Changelog for Unfurl

v1.1.0 - 2024-4-02

Compare with 1.0.0

Major changes

Ansible configurator

  • A playbook’s host is now set automatically if not explicitly specified. See for the selection rules.

  • If playbook and inventory input parameters have a string value, detect whether to treat as a file path or parse to YAML.

  • Fix rendering of inventory.yml when the inventory input parameter is set to inline YAML.

  • If the ansible_connection host var is not explicitly set, default to “ssh” if we’re also setting ansible_host to the node’s ip_address.

Update tosca package to v0.0.8

Release includes the following fixes and enhancements:

  • yaml to python: more idiomatic Python when importing __init__.yaml

  • yaml to python: use import namespace when following imports

  • yaml to python: don’t forward reference built-in types

  • overwrite policy: don’t overwrite if converted contents didn’t change

  • remove dependency on unfurl package.

  • support array and key access in field projections

  • allow regular data as arguments to boolean expressions.

  • add fallback(left: Optional[T], right: T) -> T to unfurl.tosca_plugins.expr for type-safe default expressions.

  • move tfoutput and tfvar Options to unfurl.tosca_plugins.expr (this make them available in safe mode).

Server enhancements and fixes

  • Add a /empty_cache POST endpoint for clearing entire cache (using optional prefix parameter). Access requires UNFURL_SERVER_ADMIN_PROJECT environment variable to be set and the auth_project URL parameter to match it.

  • Patch and export now support the “branch” field on DeploymentTemplate

  • Invalidate cached blueprint export even if file in the key didn’t change.

Other Notable bug fixes

897c32ba testing: add mypy testing api and make lifecycle api more flexible

3f80e5bf job: fix for TaskView.find_connection()

269f981c runtime: refine “is resource computed” heuristic

253f55aa cloudmap: fix tag refs on gitlab hosts

61da0c6f clone: apply revision if found in fragment of the cloned url

63e2f48a kompose: have KomposeInputs use unfurl_datatypes_DockerContainer

03feb7d7 plan: exclude replaced node templates from relationships

0922b557 logging: smarter truncation of log messages with stack traces

aaabb8a2 packages: fix resolving package compatibility

667c59d9 export: stop clearing out requirement match pointing to nested templates

5bfef1d8 export: fix get_nodes_of_type in match filters

a3c8fb54 export: stop hoisting default templates as requirements.

df2d3e97 parser: fix matching when a requirement target is substituted by the outer topology template.

a3f7feae parser: fix typename checking when evaluating ‘node’ field on requirements

b570246f parser: use the type’s namespace when creating properties.

1b965e5b parser: fix Namespace.get_global_name_and_prefix()

Breaking changes

Drop support for Python 3.7

v1.0.0 - 2024-2-26

Compare with 0.9.1


We’ve strived to maintain backwards compatibility and API stability for a while now, so for this release we decided to go ahead and christen it 1.0 🎉.

Major new features include:

TOSCA namespaces and global type identifiers

This release adds features designed to enable 3rd party type libraries to be shared within the same TOSCA topology and for Unfurl API consumers (such as Unfurl Cloud) to manage them.

  • Namespace isolation.

Each imported service template is placed in a separate namespace that is used to resolve type references in the file. It includes the types defined in that file along with the types it imports, with those type names prefixed if namespace_prefix key is set on the import definition. The namespace will be unique to that import unless an explicit namespace is declared (see below). This can be disabled or limited using the new UNFURL_GLOBAL_NAMESPACE_PACKAGES environment variable (see Breaking changes below). Note that the Python DSL already behaves this way, as documented here.

  • Support for TOSCA 1.3’s namespace field

If a service template explicitly declares a namespace using the namespace keyword, its namespace will be assigned that name and namespace isolation will be disabled for any templates it imports – so any import will share the same namespace unless it also declares its own namespace. In addition, any other template that declares the same namespace identifier will be placed in the same namespace. Shared namespaces means a template can reference types it didn’t explicitly import and overwrite existing type definitions with the same name and so declaring namespaces is not recommended.

  • Globally unique identifiers for types.

Namespaces can be used to generate globally unique type names and updated the Unfurl Server APIs and GraphQl/JSON export format to use these globally unique names. Follow the format <typename>@<namespace_id> where namespace_id is namespace the type was declared in. If a namespace id isn’t explicitly declared using the namespace keyword, one will be generated using the package id of its repository or current project, and optionally, a file path if it isn’t the root service template. For example:>/onecommons/std (a type defined in service-template.yaml) and>/onecommons/std:aws (a type defined in aws.yaml). TOSCA and unfurl types defined in the core vocabulary (and don’t need to be imported) are not qualified. Built-in unfurl types that do need to be imported use unfurl as there package id, for example: unfurl.nodes.Installer.Terraform@unfurl:tosca_plugins/artifacts. Generation of type global names with export and the APIs can be disabled by setting the UNFURL_EXPORT_LOCALNAMES environment variable (See Breaking changes below).

