ayon-core/website/docs/dev_publishing.md

id	title	sidebar_label
dev_publishing	Publishing	Publishing

Publishing workflow consist of 2 parts:

• Creation - Mark what will be published and how.
• Publishing - Use data from creation to go through pyblish process.

OpenPype is using pyblish for publishing process. OpenPype a little bit extend and modify few functions mainly for reports and UI purposes. The main differences are that OpenPype's publish UI allows to enable/disable instances or plugins during creation part instead of in publishing part and has limited plugin actions only for failed validation plugins.

Creation

Concept of creation does not have to "create" anything but prepare and store metadata about an "instance" (becomes a subset after publish process). Created instance always has family which defines what kind of data will be published, best example is workfile family. Storing of metadata is host specific and may be even a Creator plugin specific. In most of hosts are metadata stored to workfile (Maya scene, Nuke script, etc.) to an item or a node the same way so consistency of host implementation is kept, but some features may require different approach that is the reason why it is creator plugin responsibility. Storing the metadata to workfile gives ability to keep values so artist does not have to do create and set what should be published and how over and over.

Created instance

Objected representation of created instance metadata defined by class CreatedInstance. Has access to CreateContext and BaseCreator that initialized the object. Is dictionary like object with few immutable keys (marked with start * in table). The immutable keys are set by creator plugin or create context on initialization and thei values can't change. Instance can have more arbitrary data, for example ids of nodes in scene but keep in mind that some keys are reserved.

Key	Type	Description
*id	str	Identifier of metadata type. ATM constant "pyblish.avalon.instance"
*instance_id	str	Unique ID of instance. Set automatically on instance creation using str(uuid.uuid4())
*family	str	Instance's family representing type defined by creator plugin.
*creator_identifier	str	Identifier of creator that collected/created the instance.
*creator_attributes	dict	Dictionary of attributes that are defined by creator plugin (get_instance_attr_defs).
*publish_attributes	dict	Dictionary of attributes that are defined by publish plugins.
variant	str	Variant is entered by artist on creation and may affect subset.
subset	str	Name of instance. This name will be used as subset name during publishing. Can be changed on context change or variant change.
active	bool	Is instance active and will be published or not.
asset	str	Name of asset in which context was created.
task	str	Name of task in which context was created. Can be set to None.

:::note Task should not be required until subset name template expect it. :::

object of CreatedInstance has method data_to_store which returns dictionary that can be parsed to json string. This method will return all data related to instance so can be re-created using CreatedInstance.from_existing(data).

Create context

Controller and wrapper around creation is CreateContext which cares about loading of plugins needed for creation. And validates required functions in host implementation.

Context discovers creator and publish plugins. Trigger collections of existing instances on creators and trigger creation itself. Also keeps in mind instance objects by their ids.

Creator plugins can call creator_adds_instance or creator_removed_instance to add/remove instance but these methods are not meant to be called directly out of creator. The reason is that is creator's responsibility to remove metadata or decide if should remove the instance.

Required functions in host implementation

Host implementation must have implemented get_context_data and update_context_data. These two functions are needed to store metadata that are not related to any instane but are needed for creation and publishing process. Right now are there stored data about enabled/disabled optional publish plugins. When data are not stored and loaded properly reset of publishing will cause that they will be set to default value. Similar to instance data can be context data also parsed to json string.

There are also few optional functions. For UI purposes it is possible to implement get_context_title which can return string showed in UI as a title. Output string may contain html tags. It is recommended to return context path (it will be created function this purposes) in this order "{project name}/{asset hierarchy}/<b>{asset name}</b>/{task name}".

Another optional function is get_current_context. This function is handy in hosts where is possible to open multiple workfiles in one process so using global context variables is not relevant because artist can switch between opened workfiles without being acknowledged. When function is not implemented or won't return right keys the global context is used.

# Expected keys in output
{
    "project_name": "MyProject",
    "asset_name": "sq01_sh0010",
    "task_name": "Modeling"
}

Create plugin

Main responsibility of create plugin is to create, update, collect and remove instance metadata and propagate changes to create context. Has access to CreateContext (self.create_context) that discovered the plugin so has also access to other creators and instances. Create plugins have a lot of responsibility so it is recommended to implement common code per host.

