Arkindex Workers

arkindex_worker.worker

Base classes to implement Arkindex workers.

Classes

ActivityState

Bases: Enum

Processing state of an element.

Attributes

Queued class-attribute instance-attribute

Queued = 'queued'

The element has not yet been processed by a worker.

Started class-attribute instance-attribute

Started = 'started'

The element is being processed by a worker.

Processed class-attribute instance-attribute

Processed = 'processed'

The element has been successfully processed by a worker.

Error class-attribute instance-attribute

Error = 'error'

An error occurred while processing this element.

ElementsWorker

ElementsWorker(
    description: str = "Arkindex Elements Worker",
    support_cache: bool = False,
)

Bases: BaseWorker, ClassificationMixin, ElementMixin, TranscriptionMixin, WorkerVersionMixin, EntityMixin, MetaDataMixin

Base class for ML workers that operate on Arkindex elements.

This class inherits from numerous mixin classes found in other modules of arkindex.worker, which provide helpers to read and write to the Arkindex API.

Parameters:

Name	Type	Description	Default
description	str	The worker’s description	'Arkindex Elements Worker'
support_cache	bool	Whether the worker supports cache	False

Source code in arkindex_worker/worker/__init__.py

def __init__(
    self, description: str = "Arkindex Elements Worker", support_cache: bool = False
):
    """
    :param description: The worker's description
    :param support_cache: Whether the worker supports cache
    """
    super().__init__(description, support_cache)

    # Add mandatory argument to process elements
    self.parser.add_argument(
        "--elements-list",
        help="JSON elements list to use",
        type=open,
        default=os.environ.get("TASK_ELEMENTS"),
    )
    self.parser.add_argument(
        "--element",
        type=uuid.UUID,
        nargs="+",
        help="One or more Arkindex element ID",
    )

    self.classes = {}

    self.entity_types = {}
    """Known and available entity types in processed corpus
    """

    self._worker_version_cache = {}

Attributes

entity_types instance-attribute

entity_types = {}

Known and available entity types in processed corpus

store_activity property

store_activity: bool

Whether or not WorkerActivity support has been enabled on the DataImport used to run this worker.

Functions

list_elements

list_elements() -> Iterable[CachedElement] | list[str]

List the elements to be processed, either from the CLI arguments or the cache database when enabled.

Returns:

Type	Description
Iterable[CachedElement] | list[str]	An iterable of CachedElement when cache support is enabled, and a list of strings representing element IDs otherwise.

Source code in arkindex_worker/worker/__init__.py

def list_elements(self) -> Iterable[CachedElement] | list[str]:
    """
    List the elements to be processed, either from the CLI arguments or
    the cache database when enabled.

    :return: An iterable of [CachedElement][arkindex_worker.cache.CachedElement] when cache support is enabled,
       and a list of strings representing element IDs otherwise.
    """
    assert not (
        self.args.elements_list and self.args.element
    ), "elements-list and element CLI args shouldn't be both set"
    out = []

    # Load from the cache when available
    # Flake8 wants us to use 'is True', but Peewee only supports '== True'
    cache_query = CachedElement.select().where(
        CachedElement.initial == True  # noqa: E712
    )
    if self.use_cache and cache_query.exists():
        return cache_query
    # Process elements from JSON file
    elif self.args.elements_list:
        data = json.load(self.args.elements_list)
        assert isinstance(data, list), "Elements list must be a list"
        assert len(data), "No elements in elements list"
        out += list(filter(None, [element.get("id") for element in data]))
    # Add any extra element from CLI
    elif self.args.element:
        out += self.args.element

    return out

configure

configure()

Setup the worker using CLI arguments and environment variables.

Source code in arkindex_worker/worker/__init__.py

def configure(self):
    """
    Setup the worker using CLI arguments and environment variables.
    """
    # CLI args are stored on the instance so that implementations can access them
    self.args = self.parser.parse_args()

    if self.is_read_only:
        super().configure_for_developers()
    else:
        super().configure()
        super().configure_cache()

    # Retrieve the model configuration
    if self.model_configuration:
        self.config.update(self.model_configuration)
        logger.info("Model version configuration retrieved")

    # Retrieve the user configuration
    if self.user_configuration:
        self.config.update(self.user_configuration)
        logger.info("User configuration retrieved")

run

run()

Implements an Arkindex worker that goes through each element returned by list_elements. It calls process_element, catching exceptions, and handles saving WorkerActivity updates when enabled.

