v4.0.0

Bug fixes and enhancements combined into a single breaking release @aknysh (#202)

Breaking Changes

Terraform version 1.3.0 or later is now required.

policy input removed

The deprecated policy input has been removed. Use source_policy_documents instead.

Convert from

policy = data.aws_iam_policy_document.log_delivery.json

to

source_policy_documents = [data.aws_iam_policy_document.log_delivery.json]

Do not use list modifiers like sort, compact, or distinct on the list, or it will trigger an Error: Invalid count argument. The length of the list must be known at plan time.

Logging configuration converted to list

To fix #182, the logging input has been converted to a list. If you have a logging configuration, simply surround it with brackets.

Replication rules brought into alignment with Terraform resource

Previously, the s3_replication_rules input had some deviations from the aws_s3_bucket_replication_configuration Terraform resource. Via the use of optional attributes, the input now closely matches the resource while providing backward compatibility, with a few exceptions.

• Replication source_selection_criteria.sse_kms_encrypted_objects was documented as an object with one member, enabled, of type bool. However, it only worked when set to the string "Enabled". It has been replaced with the resource's choice of status of type String.
• Previously, Replication Time Control could not be set directly. It was implicitly enabled by enabling Replication Metrics. We preserve that behavior even though we now add a configuration block for replication_time. To enable Metrics without Replication Time Control, you must set replication_time.status = "Disabled".

These are not changes, just continued deviations from the resources:

• existing_object_replication cannot be set.
• token to allow replication to be enabled on an Object Lock-enabled bucket cannot be set.

what

• Remove local local.source_policy_documents and deprecated variable policy (because of that, pump the module to a major version)
• Convert lifecycle_configuration_rules and s3_replication_rules from loosely typed objects to fully typed objects with optional attributes.
• Use local bucket_id variable
• Remove comments suppressing Bridgecrew rules
• Update tests to Golang 1.20

why

• The number of policy documents needs to be known at plan time. Default value of policy was empty, meaning it had to be removed based on content, which would not be known at plan time if the policy input was being generated.
• Closes #167, supersedes and closes #163, and generally makes these inputs easier to deal with, since they now have type checking and partial defaults, meaning the inputs can be much smaller.
• Incorporates and closes #197. Thank you @nikpivkin
• Suppressing Bridgecrew rules Cloud Posse does not like should be done via external configuration so that users of this module can have the option of having those rules enforced.
• Security and bug fixes

explanation

Any list manipulation functions should not be used in count since it can lead to the error:

│ Error: Invalid count argument
│
│   on ./modules/s3_bucket/main.tf line 462, in resource "aws_s3_bucket_policy" "default":
│  462:   count      = local.enabled && (var.allow_ssl_requests_only || var.allow_encrypted_uploads_only || length(var.s3_replication_source_roles) > 0 || length(var.privileged_principal_arns) > 0 || length(local.source_policy_documents) > 0) ? 1 : 0
│
│ The "count" value depends on resource attributes that cannot be determined until apply, so Terraform cannot predict how many instances will be created. To work around this, use the -target argument to
│ first apply only the resources that the count depends on.

Using the local like this

source_policy_documents = var.policy != "" && var.policy != null ? concat([var.policy], var.source_policy_documents) : var.source_policy_documents

would not work either if var.policy depends on apply-time resources from other TF modules.

General rules:

• When using for_each, the map keys have to be known at plan time (the map values are not required to be know at plan time)

• When using count, the length of the list must be know at plan time, the items inside the list are not. That does not mean that the list must be static with the length known in advance, the list can be dynamic and come from a remote state or data sources which Terraform evaluates first during plan, it just can’t come from other resources (which are only known after apply)

• When using count, no list manipulating functions can be used in count - it will lead to the The "count" value depends on resource attributes that cannot be determined until apply error in some cases