  • Cross-package interoperability with type metadata.

A package can declare compatibility with types in different packages without having to import those packages using the aliases and deprecates keywords in the metadata section of a type definition. The keywords value can be either a fully qualified type name or a list of fully qualified type names and indicate that the type is equivalent to the listed types. This is used both by the parser and by the API (export includes those types in exported types extends section) used by Unfurl Cloud’s UI.

Unfurl Server and export APIs

  • Unfurl Server (and the Unfurl Cloud front-end) patching API now uses global type names to generate import statements with prefixes as needed to prevent clashes packages with the same name.

  • Support for HTTP caching: Improved etag generation and Cache-Control headers that enable the browser and proxy caches to use stale content while processing slow requests. Use the CACHE_CONTROL_SERVE_STALE environment variable to set.

  • Add a Dockerfile that includes a nginx caching proxy in front of unfurl server. Provide prebuilt container images as onecommons/unfurl:v1.0.0-server-cached on and

  • Improvements to Redis caching: Track dependencies better, cache commit dates for more efficient shallow clones, improved error handling and recovery.

  • Local developer mode will now serve content from any local repository tracked by the current project or the home project (in local/unfurl.yaml). It also improves handling local changes and error recovery.

  • Improve type annotations for new Graphql types and consolidated Graphql Schema.

  • Add support for a property_metadata metadata key to apply metadata to individual properties on a TOSCA datatype.

For example, this property declaration applies the user_settable metadata key to the environment property on unfurl.datatypes.DockerContainer:

        type: unfurl.datatypes.DockerContainer
              user_settable: true

Python DSL

Service templates written in Python now have the following integration points with Unfurl’s runtime:

  • ToscaType implementations of TOSCA operations can return a Python method instead of an artifact or configurator and that method will converted to a configurator.

  • If TOSCA operation attempts to execute runtime-only functionality it will be invoked during the job’s render phase instead of converted to YAML.

  • Introduce a Computed() field specifier for TOSCA properties whose value is computed at runtime with the given method.

  • Add a unfurl.tosca_plugins.functions module containing utility functions that can be executed in the safe mode Python sandbox. This allows these functions to be executed in “spec” mode (e.g. as part of a class definition or in _class_init).

  • Add a unfurl.expr module containing type-safe python equivalents to Unfurl’s eval expression functions.

  • Add methods and free functions providing type-safes equivalents to Unfurl’s query expressions: find_configured_by, find_hosted_on, find_required_by and find_all_required_by APIs.

  • Providing these apis required synchronizing a job’s task context as a per-thread global runtime state, add public apis for querying global state and retrieving the current RefContext.

  • Treat non-required instance properties as optional (return None instead of raising KeyError)

  • Add a public unfurl.testing.create_runner() api for writing unit tests for Python DSL service templates.

Various improvements to YAML-to-Python and Python-to-YAML conversion, including:

  • Better support for artifacts

  • Better default operation conversion

  • Support for aliased references (multiple variables assigned to the same type).

  • Special case module attributes named __root__ to generate TOSCA substitution_mappings (aka root). For example:

    __root__ = my_template


    __root__ = MyNodeType

DSL API improvements:

  • Add a NodeTemplateDirective string enum for type-safe node template directives.

  • Add a @anymethod method decorator for creating methods that can act as both a classmethods and regular method.

  • Revamp Options api for typed and validated metadata on field specifiers.

  • Add a DEFAULT sentinel value to indicate that a field should have a default value constructed from the type annotation. This helps when using forward references to types that aren’t defined yet as well as prevent a bit of DRY.

  • Add a similar CONSTRAINED sentinel value to indicate that the property’s value will be set in _class_init.

  • Add a API for registering custom TOSCA property constraint types.

  • Add UNFURL_TEST_SAFE_LOADER environment variable to force a runtime exception if the tosca loader isn’t in safe mode or to disable safe mode (for testing).UNFURL_TEST_SAFE_LOADER=never option to disable, any other non-empty value to enforce safe mode.

  • Improve the API documentation.

  • Release tosca package 0.0.7


  • Package version resolution during loading of an ensemble now honors the lock section of the ensemble if present. After an ensemble is deployed, the version tag recorded in the lock section for a package will take precedence.