BaseCreator

Base implementation of creator plugin. It is not recommended to use this class as base for production plugins but rather use one of AutoCreator and Creator variants.

Abstractions

• family (class attr) - Tells what kind of instance will be created.

class WorkfileCreator(Creator):
    family = "workfile"

• collect_instances (method) - Collect already existing instances from workfile and add them to create context. This method is called on initialization or reset of CreateContext. Each creator is responsible to find it's instances metadata, convert them to CreatedInstance object and add the to create context (self._add_instance_to_context(instnace_obj)).

def collect_instances(self):
    # Using 'pipeline.list_instances' is just example how to get existing instances from scene
    # - getting existing instances is different per host implementation
    for instance_data in pipeline.list_instances():
        # Process only instances that were created by this creator
        creator_id = instance_data.get("creator_identifier")
        if creator_id == self.identifier:
            # Create instance object from existing data
            instance = CreatedInstance.from_existing(
                instance_data, self
            )
            # Add instance to create context
            self._add_instance_to_context(instance)

• create (method) - Create new object of CreatedInstance store it's metadata to workfile and add the instance into create context. Failed creation should raise CreatorError if happens error that can artist fix or give him some useful information. Trigger and implementation differs for Creator and AutoCreator.

• update_instances (method) - Update data of instances. Receives tuple with instance and changes.

def update_instances(self, update_list):
    # Loop over changed instances
    for instance, changes in update_list:
        # Example possible usage of 'changes' to use different node on change
        #   of node id in instance data (MADE UP)
        node = None
        if "node_id" in changes:
            old_value, new_value = changes["node_id"]
            if new_value is not None:
                node = pipeline.get_node_by_id(new_value)

        if node is None:
            node = pipeline.get_node_by_instance_id(instance.id)
        # Get node in scene that represents the instance
        # Imprind data to a node
        pipeline.imprint(node, instance.data_to_store())


# Most implementations will probably ignore 'changes' completely
def update_instances(self, update_list):
    for instance, _ in update_list:
        # Get node from scene
        node = pipeline.get_node_by_instance_id(instance.id)
        # Imprint data to node
        pipeline.imprint(node, instance.data_to_store())

• remove_instances (method) - Remove instance metadata from workfile and from create context.

# Possible way how to remove instance
def remove_instances(self, instances):
    for instance in instances:
        # Remove instance metadata from workflle
        pipeline.remove_instance(instance.id)
        # Remove instance from create context
        self._remove_instance_from_context(instance)


# Default implementation of `AutoCreator`
def remove_instances(self, instances):
    pass

:::note When host implementation use universal way how to store and load instances you should implement host specific creator plugin base class with implemented collect_instances, update_instances and remove_instances. :::

Optional implementations

• enabled (attr) - Boolean if creator plugin is enabled and used.
• identifier (class attr) - Consistent unique string identifier of the creator plugin. Is used to identify source plugin of existing instances. There can't be 2 creator plugins with same identifier. Default implementation returns family attribute.

class RenderLayerCreator(Creator):
    family = "render"
    identifier = "render_layer"


class RenderPassCreator(Creator):
    family = "render"
    identifier = "render_pass"

• label (attr) - String label of creator plugin which will showed in UI, identifier is used when not set. It should be possible to use html tags.

class RenderLayerCreator(Creator):
    label = "Render Layer"

• get_icon (attr) - Icon of creator and it's instances. Value can be a path to image file, full name of qtawesome icon, QPixmap or QIcon. For complex cases or cases when Qt objects are returned it is recommended to override get_icon method and handle the logic or import Qt inside the method to not break headless usage of creator plugin. For list of qtawesome icons check qtawesome github repository (look for used version in pyproject.toml). Default implementation return icon attribute.
• icon (method) - Attribute for default implementation of get_icon.

class RenderLayerCreator(Creator):
    # Use font awesome 5 icon
    icon = "fa5.building"

