futurehouse-client 0.3.18.dev186__py3-none-any.whl → 0.3.19__py3-none-any.whl

futurehouse_client/clients/rest_client.py

@@ -13,9 +13,10 @@ import tempfile
 import time
 import uuid
 from collections.abc import Collection
+from concurrent.futures import ThreadPoolExecutor, as_completed
 from pathlib import Path
 from types import ModuleType
-from typing import Any, ClassVar, assert_never, cast
+from typing import Any, ClassVar, cast
 from uuid import UUID
 
 import cloudpickle
@@ -31,6 +32,7 @@ from httpx import (
     ReadError,
     ReadTimeout,
     RemoteProtocolError,
+    codes,
 )
 from ldp.agent import AgentConfig
 from requests.exceptions import RequestException, Timeout
@@ -45,21 +47,15 @@ from tqdm.asyncio import tqdm
 
 from futurehouse_client.clients import JobNames
 from futurehouse_client.models.app import (
-    APIKeyPayload,
     AuthType,
     JobDeploymentConfig,
-    PQATaskResponse,
     Stage,
     TaskRequest,
     TaskResponse,
     TaskResponseVerbose,
 )
 from futurehouse_client.models.rest import ExecutionStatus
-from futurehouse_client.utils.auth import (
-    AUTH_ERRORS_TO_RETRY_ON,
-    AuthError,
-    refresh_token_on_auth_error,
-)
+from futurehouse_client.utils.auth import RefreshingJWT
 from futurehouse_client.utils.general import gather_with_concurrency
 from futurehouse_client.utils.module_utils import (
     OrganizationSelector,
@@ -128,8 +124,6 @@ retry_if_connection_error = retry_if_exception_type((
     FileUploadError,
 ))
 
-# 5 minute default for JWTs
-JWT_TOKEN_CACHE_EXPIRY: int = 300  # seconds
 DEFAULT_AGENT_TIMEOUT: int = 2400  # seconds
 
 
@@ -140,6 +134,9 @@ class RestClient:
     MAX_RETRY_WAIT: ClassVar[int] = 10
     DEFAULT_POLLING_TIME: ClassVar[int] = 5  # seconds
     CHUNK_SIZE: ClassVar[int] = 16 * 1024 * 1024  # 16MB chunks
+    ASSEMBLY_POLLING_INTERVAL: ClassVar[int] = 10  # seconds
+    MAX_ASSEMBLY_WAIT_TIME: ClassVar[int] = 1800  # 30 minutes
+    MAX_CONCURRENT_CHUNKS: ClassVar[int] = 12  # Maximum concurrent chunk uploads
 
     def __init__(
         self,
@@ -163,69 +160,87 @@ class RestClient:
         self.api_key = api_key
         self._clients: dict[str, Client | AsyncClient] = {}
         self.headers = headers or {}
-        self.auth_jwt = self._run_auth(jwt=jwt)
+        self.jwt = jwt
         self.organizations: list[str] = self._filter_orgs(organization)
 
     @property
     def client(self) -> Client:
-        """Lazily initialized and cached HTTP client with authentication."""
-        return cast(Client, self.get_client("application/json", with_auth=True))
+        """Authenticated HTTP client for regular API calls."""
+        return cast(Client, self.get_client("application/json", authenticated=True))
 
     @property
     def async_client(self) -> AsyncClient:
-        """Lazily initialized and cached HTTP client with authentication."""
+        """Authenticated async HTTP client for regular API calls."""
         return cast(
             AsyncClient,
-            self.get_client("application/json", with_auth=True, with_async=True),
+            self.get_client("application/json", authenticated=True, async_client=True),
         )
 
     @property
-    def auth_client(self) -> Client:
-        """Lazily initialized and cached HTTP client without authentication."""
-        return cast(Client, self.get_client("application/json", with_auth=False))
+    def unauthenticated_client(self) -> Client:
+        """Unauthenticated HTTP client for auth operations."""
+        return cast(Client, self.get_client("application/json", authenticated=False))
 
     @property
     def multipart_client(self) -> Client:
-        """Lazily initialized and cached HTTP client for multipart uploads."""
-        return cast(Client, self.get_client(None, with_auth=True))
+        """Authenticated HTTP client for multipart uploads."""
+        return cast(Client, self.get_client(None, authenticated=True))
 
     def get_client(
         self,
         content_type: str | None = "application/json",
-        with_auth: bool = True,
-        with_async: bool = False,
+        authenticated: bool = True,
+        async_client: bool = False,
     ) -> Client | AsyncClient:
         """Return a cached HTTP client or create one if needed.
 
