
--- FAIL: TestAccAzureDnsServerBasic (2.17s)
testing.go:172: Step 0 error: Error applying: 1 error(s) occurred:
* azure_dns_server.foo: Failed issuing update to network
configuration: Error response from Azure. Code: BadRequest,
Message: Multiple DNS servers specified with the same name
'terraform-dns-server'.
=== RUN TestAccAzureDnsServerUpdate
--- FAIL: TestAccAzureDnsServerUpdate (2.04s)
testing.go:172: Step 0 error: Error applying: 1 error(s) occurred:
* azure_dns_server.foo: Failed issuing update to network
configuration: Error response from Azure. Code: BadRequest,
Message: Multiple DNS servers specified with the same name
'terraform-dns-server'.
This change adds the support for the proxied configuration option for a
record which enables origin protection for CloudFlare records.
In order to do so the golang library needed to be changed as the old did
not support the option and was using and outdated API version.
Open issues which ask for this (#5049, #3805).
User may specify a vmdk in their disk definition.
The options size, template, and vmdk are considered
to be mutually exclusive. User may also set whether each disk
associated with the vm should try to boot after creation.
Todo: Enforce mutual exclusivity, validate the bootable_vmdk_path
The "find route in table" helper code was not properly handling routes
with no destination CIDR block - like vpc_endpoint routes - so if one of
those routes would come up before the target route in the loop, we'd get
a crash.
Fixes#6337
* Updated `aws_elastic_beanstalk_environment` to update environment when the `template_name` attribute has a change. Consildated update functions to use a single update call and added state change conf to wait until environment is in a "Ready" state.
* Adding tests for `aws_elastic_beanstalk_configuration_template` use with the `aws_elastic_beanstalk_environment` resource.
* Verifies option settings from an `aws_elastic_beanstalk_configuration_template` resource are applied to the associated `aws_elastic_beanstalk_environment` resource
* Verifies updated name of an `aws_elastic_beanstalk_configuration_template` resource triggers an update for the associated `aws_elastic_beanstalk_environment` resource
* Verifies that option settings set in the `aws_elastic_beanstalk_environment` resource override settings in the `aws_elastic_beanstalk_configuration_template` resource
Currently, the number of nodes was broken due to not passing the
node_type with the update. This PR adds the correct parameters and a
test to prove this works as expected
The validation as part of #6330 was only for length. This PR adds the
rules for alphanumeric, not having -- within, not ending with a - and
that the id must start with a letter.
The PR also adds tests for these rules
The azure tests relating to cdn endpoints (TestAccAzureRMCdnEndpoint_basic
etc) are breaking because ARM is not accepting port values of 0. The
following error is received:
statusCode:BadRequest
statusMessage:{"error":{"code":"BadRequest","message":"Invalid port \"0\". Port value must be a number between 1 and 65535."}}
This patch sets the ports for example.com to 443 and 80.
When a directory service was not found, Terraform was panicking due to
`dir := out.DirectoryDescriptions[0]`. The AWS API doesn't throw an
Error in this case. IT just return s0 results. Therefore, we should
check for 0 results in the return and remove the directory from the
state
This commit uses Riviera to register the Microsoft.Compute provider as a
canary for whether or not the Azure account credentials are set up. It
used to use the MS client, but that appeared to panic internally if the
credentials were bad. It's possible that we were using it wrong, but
there are no docs so ¯\_(ツ)_/¯.
As part of this, we parellelise the registration of the other providers.
This shaves the latency of each provider request times the number of
providers minus 1 off the "startup" time of the AzureRM provider. The
result is quite noticeable.
Official OpenStack clients commonly support specifing a client
certificate/key to enable SSL client authentication when communicating
with OpenStack services. This patch enables such feature in Terraform
with new parameters and environment variables:
* 'cert' provider parameter or OS_CERT env variable to specify client
certificate file,
* 'key' provider parameter or OS_KEY env variable to specify client
certificate private key file.
* Add Triton Metadata modification AccTest.
The test starts the basic machine and then adds the metadata field
user_data.
Test fails if the user_data field does not match what we expect
OR it times out.
Related to hashicorp/terraform#6148
* Fix the non-convergence of Triton metadata changes
The code waiting for the entire Machine Metadata to "deep equal" the Terraform
metadata modifications. These two sets will only be the same if the user
changes all metadata fields of the resource before calling `apply`.
Closeshashicorp/terraform#6148
It can come in handy to be able to mount ISOs programmatically.
For instance if you're developing a custom appliance (that automatically installs itself on the hard drive volume)
that you want to automatically test on every successful build (given the ISO is uploaded to the vmware datastore).
There are probably lots of other reasons for using this functionality.
* provider/aws: Fix hashing on CloudFront certificate parameters
Adding necessary type assertion to values on the viewer_certificate hash
function to ensure that certain fields are indeed not zero string
values, versus simply zero interface{} values (aka nil, as is such for a
map[string]interface{}).