• get_instance_attr_defs (method) - Attribute definitions of instance. Creator can define attribute values with default values for each instance. These attributes may affect how will be instance processed during publishing. Attribute defiitions can be used from openpype.pipeline.lib.attribute_definitions (NOTE: Will be moved to openpype.lib.attribute_definitions soon). Attribute definitions define basic type of values for different cases e.g. boolean, number, string, enumerator, etc. Default implementations returns instance_attr_defs.
• instance_attr_defs (attr) - Attribute for default implementation of get_instance_attr_defs.

from openpype.pipeline import attribute_definitions


class RenderLayerCreator(Creator):
    def get_instance_attr_defs(self):
        # Return empty list if '_allow_farm_render' is not enabled (can be set during initialization)
        if not self._allow_farm_render:
            return []
        # Give artist option to change if should be rendered on farm or locally
        return [
            attribute_definitions.BoolDef(
                "render_farm",
                default=False,
                label="Render on Farm"
            )
        ]

• get_subset_name (method) - Calculate subset name based on passed data. Data can be extended using get_dynamic_data method. Default implementation is using get_subset_name from openpype.lib which is recommended.

• get_dynamic_data (method) - Can be used to extend data for subset template which may be required in some cases.

AutoCreator

Creator that is triggered on reset of create context. Can be used for families that are expected to be created automatically without artist interaction (e.g. workfile). Method create is triggered after collecting of all creators.

:::important AutoCreator has implemented remove_instances to do nothing as removing of auto created instances would lead to create new instance immediately or on refresh. :::

def __init__(
    self, create_context, system_settings, project_settings, *args, **kwargs
):
    super(MyCreator, self).__init__(
        create_context, system_settings, project_settings, *args, **kwargs
    )
    # Get variant value from settings
    variant_name = (
        project_settings["my_host"][self.identifier]["variant"]
    ).strip()
    if not variant_name:
        variant_name = "Main"
    self._variant_name = variant_name

# Create does not expect any arguments
def create(self):
    # Look for existing instance in create  context
    existing_instance = None
    for instance in self.create_context.instances:
        if instance.creator_identifier == self.identifier:
            existing_instance = instance
            break

    # Collect current context information
    # - variant can be filled from settings
    variant = self._variant_name
    # Only place where we can look for current context
    project_name = io.Session["AVALON_PROJECT"]
    asset_name = io.Session["AVALON_ASSET"]
    task_name = io.Session["AVALON_TASK"]
    host_name = io.Session["AVALON_APP"]

    # Create new instance if does not exist yet
    if existing_instance is None:
        asset_doc = io.find_one({"type": "asset", "name": asset_name})
        subset_name = self.get_subset_name(
            variant, task_name, asset_doc, project_name, host_name
        )
        data = {
            "asset": asset_name,
            "task": task_name,
            "variant": variant
        }
        data.update(self.get_dynamic_data(
            variant, task_name, asset_doc, project_name, host_name
        ))

        new_instance = CreatedInstance(
            self.family, subset_name, data, self
        )
        self._add_instance_to_context(new_instance)

    # Update instance context if is not the same
    elif (
        existing_instance["asset"] != asset_name
        or existing_instance["task"] != task_name
    ):
        asset_doc = io.find_one({"type": "asset", "name": asset_name})
        subset_name = self.get_subset_name(
            variant, task_name, asset_doc, project_name, host_name
        )
        existing_instance["asset"] = asset_name
        existing_instance["task"] = task_name

Creator

Implementation of creator plugin that is triggered manually by artist in UI (or by code). Has extended options for UI purposes than AutoCreator and create method expect more arguments.

Abstractions

• create (method) - Code where creation of metadata

Optional implementations

• create_allow_context_change (class attr) - Allow to set context in UI before creation. Some creator may not allow it or their logic would not use the context selection (e.g. bulk creators). Is set to True but default.

class BulkRenderCreator(Creator):
    create_allow_context_change = False

• get_default_variants (method) - Returns list of default variants that are listed in create dialog for user. Returns default_variants attribute by default.