Source code in arkindex_worker/worker/__init__.py

def run(self):
    """
    Implements an Arkindex worker that goes through each element returned by
    [list_elements][arkindex_worker.worker.ElementsWorker.list_elements].
    It calls [process_element][arkindex_worker.worker.ElementsWorker.process_element],
    catching exceptions, and handles saving WorkerActivity updates when enabled.
    """
    self.configure()

    # List all elements either from JSON file
    # or direct list of elements on CLI
    elements = self.list_elements()
    if not elements:
        logger.warning("No elements to process, stopping.")
        sys.exit(1)

    if not self.store_activity:
        logger.info(
            "No worker activity will be stored as it is disabled for this process"
        )

    # Process every element
    count = len(elements)
    failed = 0
    for i, item in enumerate(elements, start=1):
        element = None
        try:
            if self.use_cache:
                # Just use the result of list_elements as the element
                element = item
            else:
                # Load element using the Arkindex API
                element = Element(**self.request("RetrieveElement", id=item))

            logger.info(f"Processing {element} ({i}/{count})")

            # Process the element and report its progress if activities are enabled
            if self.update_activity(element.id, ActivityState.Started):
                self.process_element(element)
                self.update_activity(element.id, ActivityState.Processed)
            else:
                logger.info(
                    f"Skipping element {element.id} as it was already processed"
                )
                continue
        except Exception as e:
            # Handle errors occurring while retrieving, processing or patching the activity for this element.
            # Count the element as failed in case the activity update to "started" failed with no conflict.
            # This prevent from processing the element
            failed += 1

            # Handle the case where we failed retrieving the element
            element_id = element.id if element else item

            if isinstance(e, ErrorResponse):
                message = f"An API error occurred while processing element {element_id}: {e.title} - {e.content}"
            else:
                message = (
                    f"Failed running worker on element {element_id}: {repr(e)}"
                )

            logger.warning(
                message,
                exc_info=e if self.args.verbose else None,
            )
            if element:
                # Try to update the activity to error state regardless of the response
                with contextlib.suppress(Exception):
                    self.update_activity(element.id, ActivityState.Error)

    message = f'Ran on {count} element{"s"[:count>1]}: {count - failed} completed, {failed} failed'
    if failed:
        logger.error(message)
        if failed >= count:  # Everything failed!
            sys.exit(1)
    else:
        logger.info(message)

process_element

process_element(element: Element | CachedElement)

Override this method to implement your worker and process a single Arkindex element at once.

Parameters:

Name	Type	Description	Default
element	Element | CachedElement	The element to process. Will be a CachedElement instance if cache support is enabled, and an Element instance otherwise.	required

Source code in arkindex_worker/worker/__init__.py

def process_element(self, element: Element | CachedElement):
    """
    Override this method to implement your worker and process a single Arkindex element at once.

    :param element: The element to process.
       Will be a CachedElement instance if cache support is enabled,
       and an Element instance otherwise.
    """

update_activity

update_activity(
    element_id: str | uuid.UUID, state: ActivityState
) -> bool

Update the WorkerActivity for this element and worker.

Parameters:

Name	Type	Description	Default
element_id	str | UUID	ID of the element.	required
state	ActivityState	New WorkerActivity state for this element.	required

Returns:

Type	Description
bool	True if the update has been successful or WorkerActivity support is disabled. False if the update has failed due to a conflict; this worker might have already processed this element.

Source code in arkindex_worker/worker/__init__.py

def update_activity(
    self, element_id: str | uuid.UUID, state: ActivityState
) -> bool:
    """
    Update the WorkerActivity for this element and worker.