         Args:
             content_type: The desired content type header. Use None for multipart uploads.
-            with_auth: Whether the client should include an Authorization header.
-            with_async: Whether to use an async client.
+            authenticated: Whether the client should include authentication.
+            async_client: Whether to use an async client.
 
         Returns:
             An HTTP client configured with the appropriate headers.
         """
-        # Create a composite key based on content type and auth flag.
-        key = f"{content_type or 'multipart'}_{with_auth}_{with_async}"
+        # Create a composite key based on content type and auth flag
+        key = f"{content_type or 'multipart'}_{authenticated}_{async_client}"
+
         if key not in self._clients:
             headers = copy.deepcopy(self.headers)
-            if with_auth:
-                headers["Authorization"] = f"Bearer {self.auth_jwt}"
+            auth = None
+
+            if authenticated:
+                auth = RefreshingJWT(
+                    # authenticated=False will always return a synchronous client
+                    auth_client=cast(
+                        Client, self.get_client("application/json", authenticated=False)
+                    ),
+                    auth_type=self.auth_type,
+                    api_key=self.api_key,
+                    jwt=self.jwt,
+                )
+
             if content_type:
                 headers["Content-Type"] = content_type
+
+            headers["x-client"] = "sdk"
+
             self._clients[key] = (
                 AsyncClient(
                     base_url=self.base_url,
                     headers=headers,
                     timeout=self.REQUEST_TIMEOUT,
+                    auth=auth,
                 )
-                if with_async
+                if async_client
                 else Client(
                     base_url=self.base_url,
                     headers=headers,
                     timeout=self.REQUEST_TIMEOUT,
+                    auth=auth,
                 )
             )
+
         return self._clients[key]
 
     def close(self):
@@ -255,32 +270,6 @@ class RestClient:
             raise ValueError(f"Organization '{organization}' not found.")
         return filtered_orgs
 
-    def _run_auth(self, jwt: str | None = None) -> str:
-        auth_payload: APIKeyPayload | None
-        if self.auth_type == AuthType.API_KEY:
-            auth_payload = APIKeyPayload(api_key=self.api_key)
-        elif self.auth_type == AuthType.JWT:
-            auth_payload = None
-        else:
-            assert_never(self.auth_type)
-        try:
-            # Use the unauthenticated client for login
-            if auth_payload:
-                response = self.auth_client.post(
-                    "/auth/login", json=auth_payload.model_dump()
-                )
-                response.raise_for_status()
-                token_data = response.json()
-            elif jwt:
-                token_data = {"access_token": jwt, "expires_in": JWT_TOKEN_CACHE_EXPIRY}
-            else:
-                raise ValueError("JWT token required for JWT authentication.")
-
-            return token_data["access_token"]
-        except Exception as e:
-            raise RestClientError(f"Error authenticating: {e!s}") from e
-
-    @refresh_token_on_auth_error()
     def _check_job(self, name: str, organization: str) -> dict[str, Any]:
         try:
             response = self.client.get(
@@ -288,25 +277,113 @@ class RestClient:
             )
             response.raise_for_status()
             return response.json()
-        except HTTPStatusError as e:
-            if e.response.status_code in AUTH_ERRORS_TO_RETRY_ON:
-                raise AuthError(
-                    e.response.status_code,
-                    f"Authentication failed: {e}",
-                    request=e.request,
-                    response=e.response,
-                ) from e
-            raise
         except Exception as e:
             raise JobFetchError(f"Error checking job: {e!s}") from e
 
-    @refresh_token_on_auth_error()
     def _fetch_my_orgs(self) -> list[str]:
         response = self.client.get(f"/v0.1/organizations?filter={True}")
         response.raise_for_status()
         orgs = response.json()
         return [org["name"] for org in orgs]
 
