arkindex-base-worker 0.5.2a1__tar.gz → 0.5.2b1__tar.gz

{arkindex_base_worker-0.5.2a1 → arkindex_base_worker-0.5.2b1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: arkindex-base-worker
-Version: 0.5.2a1
+Version: 0.5.2b1
 Summary: Base Worker to easily build Arkindex ML workflows
 Author-email: Teklia <contact@teklia.com>
 Maintainer-email: Teklia <contact@teklia.com>
@@ -23,8 +23,9 @@ Requires-Dist: humanize==4.15.0
 Requires-Dist: peewee~=3.17
 Requires-Dist: Pillow==11.3.0
 Requires-Dist: python-gnupg==0.5.6
+Requires-Dist: python-magic==0.4.27
 Requires-Dist: shapely==2.0.6
-Requires-Dist: teklia-toolbox==0.1.12
+Requires-Dist: teklia-toolbox==0.1.13
 Requires-Dist: zstandard==0.25.0
 Provides-Extra: tests
 Requires-Dist: pytest-mock==3.15.1; extra == "tests"

{arkindex_base_worker-0.5.2a1 → arkindex_base_worker-0.5.2b1}/arkindex_base_worker.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: arkindex-base-worker
-Version: 0.5.2a1
+Version: 0.5.2b1
 Summary: Base Worker to easily build Arkindex ML workflows
 Author-email: Teklia <contact@teklia.com>
 Maintainer-email: Teklia <contact@teklia.com>
@@ -23,8 +23,9 @@ Requires-Dist: humanize==4.15.0
 Requires-Dist: peewee~=3.17
 Requires-Dist: Pillow==11.3.0
 Requires-Dist: python-gnupg==0.5.6
+Requires-Dist: python-magic==0.4.27
 Requires-Dist: shapely==2.0.6
-Requires-Dist: teklia-toolbox==0.1.12
+Requires-Dist: teklia-toolbox==0.1.13
 Requires-Dist: zstandard==0.25.0
 Provides-Extra: tests
 Requires-Dist: pytest-mock==3.15.1; extra == "tests"

{arkindex_base_worker-0.5.2a1 → arkindex_base_worker-0.5.2b1}/arkindex_base_worker.egg-info/requires.txt

@@ -2,8 +2,9 @@ humanize==4.15.0
 peewee~=3.17
 Pillow==11.3.0
 python-gnupg==0.5.6
+python-magic==0.4.27
 shapely==2.0.6
-teklia-toolbox==0.1.12
+teklia-toolbox==0.1.13
 zstandard==0.25.0
 
 [tests]

{arkindex_base_worker-0.5.2a1 → arkindex_base_worker-0.5.2b1}/arkindex_worker/image.py

@@ -38,7 +38,7 @@ if TYPE_CHECKING:
     from arkindex_worker.models import Element
 
 # See http://docs.python-requests.org/en/master/user/advanced/#timeouts
-DOWNLOAD_TIMEOUT = (30, 60)
+REQUEST_TIMEOUT = (30, 60)
 
 BoundingBox = namedtuple("BoundingBox", ["x", "y", "width", "height"])
 
@@ -346,7 +346,7 @@ def _retried_request(url, *args, method=requests.get, **kwargs):
         url,
         *args,
         headers={"User-Agent": IIIF_USER_AGENT},
-        timeout=DOWNLOAD_TIMEOUT,
+        timeout=REQUEST_TIMEOUT,
         verify=should_verify_cert(url),
         **kwargs,
     )

{arkindex_base_worker-0.5.2a1 → arkindex_base_worker-0.5.2b1}/arkindex_worker/utils.py

@@ -163,12 +163,12 @@ def zstd_compress(
 
 def create_tar_archive(
     path: Path, destination: Path | None = None
-) -> tuple[int | None, Path, str]:
+) -> tuple[int | None, Path]:
     """Create a tar archive using the content at specified location.
 
     :param path: Path to the file to archive
     :param destination: Optional path for the created TAR archive. A tempfile will be created if this is omitted.
-    :return: The file descriptor (if one was created) and path to the TAR archive, hash of its content.
+    :return: The file descriptor (if one was created) and path to the TAR archive.
     """
     # Parse destination and create a tmpfile if none was specified
     file_d, destination = (
@@ -204,26 +204,26 @@ def create_tar_archive(
         with file_path.open("rb") as file_data:
             for chunk in iter(lambda: file_data.read(CHUNK_SIZE), b""):
                 content_hasher.update(chunk)
-    return file_d, destination, content_hasher.hexdigest()
+    return file_d, destination
 
 
 def create_tar_zst_archive(
     source: Path, destination: Path | None = None
-) -> tuple[int | None, Path, str, str]:
+) -> tuple[int | None, Path, str]:
     """Helper to create a TAR+ZST archive from a source folder.
 