    :param element_id: ID of the element.
    :param state: New WorkerActivity state for this element.
    :returns: True if the update has been successful or WorkerActivity support is disabled.
       False if the update has failed due to a conflict; this worker might have already processed
       this element.
    """
    if not self.store_activity:
        logger.debug(
            "Activity is not stored as the feature is disabled on this process"
        )
        return True

    assert element_id and isinstance(
        element_id, uuid.UUID | str
    ), "element_id shouldn't be null and should be an UUID or str"
    assert isinstance(state, ActivityState), "state should be an ActivityState"

    try:
        self.request(
            "UpdateWorkerActivity",
            id=self.worker_run_id,
            body={
                "element_id": str(element_id),
                "process_id": self.process_information["id"],
                "state": state.value,
            },
        )
    except ErrorResponse as e:
        if state == ActivityState.Started and e.status_code == 409:
            # 409 conflict error when updating the state of an activity to "started" mean that we
            # cannot process this element. We assume that the reason is that the state transition
            # was forbidden i.e. that the activity was already in a started or processed state.
            # This allow concurrent access to an element activity between multiple processes.
            # Element should not be counted as failed as it is probably handled somewhere else.
            logger.debug(
                f"Cannot start processing element {element_id} due to a conflict. "
                f"Another process could have processed it with the same version already."
            )
            return False
        logger.warning(
            f"Failed to update activity of element {element_id} to {state.value} due to an API error: {e.content}"
        )
        raise e

    logger.debug(f"Updated activity of element {element_id} to {state}")
    return True

MissingDatasetArchive

Bases: Exception

Exception raised when the compressed archive associated to a dataset isn’t found in its task artifacts.

DatasetWorker

DatasetWorker(
    description: str = "Arkindex Dataset Worker",
    support_cache: bool = False,
)

Bases: BaseWorker, DatasetMixin, TaskMixin

Base class for ML workers that operate on Arkindex dataset sets.

This class inherits from numerous mixin classes found in other modules of arkindex.worker, which provide helpers to read and write to the Arkindex API.

Parameters:

Name	Type	Description	Default
description	str	The worker’s description.	'Arkindex Dataset Worker'
support_cache	bool	Whether the worker supports cache.	False

Source code in arkindex_worker/worker/__init__.py

def __init__(
    self,
    description: str = "Arkindex Dataset Worker",
    support_cache: bool = False,
):
    """
    :param description: The worker's description.
    :param support_cache: Whether the worker supports cache.
    """
    super().__init__(description, support_cache)

    # Path to the dataset compressed archive (containing images and a SQLite database)
    # Set as an instance variable as dataset workers might use it to easily extract its content
    self.downloaded_dataset_artifact: Path | None = None

    self.parser.add_argument(
        "--set",
        type=check_dataset_set,
        nargs="+",
        help="""
            One or more Arkindex dataset sets, format is <dataset_uuid>:<set_name>
            (e.g.: "12341234-1234-1234-1234-123412341234:train")
        """,
        default=[],
    )

Functions

configure

configure()

Setup the worker using CLI arguments and environment variables.

Source code in arkindex_worker/worker/__init__.py

def configure(self):
    """
    Setup the worker using CLI arguments and environment variables.
    """
    # CLI args are stored on the instance so that implementations can access them
    self.args = self.parser.parse_args()

    if self.is_read_only:
        super().configure_for_developers()
    else:
        super().configure()
        super().configure_cache()

    # Retrieve the model configuration
    if self.model_configuration:
        self.config.update(self.model_configuration)
        logger.info("Model version configuration retrieved")

    # Retrieve the user configuration
    if self.user_configuration:
        self.config.update(self.user_configuration)
        logger.info("User configuration retrieved")

cleanup_downloaded_artifact

cleanup_downloaded_artifact() -> None

Cleanup the downloaded dataset artifact if any

Source code in arkindex_worker/worker/__init__.py

def cleanup_downloaded_artifact(self) -> None:
    """
    Cleanup the downloaded dataset artifact if any
    """
    if not self.downloaded_dataset_artifact:
        return

    self.downloaded_dataset_artifact.unlink(missing_ok=True)

download_dataset_artifact

download_dataset_artifact(dataset: Dataset) -> None

Find and download the compressed archive artifact describing a dataset using the list_artifacts and download_artifact methods.

Parameters:

Name	Type	Description	Default
dataset	Dataset	The dataset to retrieve the compressed archive artifact for.	required

Raises:

Type	Description
MissingDatasetArchive	When the dataset artifact is not found.