+    def _check_assembly_status(
+        self, job_name: str, upload_id: str, file_name: str
+    ) -> dict[str, Any]:
+        """Check the assembly status of an uploaded file.
+
+        Args:
+            job_name: The name of the futurehouse job
+            upload_id: The upload ID
+            file_name: The name of the file
+
+        Returns:
+            Dict containing status information
+
+        Raises:
+            RestClientError: If there's an error checking status
+        """
+        try:
+            url = f"/v0.1/crows/{job_name}/assembly-status/{upload_id}/{file_name}"
+            response = self.client.get(url)
+            response.raise_for_status()
+            return response.json()
+        except Exception as e:
+            raise RestClientError(f"Error checking assembly status: {e}") from e
+
+    def _wait_for_all_assemblies_completion(
+        self,
+        job_name: str,
+        upload_id: str,
+        file_names: list[str],
+        timeout: int = MAX_ASSEMBLY_WAIT_TIME,
+    ) -> bool:
+        """Wait for all file assemblies to complete.
+
+        Args:
+            job_name: The name of the futurehouse job
+            upload_id: The upload ID
+            file_names: List of file names to wait for
+            timeout: Maximum time to wait in seconds
+
+        Returns:
+            True if all assemblies succeeded, False if any failed or timed out
+
+        Raises:
+            RestClientError: If any assembly fails
+        """
+        if not file_names:
+            return True
+
+        start_time = time.time()
+        logger.info(f"Waiting for assembly of {len(file_names)} file(s) to complete...")
+
+        completed_files: set[str] = set()
+
+        while (time.time() - start_time) < timeout and len(completed_files) < len(
+            file_names
+        ):
+            for file_name in file_names:
+                if file_name in completed_files:
+                    continue
+
+                try:
+                    status_data = self._check_assembly_status(
+                        job_name, upload_id, file_name
+                    )
+                    status = status_data.get("status")
+
+                    if status == ExecutionStatus.SUCCESS.value:
+                        logger.info(f"Assembly completed for {file_name}")
+                        completed_files.add(file_name)
+                    elif status == ExecutionStatus.FAIL.value:
+                        error_msg = status_data.get("error", "Unknown assembly error")
+                        raise RestClientError(
+                            f"Assembly failed for {file_name}: {error_msg}"
+                        )
+                    elif status == ExecutionStatus.IN_PROGRESS.value:
+                        logger.debug(f"Assembly in progress for {file_name}...")
+
+                except RestClientError:
+                    raise  # Re-raise assembly errors
+                except Exception as e:
+                    logger.warning(
+                        f"Error checking assembly status for {file_name}: {e}"
+                    )
+
+            # Don't sleep if all files are complete
+            if len(completed_files) < len(file_names):
+                time.sleep(self.ASSEMBLY_POLLING_INTERVAL)
+
+        if len(completed_files) < len(file_names):
+            remaining_files = set(file_names) - completed_files
+            logger.warning(
+                f"Assembly timeout for files: {remaining_files} after {timeout} seconds"
+            )
+            return False
+
+        logger.info(f"All {len(file_names)} file assemblies completed successfully")
+        return True
+
     @staticmethod
     def _validate_module_path(path: Path) -> None:
         """Validates that the given path exists and is a directory.
@@ -358,7 +435,6 @@ class RestClient:
         if not files:
             raise TaskFetchError(f"No files found in {path}")
 
-    @refresh_token_on_auth_error()
     @retry(
         stop=stop_after_attempt(MAX_RETRY_ATTEMPTS),
         wait=wait_exponential(multiplier=RETRY_MULTIPLIER, max=MAX_RETRY_WAIT),
@@ -368,51 +444,37 @@ class RestClient:
         self, task_id: str | None = None, history: bool = False, verbose: bool = False
     ) -> "TaskResponse":
         """Get details for a specific task."""
-        try:
-            task_id = task_id or self.trajectory_id
-            url = f"/v0.1/trajectories/{task_id}"
-            full_url = f"{self.base_url}{url}"
-
-            with (
-                external_trace(
-                    url=full_url,
-                    method="GET",
-                    library="httpx",
-                    custom_params={
-                        "operation": "get_job",
-                        "job_id": task_id,
-                    },
-                ),
-                self.client.stream("GET", url, params={"history": history}) as response,
-            ):
-                response.raise_for_status()
-                json_data = "".join(response.iter_text(chunk_size=1024))
-                data = json.loads(json_data)
-                if "id" not in data:
-                    data["id"] = task_id
-                verbose_response = TaskResponseVerbose(**data)
+        task_id = task_id or self.trajectory_id
+        url = f"/v0.1/trajectories/{task_id}"
+        full_url = f"{self.base_url}{url}"
 