  • The lock section will record if a repository was missing version tags, in that case, subsequent deployments will attempt to fetch the first version tag. (As opposed to the default behavior of fetching the latest version tag if one wasn’t explicitly specified.)

  • Package rules are now serialized in the lock section, following the format used for UNFURL_PACKAGE_RULES environment variables. These rules will be applied in future unfurl invocation if the UNFURL_PACKAGE_RULES environment variable is missing.

  • Package rules are applied more widely, for example when matching repository URLs and to namespace ids.

  • A “packages” key has be added to Deployment graphql object containing the packages and versions used by the deployment.

Artifact enhancements

Support for the following built-in keywords on artifact definitions have been added as TOSCA extensions:

  • contents: Artifacts can now specify their contents inline using the new field. Contents can be auto-generated using eval expression or jinja templates.

  • target: The node template to deploy the artifact to (default is the node template the artifact is defined in).

  • permissions: A string field indicating the file permissions to apply when an artifact is deployed as a file.

  • order: A number field that provides a weight for establishing the order in which declared artifacts are processed.

  • intent: A string field for declaring the (user-defined) intent for the artifact.

  • Built-in artifact fields will now be evaluated like user-defined properties at runtime (e.g. they can contain eval expressions).

  • Artifact can now be dynamically created when declared in resultTemplates.

Other Features

  • Support Python 3.12

  • Add an unfurl.testing module for unit tests.

  • Update the dashboard skeleton to use the new onecommons/std repository.

  • The runtime will add a default relationship of type of unfurl.relationships.ConnectsTo.ComputeMachines to the current environment when a primary provider isn’t defined in that environment.

  • Add support for a validation key in type metadata to provide user-defined eval expressions for validating properties and custom datatypes (equivalent to the validation key in property metadata).

  • Add required and notnull options to the “strict” field on “eval” expressions. This provides simple validation of eval expressions, e.g.:

    eval: .::foo
    strict: required  # raise error if no results are found


    eval: .::foo
    strict: notnull # raise error if result isn't found or is null

Breaking changes

  • Prior to this release all imports shared the same namespace for types. For example, one imported file could reference a type declared in another file if both those files happened to be both imported by another service template.

With this release this behavior is only enabled if the importing template “namespace” key and the imported templates don’t declare a different namespace. Otherwise, each TOSCA import now has its own namespace. Types only have visibility and templates.

If it is a package you can’t modify, you can use UNFURL_GLOBAL_NAMESPACE_PACKAGES expects space separated list of namespaces names with wildcards. So UNFURL_GLOBAL_NAMESPACE_PACKAGES='*' would enable that flag for all service templates.

If missing, defaults to “*”

  • The unfurl server API and the export cli command now exports type names as fully qualified. This can be disabled by setting the UNFURL_EXPORT_LOCALNAMES environment variable. This will environment variable will be removed in a future release.

  • To enable Python 3.12, Unfurl now depends on ansible-core version 2.15.9.

Minor Enhancements and Notable Bug Fixes

  • job If a node template is created for a “discovered” instance that has a parent, set that parent as the “host” requirement for the template.

  • job Improve the job summary table and other logging tweaks.

  • job Report blocked tasks separately from failed jobs (and mark more tasks as blocked).

  • plan Improve heuristic for detecting circular dependencies and deadlocks (fixes #281)

  • job don’t silently skip operations when instantiating a configurator class fails.

  • runtime Fix to assure the reevaluation of computed results when the computation’s dependencies change.

  • cli Add verbose output to the status command if -v flag is used.

  • cloudmap add –repository cli option

  • cloudmap add a save_internal config setting for Gitlab hosts.

  • job Add heuristic for setting the status on relationship instances when the task’s configurator doesn’t explicitly set it.

  • job add unfurl install path to the $PATH passed to tasks.

  • job By default, dry run jobs print the modified ensemble.yaml to the console instead of saving to disk. Use UNFURL_SKIP_SAVE=never to force them to be saved.

  • parser Remove attribute definition if a derived type re-declares it as a property.

  • parser: Move default node template (those with a “default” directive) completely from inner topology to the root topology.

  • loader, export: don’t add environment instances or imports to ensemble except when generating json for the environments. By default, don’t instantiate environment instances unless they are imported from external ensembles. Similarly, don’t include imports declared in an environment unless they have a namespace_prefix that referenced by a connection in the environment.

  • logging add a UNFURL_LOG_TRUNCATE environment variable to change the maximum log message length (default: 748)

  • parser Accept relationship types defined in TOSCA extensions in requirement definitions.