* provider/aws: CloudFront complex structure error handling
Handle errors better on calls to d.Set() in the
aws_cloudfront_distribution, namely in flattenDistributionConfig(). Also
caught a bug in the setting of the origin attribute, was incorrectly
attempting to set origins.
* provider/aws: Pass pointers to set CloudFront primitives
Change a few d.Set() for primitives in aws_cloudfront_distribution and
aws_cloudfront_origin_access_identity to use the pointer versus a
dereference.
* docs: Fix CloudFront examples formatting
Ran each example thru terraform fmt to fix indentation.
* provider/aws: Remove delete retention on CloudFront tests
To play better with Travis and not bloat the test account with disabled
distributions.
Disable-only functionality has been retained - one can enable it with
the TF_TEST_CLOUDFRONT_RETAIN environment variable.
* provider/aws: CloudFront delete waiter error handling
The call to resourceAwsCloudFrontDistributionWaitUntilDeployed() on
deletion of CloudFront distributions was not trapping error messages,
causing issues with waiter failure.
hil.Eval() now returns (hil.EvaluationResult, error) instead of (value,
type, error). This commit updates the call sites, but retains all
previous behaviour. Tests are also updated.
This commit patches a few acceptance tests in order to get them to
pass under OpenStack Mitaka.
The devstack dev environment script has also been updated to reflect
OpenStack Mitaka as well as the new Terraform dependency vendoring.
* provider/fastly: Add support for managing Headers
Adds support for managing Headers in a Fastly configuration.
* update acc test
* update website with example of adding a header block
* provider/aws: Default Network ACL resource
Provides a resource to manage the default AWS Network ACL. VPC Only.
* Remove subnet_id update, mark as computed value. Remove extra tag update
* refactor default rule number to be a constant
* refactor revokeRulesForType to be revokeAllNetworkACLEntries
Refactor method to delete all network ACL entries, regardless of type. The
previous implementation was under the assumption that we may only eliminate some
rule types and possibly not others, so the split was necessary.
We're now removing them all, so the logic isn't necessary
Several doc and test cleanups are here as well
* smite subnet_id, improve docs
According to the libpq documentation, "prefer" is the default in the
underlying library and so setting a different default in the Terraform
layer would be a breaking change for existing users of this provider
whose servers do not have TLS correctly configured.
The docs now link to the libpq manual's discussion of the security
implications of each of the ssl_mode options, so the user can understand
the limitations of the "prefer" default and can make an informed decision
about which setting is appropriate for their situation.
As with several other sensitive values in Opsworks, the API returns a
placeholder value rather than a nil. To avoid writing the placeholder
value into the state we just skip updating the password on read, letting
whatever value was in the state persist.
This means that Terraform can't detect configuration drift where someone
has changed the password via some other means, but Terraform will still
be able to recognize changes to the password made within Terraform itself
due to the "last-written" value in the state.
This fixes#6192.
Other separate changes to testAccOpsworksStackConfigNoVpcCreate caused
this to begin failing because it was attempting to create a stack with
an empty name.
Previously in Update we would only set req.CustomJson if a non-empty
value was provided in the config. It seems that the Opsworks API considers
a null CustomJson to mean "do not change" rather than "set to empty",
so we need to explicitly set the empty string in the request body in
order to successfully remove an already-configured custom JSON.
This introduces a provider for Cobbler. Cobbler manages bare-metal
deployments and, to some extent, virtual machines. This initial
commit supports the following resources: distros, profiles, systems,
kickstart files, and snippets.
* CloudFront implementation v3
* Update tests
* Refactor - new resource: aws_cloudfront_distribution
* Includes a complete re-write of the old aws_cloudfront_web_distribution
resource to bring it to feature parity with API and CloudFormation.
* Also includes the aws_cloudfront_origin_access_identity resource to generate
origin access identities for use with S3.
* Improve testing of CodeDeploy DeploymentGroup Trigger Configs
- ensure updates to trigger_events are applied
- assert changes to trigger_target_arn
* Retry CodeDeploy DeploymentGroup when Trigger Config SNS Topic is not available
- increase retries from 2 => 5
* Don't Base64-encode EC2 userdata if it is already Base64 encoded
The user data may be Base64 encoded already - for example, if it has been
generated by a template_cloudinit_config resource.
* Add encoded user_data to aws_instance acceptance test
* Issue #2174 Check that InternetGateway exists before returning from creation
Fix some random InvalidInternetGatewayID.NotFound errors
* Issue #2174 Reuse IGStateRefreshFunc
* Issue #2174 Need to wait for creation before setting tags
Update github.com/joyent/gosdc/...
Test does the minimum described in hashicorp/terraform#6109, i.e.
- Start a small instance, t4-standard-128M
- Check firewall is enabled
- Change configuration to disable firewall
- Check firewall is disabled.
Fixes#6119.