-            if verbose:
-                return verbose_response
-            if any(
-                JobNames.from_string(job_name) in verbose_response.job_name
-                for job_name in ["crow", "falcon", "owl", "dummy"]
-            ):
-                return PQATaskResponse(**data)
-            return TaskResponse(**data)
-        except HTTPStatusError as e:
-            if e.response.status_code in AUTH_ERRORS_TO_RETRY_ON:
-                raise AuthError(
-                    e.response.status_code,
-                    f"Authentication failed: {e}",
-                    request=e.request,
-                    response=e.response,
-                ) from e
-            raise
-        except Exception as e:
-            raise TaskFetchError(f"Error getting task: {e!s}") from e
+        with (
+            external_trace(
+                url=full_url,
+                method="GET",
+                library="httpx",
+                custom_params={
+                    "operation": "get_job",
+                    "job_id": task_id,
+                },
+            ),
+            self.client.stream("GET", url, params={"history": history}) as response,
+        ):
+            if response.status_code in {401, 403}:
+                raise PermissionError(
+                    f"Error getting task: Permission denied for task {task_id}"
+                )
+            response.raise_for_status()
+            json_data = "".join(response.iter_text(chunk_size=1024))
+            data = json.loads(json_data)
+            if "id" not in data:
+                data["id"] = task_id
+            verbose_response = TaskResponseVerbose(**data)
+
+        if verbose:
+            return verbose_response
+        return JobNames.get_response_object_from_job(verbose_response.job_name)(**data)
 
-    @refresh_token_on_auth_error()
     @retry(
         stop=stop_after_attempt(MAX_RETRY_ATTEMPTS),
         wait=wait_exponential(multiplier=RETRY_MULTIPLIER, max=MAX_RETRY_WAIT),
@@ -422,53 +484,37 @@ class RestClient:
         self, task_id: str | None = None, history: bool = False, verbose: bool = False
     ) -> "TaskResponse":
         """Get details for a specific task asynchronously."""
-        try:
-            task_id = task_id or self.trajectory_id
-            url = f"/v0.1/trajectories/{task_id}"
-            full_url = f"{self.base_url}{url}"
+        task_id = task_id or self.trajectory_id
+        url = f"/v0.1/trajectories/{task_id}"
+        full_url = f"{self.base_url}{url}"
+
+        with external_trace(
+            url=full_url,
+            method="GET",
+            library="httpx",
+            custom_params={
+                "operation": "get_job",
+                "job_id": task_id,
+            },
+        ):
+            async with self.async_client.stream(
+                "GET", url, params={"history": history}
+            ) as response:
+                if response.status_code in {401, 403}:
+                    raise PermissionError(
+                        f"Error getting task: Permission denied for task {task_id}"
+                    )
+                response.raise_for_status()
+                json_data = "".join([chunk async for chunk in response.aiter_text()])
+                data = json.loads(json_data)
+                if "id" not in data:
+                    data["id"] = task_id
+                verbose_response = TaskResponseVerbose(**data)
 
-            with external_trace(
-                url=full_url,
-                method="GET",
-                library="httpx",
-                custom_params={
-                    "operation": "get_job",
-                    "job_id": task_id,
-                },
-            ):
-                async with self.async_client.stream(
-                    "GET", url, params={"history": history}
-                ) as response:
-                    response.raise_for_status()
-                    json_data = "".join([
-                        chunk async for chunk in response.aiter_text()
-                    ])
-                    data = json.loads(json_data)
-                    if "id" not in data:
-                        data["id"] = task_id
-                    verbose_response = TaskResponseVerbose(**data)
-
-            if verbose:
-                return verbose_response
-            if any(
-                JobNames.from_string(job_name) in verbose_response.job_name
-                for job_name in ["crow", "falcon", "owl", "dummy"]
-            ):
-                return PQATaskResponse(**data)
-            return TaskResponse(**data)
-        except HTTPStatusError as e:
-            if e.response.status_code in AUTH_ERRORS_TO_RETRY_ON:
-                raise AuthError(
-                    e.response.status_code,
-                    f"Authentication failed: {e}",
-                    request=e.request,
-                    response=e.response,
-                ) from e
-            raise
-        except Exception as e:
-            raise TaskFetchError(f"Error getting task: {e!s}") from e
+        if verbose:
+            return verbose_response
+        return JobNames.get_response_object_from_job(verbose_response.job_name)(**data)
 
-    @refresh_token_on_auth_error()
     @retry(
         stop=stop_after_attempt(MAX_RETRY_ATTEMPTS),
         wait=wait_exponential(multiplier=RETRY_MULTIPLIER, max=MAX_RETRY_WAIT),
@@ -485,27 +531,18 @@ class RestClient:
                 self.stage,
             )
 