• default_variants (attr) - Attribute for default implementation of get_default_variants.

• get_default_variant (method) - Returns default variant that is prefilled in UI (value does not have to be in default variants). By default returns default_variant attribute. If returns None then UI logic will take first item from get_default_variants if there is any otherwise "Main" is used.

• default_variant (attr) - Attribute for default implementation of get_default_variant.

• get_description (method) - Returns short string description of creator. Returns description attribute by default.

• description (attr) - Attribute for default implementation of get_description.

• get_detailed_description (method) - Returns detailed string description of creator. Can contain markdown. Returns detailed_description attribute by default.

• detailed_description (attr) - Attribute for default implementation of get_detailed_description.

• get_pre_create_attr_defs (method) - Similar to get_instance_attr_defs returns attribute definitions but they are filled before creation. When creation is called from UI the values are passed to create method. Returns pre_create_attr_defs attribute by default.

• pre_create_attr_defs (attr) - Attribute for default implementation of get_pre_create_attr_defs.

from openpype.pipeline import Creator, attribute_definitions


class CreateRender(Creator):
    family = "render"
    label = "Render"
    icon = "fa.eye"
    description = "Render scene viewport"

    def __init__(
        self, context, system_settings, project_settings, *args, **kwargs
    ):
        super(CreateRender, self).__init__(
            context, system_settings, project_settings, *args, **kwargs
        )
        plugin_settings = (
            project_settings["my_host"]["create"][self.__class__.__name__]
        )
        # Get information if studio has enabled farm publishing
        self._allow_farm_render = plugin_settings["allow_farm_render"]
        # Get default variants from settings
        self.default_variants = plugin_settings["variants"]

    def get_instance_attr_defs(self):
        # Return empty list if '_allow_farm_render' is not enabled (can be set during initialization)
        if not self._allow_farm_render:
            return []
        # Give artist option to change if should be rendered on farm or locally
        return [
            attribute_definitions.BoolDef(
                "render_farm",
                default=False,
                label="Render on Farm"
            )
        ]

    def get_pre_create_attr_defs(self):
        # Give user option to use selection or not
        attrs = [
            attribute_definitions.BoolDef(
                "use_selection",
                default=False,
                label="Use selection"
            )
        ]
        if self._allow_farm_render:
            # Set to render on farm in creator dialog
            # - this value is not automatically passed to instance attributes
            #   creator must do that during creation
            attrs.append(
                attribute_definitions.BoolDef(
                    "render_farm",
                    default=False,
                    label="Render on Farm"
                )
            )
        return attrs

    def create(self, subset_name, instance_data, pre_create_data):
        # ARGS:
        # - 'subset_name' - precalculated subset name
        # - 'instance_data' - context data
        #    - 'asset' - asset name
        #    - 'task' - task name
        #    - 'variant' - variant
        #    - 'family' - instnace family

        # Check if should use selection or not
        if pre_create_data.get("use_selection"):
            items = pipeline.get_selection()
        else:
            items = [pipeline.create_write()]

        # Validations related to selection
        if len(items) > 1:
            raise CreatorError("Please select only single item at time.")

        elif not items:
            raise CreatorError("Nothing to create. Select at least one item.")

        # Create instence object
        new_instance = CreatedInstance(self.family, subset_name, data, self)
        # Pass value from pre create attribute to instance
        # - use them only when pre create date contain the data
        if "render_farm" in pre_create_data:
            use_farm = pre_create_data["render_farm"]
            new_instance.creator_attributes["render_farm"] = use_farm

        # Store metadata to workfile
        pipeline.imprint(new_instance.id, new_instance.data_to_store())

        # Add instance to context
        self._add_instance_to_context(new_instance)

Publish

Exceptions

OpenPype define few specific exceptions that should be used in publish plugins.

Validation exception

Validation plugins should raise PublishValidationError to show to an artist what's wrong and give him actions to fix it. The exception says that error happened in plugin can be fixed by artist himself (with or without action on plugin). Any other errors will stop publishing immediately. Exception PublishValidationError raised after validation order has same effect as any other exception.