* provider/aws: CodeDeploy Deployment Group Triggers
- Create a Trigger to Send Notifications for AWS CodeDeploy Events
- Update aws_codedeploy_deployment_group docs
* Refactor validateTriggerEvent function and test
- also rename TestAccAWSCodeDeployDeploymentGroup_triggerConfiguration test
* Enhance existing Deployment Group integration tests
- by using built in resource attribute helpers
- these can get quite verbose and repetitive, so passing the resource to a function might be better
- can't use these (yet) to assert trigger configuration state
* Unit tests for conversions between aws TriggerConfig and terraform resource schema
- buildTriggerConfigs
- triggerConfigsToMap
* provider/aws: Add more Randomization to DB Parameter Group Tests, to avoid collisions
* provider/aws: Add more randomization to Autoscaling group tests
We have a curtesy function in place allowing you to specify both a
`name` of `ID`. But in order for the graph to be build correctly when
you recreate or taint stuff that other resources depend on, we need to
reference the `ID` and *not* the `name`.
So in order to enforce this and by that help people to not make this
mistake unknowingly, I deprecated all the parameters this allies to and
changed the logic, docs and tests accordingly.
Added the ability to set the "privacy" of a github_team resource so all teams won't automatically set to private.
* Added the privacy argument to github_team
* Refactored parameter validation to be general for any argument
* Updated testing
This is the first step in removing the config dependency on "project".
This change is backwards-compatible because the value for this new
attribute defaults to the value from the provider.
This commit enables the ability to authenticate to OpenStack by way
of a Keystone Token. Tokens can provide a way to use Terraform and
OpenStack with an expiring, temporary credential. The token will need
to be generated out of band from Terraform.
* provider/aws: test empty plan with sns_topic policy with random order
If we setup a sns_topic policy with a policy with a different order
to the one set by the AWS API, terraform plan will be not empty between
runs.
* provider/aws: normalize json policy for sns topic
For the policy attribute of the resource aws_sns_topic, AWS returns the policy
in JSON format with the fields in a different order.
If we store and compare the values without normalizing, terraform
will unnecesary trigger and update of the resource.
To avoid that, we must add a normalization function in the StateFunc of
the policy attribute and also when we read the attribute from AWS.
This commit adds a no_gateway attribute. When set, the subnet will
not have a gateway. This is different than not specifying a
gateway_ip since that will cause a default gateway of .1 to be used.
This behavior mirrors the OpenStack Neutron command-line tool.
Fixes#6031
When calling AssociateAddress, the PrivateIpAddress parameter must be
used to select which private IP the EIP should associate with, otherwise
the EIP always associates with the _first_ private IP.
Without this parameter, multiple EIPs couldn't be assigned to a single
ENI. Includes covering test and docs update.
Fixes#2997
Previously the format string was using %#v, which prints the whole data structure given.
Instead we want to use %s to get the string representation of the error.
This fixes#6038.
Normalise the event_pattern of the aws_cloudwatch_event_rule resource
before uploading it to AWS.
AWS seems to accept a event_pattern with a JSON with new lines, but then
the rule does not seem to work. Creating the rule in the AWS console works,
but will setup the pattern as a json without newlines or spaces, and
display a formatted JSON.
Previously, resizing would only work if the flavor_id changed and
would create an error if the flavor_name changes. This commit fixes
this behavior.
It also quickly refactors the getFlavorID function to use
Gophercloud's IDFromName function. getFlavorID was the basis of
IDFromName so the exact same code is used.
Fixes#5780
Fix retry after removing associations by correctly checking and returning an
error. This should patch the VPC/Resource leak in our nightly acceptance tests.
It turns out all other providers use `ip_address` where the CloudStack
provider uses `ipaddress`. To make this more consistent this PR
deprecates `ipaddress` and adds `ip_address` where needed…
This new resource is an alternative to consul_keys that manages all keys
under a given prefix, rather than arbitrary single keys across the entire
store.
The key advantage of this resource over consul_keys is that it is able to
detect and delete keys that are added outside of Terraform, whereas
consul_keys is only able to detect changes to keys it is explicitly
managing.
The provider should, when working on a new repository without branches:
* Able to create a new repository even with default_branch defined.
* Able to create a new repository without default_branch, and do not fail
if default_branch is defined.
In AWS codecommit the default branch must have a value unless there are
no branches created, in which case it is not possible to set it to any value.
We query the existing branches and do not update the default branch
if there are none defined remotely.
This solves the issue of the initial creation of the repository with a
resource with `default_branch` defined.
AWS changed their error message, which was being used for detection of
the specific error that indicates we need to wait for IAM propagation.
Behavior is covered by a test now.
Fixes#5862
Unlike SimpleScaling policies, StepScaling policies require one or more
"steps", which are interval ranges in which a tracked metric can lie.
Policies can then execute scaling adjustments wedded to these steps.
This commit also adds a slew of additional policy attributes which are
only applicable to step policies.
Clint