  • eval have template function’s path parameter make relative paths relative to source location

  • dns install octodns artifact if needed

  • ansible load ansible collection artifact as soon as they are installed.

  • kompose make sure service names should conform to dns host name syntax

  • kompose fix expose input parameter

  • kompose implement the env input parameter

  • kompose update artifact version

  • package wide many improvements to type annotations; for example, introducing TypedDicts and overloaded signatures for the to_label family of functions.

v0.9.1 - 2023-10-25

Compare with 0.9.0


TOSCA and Python DSL

  • Add ToscaInputs classes for Unfurl’s built-in configurators to enable static type checking of configurator inputs.

  • Introduce the Options class to enable typed metadata for TOSCA fields and have the Terraform configurator add tfvar and tfoutput options to automatically map properties and attributes to Terraform variables and outputs, respectively.

This example uses the above features to integrate a Terraform module.

from unfurl.configurators.terraform import TerraformConfigurator, TerraformInputs, tfvar, tfoutput

class GenericTerraformManagedResource(tosca.nodes.Root):
    example_terraform_var: str = Property(options=tfvar)
    example_terraform_output: str = Attribute(options=tfoutput)

    @operation(apply_to=["Install.check", "Standard.configure", "Standard.delete"])
    def default(self, **kw):
        return TerraformConfigurator(TerraformInputs(main="terraform_dir"))
  • Introduce unfurl.datatypes.EnvironmentVariables, a TOSCA datatype that converts to map of environment variables. Subclass this type to enable statically typed environment variables.

  • Allow TOSCA data types to declare a “transform” in metadata that is applied as a property transform.

  • node_filter improvements:

    • Recursively merge requirements keys in node filters when determining the node_filter for a requirement.

    • Allow get_nodes_of_type TOSCA function in node_filter match expressions.

  • Release 0.0.5 version of the Python tosca package.


  • Allow service templates to declare the unfurl version they are compatible with.

They can do this be declaring a repository for the unfurl package like so:

            revision: v0.9.1

Unfurl will still resolve imports in the unfurl package using the local installed version of unfurl but it will raise an error if it isn’t compatible with the version declared here.

  • Do semver compatibility check for 0.* versions.

Even though pre 1.0 versions aren’t expected to provide semver guarantees the alternative to doing the semver check is to treat every version as incompatible with another thus requiring every version reference to a package to be updated with each package update. This isn’t very useful, especially when developing against an unstable package.

  • Support file URLs in package rules.

Minor Enhancements and Notable Bug Fixes

  • parser allow merge keys to be optional, e.g. “+?/a/d”

  • loader: Add a UNFURL_OVERWRITE_POLICY environment variable to guide the loader’s python to yaml converter.

  • loader: Relax restrictions on from foo import * and other bug fixes with the DSL sandbox’s Python import loader.

  • init: apply UNFURL_SEARCH_ROOT to unfurl project search.

  • packages: If a package rule specifies a full url, preserve it when applying the rule.

v0.9.0 - Friday 10-13-23

Compare with 0.8.0


Introduce Python DSL for TOSCA

Write TOSCA as Python modules instead of YAML. Features include:

  • Static type checking of your TOSCA model.

  • IDE integration.

  • Export command now support conversion from YAML to Python and Python to YAML.

  • Python data model simplifies TOSCA YAML

  • But allows advanced constraints that encapsulate verbose relationships and node filters.

  • Python executes in a sandbox to safely parse untrusted TOSCA service templates.

See for more information.

Unfurl Cloud local development mode

You can now see local changes to a blueprint project under development on Unfurl Cloud by running unfurl serve . in the project directory. If the project was cloned from Unfurl Cloud, it will connect to that local server to render and deploy that local copy of the blueprint (for security, on your local browser only). Use the --cloud-server option to specify an alternative instance of Unfurl Cloud.

Embedded Blueprints (TOSCA substitution mapping)

Support for TOSCA substitution mapping has been stabilized and integrated into Unfurl Cloud.

One new feature level enhancement an extension to TOSCA’s requirements mapping to that enables you to essentially parameterize an embedded template by letting the outer (the embedding template) substitute node templates in the embedded template.

When substituted node template (in the outer topologies) declares requirements with whose name matches the name of a node template in the substituted (inner) topology then that node template will be replaced by the node template targeted by the requirement.

For example, if the substituted (inner) topology looked like:

    derived_from: tosca:Root
    - placeholder:
        node: PlaceHolderType

      node_type: NestedWithPlaceHolder

        type: PlaceHolderType

Then another topology that is embedding it can replace the “placeholder” template like so:

      type: NestedWithPlaceHolder
      - substitute
        - placeholder: replacement

      type: PlaceHolderType

CLI Improvements

  • Add --skip-upstream-check global option which skips pulling latest upstream changes from existing repositories and checking remote repositories for version tags.