     :param source: Path to the folder whose content should be archived.
     :param destination: Path to the created archive, defaults to None. If unspecified, a temporary file will be created.
-    :return: The file descriptor of the created tempfile (if one was created), path to the archive, its hash and the hash of the tar archive's content.
+    :return: The file descriptor of the created tempfile (if one was created), path to the archive and its hash.
     """
     # Create tar archive
-    tar_fd, tar_archive, tar_hash = create_tar_archive(source)
+    tar_fd, tar_archive = create_tar_archive(source)
 
     zst_fd, zst_archive, zst_hash = zstd_compress(tar_archive, destination)
 
     close_delete_file(tar_fd, tar_archive)
 
-    return zst_fd, zst_archive, zst_hash, tar_hash
+    return zst_fd, zst_archive, zst_hash
 
 
 def create_zip_archive(source: Path, destination: Path | None = None) -> Path:

{arkindex_base_worker-0.5.2a1 → arkindex_base_worker-0.5.2b1}/arkindex_worker/worker/__init__.py

@@ -424,12 +424,13 @@ class DatasetWorker(DatasetMixin, BaseWorker, TaskMixin):
         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)
+                if dataset_set.dataset.state == DatasetState.Complete.value:
+                    logger.info(f"Retrieving data for {dataset_set} ({i}/{count})")
+                    self.download_dataset_artifact(dataset_set.dataset)
+                else:
+                    logger.warning(
+                        f"The dataset {dataset_set.dataset} has its state set to `{dataset_set.dataset.state}`, its archive will not be downloaded"
+                    )
 
                 logger.info(f"Processing {dataset_set} ({i}/{count})")
                 self.process_set(dataset_set)
@@ -444,7 +445,7 @@ class DatasetWorker(DatasetMixin, BaseWorker, TaskMixin):
 
                 logger.warning(message, exc_info=e if self.args.verbose else None)
 
-        # Cleanup the latest downloaded dataset artifact
+        # Cleanup the latest downloaded dataset artifact (if needed)
         self.cleanup_downloaded_artifact()
 
         message = f"Ran on {count} {pluralize('set', count)}: {count - failed} completed, {failed} failed"

{arkindex_base_worker-0.5.2a1 → arkindex_base_worker-0.5.2b1}/arkindex_worker/worker/base.py

@@ -15,7 +15,7 @@ import gnupg
 import yaml
 
 from arkindex import options_from_env
-from arkindex.exceptions import ClientError, ErrorResponse
+from arkindex.exceptions import ErrorResponse
 from arkindex_worker import logger
 from arkindex_worker.cache import (
     check_version,
@@ -261,6 +261,10 @@ class BaseWorker:
 
         logger.info(f"Loaded {worker_run['summary']} from API")
 
+        # The `RetrieveSecret` endpoint is only available in Arkindex EE.
+        # In CE, the values of `secret` fields should be used directly without calling `RetrieveSecret`.
+        can_retrieve_secret = "RetrieveSecret" in self.api_client.document.links
+
         def _process_config_item(item: dict) -> tuple[str, Any]:
             if not item["secret"]:
                 return (item["key"], item["value"])
@@ -270,16 +274,12 @@ class BaseWorker:
                 logger.info(f"Optional secret `{item['key']}` is not set")
                 return (item["key"], None)
 
-            # Load secret, only available in Arkindex EE
-            try:
-                secret = self.load_secret(Path(item["value"]))
-            except ClientError as e:
-                logger.error(
-                    f"Failed to retrieve the secret {item['value']}, probably an Arkindex Community Edition: {e}"
-                )
-                return (item["key"], None)
+            value = item["value"]
+            # Load secret when `RetrieveSecret` is available
+            if can_retrieve_secret:
+                value = self.load_secret(Path(item["value"]))
 
-            return (item["key"], secret)
+            return (item["key"], value)
 