Exception PublishValidationError expects 4 arguments:

• message Which is not used in UI but for headless publishing.
• title Short description of error (2-5 words). Title is used for grouping of exceptions per plugin.
• description Detailed description of happened issue where markdown and html can be used.
• detail Is optional to give even more detailed information for advanced users. At this moment is detail showed under description but it is in plan to have detail in collapsible widget.

Extended version is PublishXmlValidationError which uses xml files with stored descriptions. This helps to avoid having huge markdown texts inside code. The exception has 4 arguments:

• plugin The plugin object which raises the exception to find it's related xml file.
• message Exception message for publishing without UI or different pyblish UI.
• key Optional argument says which error from xml is used as validation plugin may raise error with different messages based on the current errors. Default is "main".
• formatting_data Optional dictionary to format data in the error. This is used to fill detailed description with data from the publishing so artist can get more precise information.

Where and how to create xml file

Xml files for PublishXmlValidationError must be located in ./help subfolder next to plugin and the filename must match the filename of plugin.

# File location related to plugin file
└ publish
  ├ help
  │ ├ validate_scene.xml
  │ └ ...
  ├ validate_scene.py
  └ ...

Xml file content has &ltroot&gt node which may contain any amount of &lterror&gt nodes, but each of them must have id attribute with unique value. That is then used for key. Each error must have &lttitle&gt and &ltdescription&gt and &ltdetail&gt. Text content may contain python formatting keys that can be filled when exception is raised.

<?xml version="1.0" encoding="UTF-8"?>
<root>
    <error id="main">
        <title>Subset context</title>
        <description>## Invalid subset context

Context of the given subset doesn't match your current scene.

### How to repair?

Yout can fix this with "Repair" button on the right. This will use '{expected_asset}' asset name and overwrite '{found_asset}' asset name in scene metadata.

After that restart publishing with Reload button.
        </description>
        <detail>
### How could this happen?

The subset was created in different scene with different context
or the scene file was copy pasted from different context.
        </detail>
    </error>
</root>

Known errors

When there is a known error that can't be fixed by user (e.g. can't connect to deadline service, etc.) KnownPublishError should be raise. The only difference is that it's message is shown in UI to artist otherwise a neutral message without context is shown.

Plugin extension

Publish plugins can be extended by additional logic when inherits from OpenPypePyblishPluginMixin which can be used as mixin (additional inheritance of class). Publish plugins that inherit from this mixin can define attributes that will be shown in CreatedInstance. One of most important usages is to be able turn on/off optional plugins.

Attributes are defined by return value of get_attribute_defs method. Attribute definitions are for families defined in plugin's families attribute if it's instance plugin or for whole context if it's context plugin. To convert existing values (or to remove legacy values) can be re-implemented convert_attribute_values. Default implementation just converts the values to right types.

:::important Values of publish attributes from created instance are never removed automatically so implementing of this method is best way to remove legacy data or convert them to new data structure. :::

Possible attribute definitions can be found in openpype/pipeline/lib/attribute_definitions.py.

import pyblish.api
from openpype.pipeline import (
    OpenPypePyblishPluginMixin,
    attribute_definitions,
)


# Example context plugin
class MyExtendedPlugin(
    pyblish.api.ContextPlugin, OpenPypePyblishPluginMixin
):
    optional = True
    active = True

    @classmethod
    def get_attribute_defs(cls):
        return [
            attribute_definitions.BoolDef(
                # Key under which it will be stored
                "process",
                # Use 'active' as default value
                default=cls.active,
                # Use plugin label as label for attribute
                label=cls.label
            )
        ]

    def process_plugin(self, context):
        # First check if plugin is optional
        if not self.optional:
            return True

        # Get 'process' key
        process_value = (
            context.data
            .get("publish_attributes", {})
            # Attribute values are stored by class names
            .get(self.__class__.__name__, {})
            # Access the key
            .get("process")
        )
        if process_value or process_value is None:
            return True
        return False

    def process(self, context):
        if not self.process_plugin(context):
            return
        # Do plugin logic
        ...