Improvements to sub commands:

  • init Add Azure and Kubernetes project skeletons (unfurl init --skeleton k8s and unfurl init --skeleton azure)

  • clone Add --design flag which to configure the cloned project for blueprint development.

  • serve Add --cloud-server option to specify Unfurl Cloud instance to connect to.

  • export Add support for exporting TOSCA YAML to Python and Python to YAML; add --overwrite and --python-target options.

  • cloudmap Also allow host urls for options that had only accepted a pre-configured name of a repository host.

Dry Run improvements

  • When deploying jobs with --dryrun if a Mock operation is defined for a node type or template it will be invoked if the configurator doesn’t support dry run mode.

  • The terraform configurator now supports dryrun_mode and dryrun_outputs input parameters.

Runtime Eval Expressions

  • Eval expressions can now be used in node_filters to query for node matches using the new match keyword in the node filter.

  • Add a “.hosted_on” key to eval expressions that (recursively) follows “.targets” filtered by the HostedOn relationship, even across topology boundaries.

  • Add optional wantList parameter to the jinja2 eval filter to guarantee the result is a list e.g. {{"expression" | eval(wantList=True)}}

  • The trace keyword in eval expressions now accept break as a value. This will invoke Python’s breakpoint() function when the expression is evaluated.

New Environment Variables

  • UNFURL_SEARCH_ROOT environment variable: When search for ensembles and unfurl projects Unfurl will stop when it reach the directory this is set to.

  • UNFURL_SKIP_SAVE environment variable: If set, skips saving the ensemble.yaml to disk after a job runs,

Minor Enhancements and Notable Bug Fixes

  • artifacts: Add an asdf artifact for kompose and have the kompose configurator schedule the artifact if kompose is missing.

  • parser: treat import failures are fatal errors (abort parsing and validation).

  • parser: better syntax validation of +include statements.

  • cloudmap: allow anonymous connections to Gitlab and Unfurl Cloud api for readonly access

  • plan: fix spurious validation failures when creating initial environment variable rules

  • tosca:: Add support for ‘unsupported’, ‘deprecated’, and ‘removed’ property statuses.

  • tosca: Add support for bitrate scalar units.

  • tosca: fix built-in storage type definition

  • plan: fix candidate status check when deleting instances.

  • server: make sure the generated local unfurl.yaml exists when patching deployments.

  • packages: warn when remote lookup of version tags fails.

  • repo: fix matching local paths with repository paths.

  • loader: fix relative path lookups when processing an +include directives during a TOSCA import of a file in a repository.

  • logging: improve readability when pretty printing dictionaries.

0.8.0 - 2023-07-26

Compare with 0.7.1

Breaking Changes

  • The Unfurl package now only depends on the ansible-core package instead of the full ansible package. Unfurl projects that depend on ansible modules installed by that package will not work with new Unfurl installations unless it is installed by some other means – or, better, declare an Ansible collection artifact as a dependency on the template that requires it. For an example, see this usage in the docker-template.yaml.


  • Allow Ansible collections to be declared as TOSCA artifacts. Some predefined ones are defined here.

  • Unfurl Server now tracks the dependent repositories accessed when generating cached representations (e.g. for /export) and uses that information to invalidate cache items when the dependencies change.

  • Unfurl Server now caches more git operations, controlled by these environment variables: CACHE_DEFAULT_PULL_TIMEOUT (default: 120s) and CACHE_DEFAULT_REMOTE_TAGS_TIMEOUT (default: 300s)

  • Unfurl Server: add a /types endpoint that can extract types from a cloudmap.

  • API: allow simpler implementations

  • cloudmap: The cloudmap command now supports --import local.

  • eval: Unfurl’s Jinja2 filters that are marked as safe can now be used in safe evaluation mode (currently: eval, map_value, sensitive, and the to_*_label family).

Bug Fixes

  • jobs: Fix issue where some tasks that failed during the render phase were missing from the job summary.

  • jobs: Don’t apply the --dryrun option to the tasks that are installing local artifacts. (Since they will most likely be needed to execute the tasks in the job even in dryrun mode.)

  • k8s: Fix evaluation of the kubernetes_current_namespace() expression function outside of a job context

  • k8s: Filter out data keys with null values when generating a Kubernetes Secret resource.

  • helm: Fix check operation for unfurl.nodes.HelmRepository

  • helm: If the Kubernetes environment has the insecure flag set, pass --kube-insecure-skip-tls-verify to helm.