         # Load model version configuration when available
         # Workers will use model version ID and details to download the model

arkindex_base_worker-0.5.2b1/arkindex_worker/worker/task.py

@@ -0,0 +1,100 @@
+"""
+BaseWorker methods for tasks.
+"""
+
+import uuid
+from collections.abc import Iterator
+from http.client import REQUEST_TIMEOUT
+from pathlib import Path
+
+import magic
+import requests
+
+from arkindex.compat import DownloadedFile
+from arkindex_worker import logger
+from arkindex_worker.models import Artifact
+from teklia_toolbox.requests import should_verify_cert
+
+
+class TaskMixin:
+    def list_artifacts(self, task_id: uuid.UUID) -> Iterator[Artifact]:
+        """
+        List artifacts associated to a task.
+
+        :param task_id: Task ID to find artifacts from.
+        :returns: An iterator of ``Artifact`` objects built from the ``ListArtifacts`` API endpoint.
+        """
+        assert task_id and isinstance(task_id, uuid.UUID), (
+            "task_id shouldn't be null and should be an UUID"
+        )
+
+        results = self.api_client.request("ListArtifacts", id=task_id)
+
+        return map(Artifact, results)
+
+    def download_artifact(
+        self, task_id: uuid.UUID, artifact: Artifact
+    ) -> DownloadedFile:
+        """
+        Download an artifact content.
+
+        :param task_id: Task ID the Artifact is from.
+        :param artifact: Artifact to download content from.
+        :returns: A temporary file containing the ``Artifact`` downloaded from the ``DownloadArtifact`` API endpoint.
+        """
+        assert task_id and isinstance(task_id, uuid.UUID), (
+            "task_id shouldn't be null and should be an UUID"
+        )
+        assert artifact and isinstance(artifact, Artifact), (
+            "artifact shouldn't be null and should be an Artifact"
+        )
+
+        return self.api_client.request(
+            "DownloadArtifact", id=task_id, path=artifact.path
+        )
+
+    def upload_artifact(self, path: Path) -> None:
+        """
+        Upload a single file as an Artifact of the current task.
+
+        :param path: Path of the single file to upload as an Artifact.
+        """
+        assert path and isinstance(path, Path) and path.exists(), (
+            "path shouldn't be null, should be a Path and should exist"
+        )
+
+        if self.is_read_only:
+            logger.warning("Cannot upload artifact as this worker is in read-only mode")
+            return
+
+        # Get path relative to task's data directory
+        relpath = str(path.relative_to(self.work_dir))
+
+        # Get file size
+        size = path.stat().st_size
+
+        # Detect content type
+        try:
+            content_type = magic.from_file(path, mime=True)
+        except Exception as e:
+            logger.warning(f"Failed to get a mime type for {path}: {e}")
+            content_type = "application/octet-stream"
+
+        # Create artifact on API to get an S3 url
+        artifact = self.api_client.request(
+            "CreateArtifact",
+            id=self.task_id,
+            body={"path": relpath, "content_type": content_type, "size": size},
+        )
+
+        # Upload the file content to S3
+        s3_put_url = artifact["s3_put_url"]
+        with path.open("rb") as content:
+            resp = requests.put(
+                s3_put_url,
+                data=content,
+                headers={"Content-Type": content_type},
+                timeout=REQUEST_TIMEOUT,
+                verify=should_verify_cert(s3_put_url),
+            )
+            resp.raise_for_status()

{arkindex_base_worker-0.5.2a1 → arkindex_base_worker-0.5.2b1}/arkindex_worker/worker/training.py

@@ -3,16 +3,15 @@ BaseWorker methods for training.
 """
 
 import functools
+from collections.abc import Generator
 from contextlib import contextmanager
 from pathlib import Path
 from typing import NewType
 from uuid import UUID
 
-import requests
-
-from arkindex.exceptions import ErrorResponse
 from arkindex_worker import logger
 from arkindex_worker.utils import close_delete_file, create_tar_zst_archive
+from teklia_toolbox.uploads import MultipartUpload
 
 DirPath = NewType("DirPath", Path)
 """Path to a directory"""
@@ -25,23 +24,21 @@ FileSize = NewType("FileSize", int)
 
 
 @contextmanager
-def create_archive(path: DirPath) -> tuple[Path, Hash, FileSize, Hash]:
+def create_archive(path: DirPath) -> Generator[tuple[Path, FileSize, Hash]]:
     """
     Create a tar archive from the files at the given location then compress it to a zst archive.
 
-    Yield its location, its hash, its size and its content's hash.
+    Yield its location, its size and its hash.
 