Source code in arkindex_worker/worker/__init__.py

def download_dataset_artifact(self, dataset: Dataset) -> None:
    """
    Find and download the compressed archive artifact describing a dataset using
    the [list_artifacts][arkindex_worker.worker.task.TaskMixin.list_artifacts] and
    [download_artifact][arkindex_worker.worker.task.TaskMixin.download_artifact] methods.

    :param dataset: The dataset to retrieve the compressed archive artifact for.
    :raises MissingDatasetArchive: When the dataset artifact is not found.
    """
    extra_dir = self.find_extras_directory()
    archive = extra_dir / dataset.filepath
    if archive.exists():
        return

    # Cleanup the dataset artifact that was downloaded previously
    self.cleanup_downloaded_artifact()

    logger.info(f"Downloading artifact for {dataset}")
    task_id = uuid.UUID(dataset.task_id)
    for artifact in self.list_artifacts(task_id):
        if artifact.path != dataset.filepath:
            continue

        archive.write_bytes(self.download_artifact(task_id, artifact).read())
        self.downloaded_dataset_artifact = archive
        return

    raise MissingDatasetArchive(
        "The dataset compressed archive artifact was not found."
    )

process_set

process_set(set: Set)

Override this method to implement your worker and process a single Arkindex dataset set at once.

Parameters:

Name	Type	Description	Default
set	Set	The set to process.	required

Source code in arkindex_worker/worker/__init__.py

def process_set(self, set: Set):
    """
    Override this method to implement your worker and process a single Arkindex dataset set at once.

    :param set: The set to process.
    """

list_sets

list_sets() -> Iterator[Set]

List the sets to be processed, either from the CLI arguments or using the list_process_sets method.

Returns:

Type	Description
Iterator[Set]	An iterator of Set objects.

Source code in arkindex_worker/worker/__init__.py

def list_sets(self) -> Iterator[Set]:
    """
    List the sets to be processed, either from the CLI arguments or using the
    [list_process_sets][arkindex_worker.worker.dataset.DatasetMixin.list_process_sets] method.

    :returns: An iterator of ``Set`` objects.
    """
    if not self.is_read_only:
        yield from self.list_process_sets()

    datasets: dict[uuid.UUID, Dataset] = {}
    for dataset_id, set_name in self.args.set:
        # Retrieving dataset information is not already cached
        if dataset_id not in datasets:
            datasets[dataset_id] = Dataset(
                **self.request("RetrieveDataset", id=dataset_id)
            )

        yield Set(name=set_name, dataset=datasets[dataset_id])

run

run()

Implements an Arkindex worker that goes through each dataset set returned by list_sets.

It calls process_set, catching exceptions.

Source code in arkindex_worker/worker/__init__.py

def run(self):
    """
    Implements an Arkindex worker that goes through each dataset set returned by
    [list_sets][arkindex_worker.worker.DatasetWorker.list_sets].

    It calls [process_set][arkindex_worker.worker.DatasetWorker.process_set],
    catching exceptions.
    """
    self.configure()

    dataset_sets: list[Set] = list(self.list_sets())
    if not dataset_sets:
        logger.warning("No sets to process, stopping.")
        sys.exit(1)

    # Process every set
    count = len(dataset_sets)
    failed = 0
    for i, dataset_set in enumerate(dataset_sets, start=1):
        try:
            assert (
                dataset_set.dataset.state == DatasetState.Complete.value
            ), "When processing a set, its dataset state should be Complete."

            logger.info(f"Retrieving data for {dataset_set} ({i}/{count})")
            self.download_dataset_artifact(dataset_set.dataset)

            logger.info(f"Processing {dataset_set} ({i}/{count})")
            self.process_set(dataset_set)
        except Exception as e:
            # Handle errors occurring while retrieving or processing this dataset set
            failed += 1

            if isinstance(e, ErrorResponse):
                message = f"An API error occurred while processing {dataset_set}: {e.title} - {e.content}"
            else:
                message = f"Failed running worker on {dataset_set}: {repr(e)}"

            logger.warning(message, exc_info=e if self.args.verbose else None)

    # Cleanup the latest downloaded dataset artifact
    self.cleanup_downloaded_artifact()

    message = f'Ran on {count} set{"s"[:count>1]}: {count - failed} completed, {failed} failed'
    if failed:
        logger.error(message)
        if failed >= count:  # Everything failed!
            sys.exit(1)
    else:
        logger.info(message)