     :param path: Create a compressed tar archive from the files
-    :returns: The location of the created archive, its hash, its size and its content's hash
+    :returns: The location of the created archive, its size and its hash
     """
     assert path.is_dir(), "create_archive needs a directory"
 
-    zst_descriptor, zst_archive, archive_hash, content_hash = create_tar_zst_archive(
-        path
-    )
+    zst_descriptor, zst_archive, archive_hash = create_tar_zst_archive(path)
 
     # Get content hash, archive size and hash
-    yield zst_archive, content_hash, zst_archive.stat().st_size, archive_hash
+    yield zst_archive, zst_archive.stat().st_size, archive_hash
 
     # Remove the zst archive
     close_delete_file(zst_descriptor, zst_archive)
@@ -112,62 +109,48 @@ class TrainingMixin:
         """
 
         configuration = configuration or {}
-        if not self.model_version:
-            self.create_model_version(
-                model_id=model_id,
-                tag=tag,
-                description=description,
-                configuration=configuration,
-                parent=parent,
-            )
-
-        elif tag or description or configuration or parent:
-            assert self.model_version.get("model_id") == model_id, (
-                "Given `model_id` does not match the current model version"
-            )
-            # If any attribute field has been defined, PATCH the current model version
-            self.update_model_version(
-                tag=tag,
-                description=description,
-                configuration=configuration,
-                parent=parent,
-            )
 
         # Create the zst archive, get its hash and size
-        # Validate the model version
         with create_archive(path=model_path) as (
             path_to_archive,
-            hash,
             size,
-            archive_hash,
+            hash,
         ):
-            # Create a new model version with hash and size
-            self.upload_to_s3(archive_path=path_to_archive)
-
-            current_version_id = self.model_version["id"]
-            # Mark the model as valid
-            self.validate_model_version(
-                size=size,
-                hash=hash,
-                archive_hash=archive_hash,
-            )
-            if self.model_version["id"] != current_version_id and (
-                tag or description or configuration or parent
-            ):
-                logger.warning(
-                    "Updating the existing available model version with the given attributes."
+            # Update an existing model version with hash, size and any other defined attribute
+            if self.model_version:
+                assert self.model_version.get("model_id") == model_id, (
+                    "Given `model_id` does not match the current model version"
                 )
                 self.update_model_version(
+                    size=size,
+                    archive_hash=hash,
+                    tag=tag,
+                    description=description,
+                    configuration=configuration,
+                    parent=parent,
+                )
+
+            # Create a new model version with hash and size
+            else:
+                self.create_model_version(
+                    model_id=model_id,
+                    size=size,
+                    archive_hash=hash,
                     tag=tag,
                     description=description,
                     configuration=configuration,
                     parent=parent,
                 )
 
+            # Upload the archive in multiple parts (supports huge files)
+            self.upload_to_s3(path_to_archive)
+
     @skip_if_read_only
     def create_model_version(
         self,
         model_id: str,
+        size: FileSize,
+        archive_hash: Hash,
         tag: str | None = None,
         description: str | None = None,
         configuration: dict | None = None,
@@ -177,6 +160,8 @@ class TrainingMixin:
         Create a new version of the specified model with its base attributes.
         Once successfully created, the model version is accessible via `self.model_version`.
 
+        :param size: Size of uploaded archive
+        :param hash: MD5 hash of the uploaded archive
         :param tag: Tag of the model version
         :param description: Description of the model version
         :param configuration: Configuration of the model version
@@ -189,6 +174,8 @@ class TrainingMixin:
             "CreateModelVersion",
             id=model_id,
             body=build_clean_payload(
+                size=size,
+                archive_hash=archive_hash,
                 tag=tag,
                 description=description,
                 configuration=configuration,
@@ -197,12 +184,14 @@ class TrainingMixin:
         )
 
         logger.info(
-            f"Model version ({self.model_version['id']}) was successfully created"
+            f"Model version ({self.model_version['id']}) was successfully created."
         )
 
     @skip_if_read_only
     def update_model_version(
         self,
+        size: FileSize,
+        archive_hash: Hash,
         tag: str | None = None,
         description: str | None = None,
         configuration: dict | None = None,
@@ -211,6 +200,8 @@ class TrainingMixin:
         """
         Update the current model version with the given attributes.
 
+        :param size: Size of uploaded archive
+        :param hash: MD5 hash of the uploaded archive
         :param tag: Tag of the model version
         :param description: Description of the model version
         :param configuration: Configuration of the model version
@@ -221,6 +212,8 @@ class TrainingMixin:
             "UpdateModelVersion",
             id=self.model_version["id"],
             body=build_clean_payload(
+                size=size,
+                archive_hash=archive_hash,
                 tag=tag,
                 description=description,
                 configuration=configuration,
@@ -228,93 +221,34 @@ class TrainingMixin:
             ),
         )
         logger.info(
-            f"Model version ({self.model_version['id']}) was successfully updated"
+            f"Model version ({self.model_version['id']}) was successfully updated."
         )
 
     @skip_if_read_only
     def upload_to_s3(self, archive_path: Path) -> None:
         """
-        Upload the archive of the model's files to an Amazon s3 compatible storage
+        Upload the archive of the model's files to an Amazon s3 compatible storage in multiple parts
         """
-
         assert self.model_version, (
             "You must create the model version before uploading an archive."
         )
         assert self.model_version["state"] != "Available", (
-            "The model is already marked as available."
+            "The model version is already marked as available."
         )
 
-        s3_put_url = self.model_version.get("s3_put_url")
-        assert s3_put_url, (
-            "S3 PUT URL is not set, please ensure you have the right to validate a model version."
-        )
-
-        logger.info("Uploading to s3...")
-        # Upload the archive on s3
-        with archive_path.open("rb") as archive:
-            r = requests.put(
-                url=s3_put_url,
-                data=archive,
-                headers={"Content-Type": "application/zstd"},
-            )
-        r.raise_for_status()
-
-    @skip_if_read_only
-    def validate_model_version(
-        self,
-        hash: str,
-        size: int,
-        archive_hash: str,
-    ):
-        """
-        Sets the model version as `Available`, once its archive has been uploaded to S3.
-
-        :param hash: MD5 hash of the files contained in the archive
-        :param size: The size of the uploaded archive
-        :param archive_hash: MD5 hash of the uploaded archive
-        """
-        assert self.model_version, (
-            "You must create the model version and upload its archive before validating it."
+        multipart = MultipartUpload(
+            client=self.api_client,
+            file_path=archive_path,
+            object_type="model_version",
+            object_id=str(self.model_version["id"]),
         )
         try:
-            self.model_version = self.api_client.request(
-                "PartialUpdateModelVersion",
-                id=self.model_version["id"],
-                body={
-                    "state": "available",
-                    "size": size,
-                    "hash": hash,
-                    "archive_hash": archive_hash,
-                },
+            multipart.upload()
+            multipart.complete()
+        except Exception:
+            multipart.abort()
+            raise
+        else:
+            logger.info(
+                f"Model version ({self.model_version['id']}) archive was successfully uploaded and is now available."
             )
-        except ErrorResponse as e:
-            model_version = e.content
-            if not model_version or "id" not in model_version:
-                raise e
-
-            logger.warning(
-                f"An available model version exists with hash {hash}, using it instead of the pending version."
-            )
-            pending_version_id = self.model_version["id"]
-            logger.warning("Removing the pending model version.")
-            try:
-                self.api_client.request("DestroyModelVersion", id=pending_version_id)
-            except ErrorResponse as e:
-                msg = getattr(e, "content", str(e))
-                logger.error(
-                    f"An error occurred removing the pending version {pending_version_id}: {msg}."
-                )
-
-            logger.info("Retrieving the existing model version.")
-            existing_version_id = model_version["id"].pop()
-            try:
-                self.model_version = self.api_client.request(
-                    "RetrieveModelVersion", id=existing_version_id
-                )
-            except ErrorResponse as e:
-                logger.error(
-                    f"An error occurred retrieving the existing version {existing_version_id}: {e.status_code} - {e.content}."
-                )
-                raise
-
-        logger.info(f"Model version {self.model_version['id']} is now available.")

{arkindex_base_worker-0.5.2a1 → arkindex_base_worker-0.5.2b1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "arkindex-base-worker"
-version = "0.5.2a1"
+version = "0.5.2b1"
 description = "Base Worker to easily build Arkindex ML workflows"
 license-files = ["LICENSE"]
 dependencies = [
@@ -12,8 +12,9 @@ dependencies = [
     "peewee~=3.17",
     "Pillow==11.3.0",
     "python-gnupg==0.5.6",
+    "python-magic==0.4.27",
     "shapely==2.0.6",
-    "teklia-toolbox==0.1.12",
+    "teklia-toolbox==0.1.13",
     "zstandard==0.25.0",
 ]
 authors = [