-        try:
-            response = self.client.post(
-                "/v0.1/crows", json=task_data.model_dump(mode="json")
+        response = self.client.post(
+            "/v0.1/crows", json=task_data.model_dump(mode="json")
+        )
+        if response.status_code in {401, 403}:
+            raise PermissionError(
+                f"Error creating task: Permission denied for task {task_data.name}"
             )
-            response.raise_for_status()
-            trajectory_id = response.json()["trajectory_id"]
-            self.trajectory_id = trajectory_id
-        except HTTPStatusError as e:
-            if e.response.status_code in AUTH_ERRORS_TO_RETRY_ON:
-                raise AuthError(
-                    e.response.status_code,
-                    f"Authentication failed: {e}",
-                    request=e.request,
-                    response=e.response,
-                ) from e
-            raise
-        except Exception as e:
-            raise TaskFetchError(f"Error creating task: {e!s}") from e
+        response.raise_for_status()
+        trajectory_id = response.json()["trajectory_id"]
+        self.trajectory_id = trajectory_id
         return trajectory_id
 
-    @refresh_token_on_auth_error()
     @retry(
         stop=stop_after_attempt(MAX_RETRY_ATTEMPTS),
         wait=wait_exponential(multiplier=RETRY_MULTIPLIER, max=MAX_RETRY_WAIT),
@@ -521,25 +558,16 @@ class RestClient:
                 task_data.name.name,
                 self.stage,
             )
-
-        try:
-            response = await self.async_client.post(
-                "/v0.1/crows", json=task_data.model_dump(mode="json")
+        response = await self.async_client.post(
+            "/v0.1/crows", json=task_data.model_dump(mode="json")
+        )
+        if response.status_code in {401, 403}:
+            raise PermissionError(
+                f"Error creating task: Permission denied for task {task_data.name}"
             )
-            response.raise_for_status()
-            trajectory_id = response.json()["trajectory_id"]
-            self.trajectory_id = trajectory_id
-        except HTTPStatusError as e:
-            if e.response.status_code in AUTH_ERRORS_TO_RETRY_ON:
-                raise AuthError(
-                    e.response.status_code,
-                    f"Authentication failed: {e}",
-                    request=e.request,
-                    response=e.response,
-                ) from e
-            raise
-        except Exception as e:
-            raise TaskFetchError(f"Error creating task: {e!s}") from e
+        response.raise_for_status()
+        trajectory_id = response.json()["trajectory_id"]
+        self.trajectory_id = trajectory_id
         return trajectory_id
 
     async def arun_tasks_until_done(
@@ -683,7 +711,6 @@ class RestClient:
             for task_id in trajectory_ids
         ]
 
-    @refresh_token_on_auth_error()
     @retry(
         stop=stop_after_attempt(MAX_RETRY_ATTEMPTS),
         wait=wait_exponential(multiplier=RETRY_MULTIPLIER, max=MAX_RETRY_WAIT),
@@ -695,19 +722,11 @@ class RestClient:
             build_id = build_id or self.build_id
             response = self.client.get(f"/v0.1/builds/{build_id}")
             response.raise_for_status()
-        except HTTPStatusError as e:
-            if e.response.status_code in AUTH_ERRORS_TO_RETRY_ON:
-                raise AuthError(
-                    e.response.status_code,
-                    f"Authentication failed: {e}",
-                    request=e.request,
-                    response=e.response,
-                ) from e
-            raise
+        except Exception as e:
+            raise JobFetchError(f"Error getting build status: {e!s}") from e
         return response.json()
 
     # TODO: Refactor later so we don't have to ignore PLR0915
-    @refresh_token_on_auth_error()
     @retry(
         stop=stop_after_attempt(MAX_RETRY_ATTEMPTS),
         wait=wait_exponential(multiplier=RETRY_MULTIPLIER, max=MAX_RETRY_WAIT),
@@ -887,13 +906,6 @@ class RestClient:
                 build_context = response.json()
                 self.build_id = build_context["build_id"]
             except HTTPStatusError as e:
-                if e.response.status_code in AUTH_ERRORS_TO_RETRY_ON:
-                    raise AuthError(
-                        e.response.status_code,
-                        f"Authentication failed: {e}",
-                        request=e.request,
-                        response=e.response,
-                    ) from e
                 error_detail = response.json()
                 error_message = error_detail.get("detail", str(e))
                 raise JobCreationError(
@@ -903,6 +915,8 @@ class RestClient:
             raise JobCreationError(f"Error generating docker image: {e!s}") from e
         return build_context
 
+    # TODO: we should have have an async upload_file, check_assembly_status,
+    # wait_for_assembly_completion, upload_directory, upload_single_file
     @retry(
         stop=stop_after_attempt(MAX_RETRY_ATTEMPTS),
         wait=wait_exponential(multiplier=RETRY_MULTIPLIER, max=MAX_RETRY_WAIT),
@@ -913,6 +927,8 @@ class RestClient:
         job_name: str,
         file_path: str | os.PathLike,
         upload_id: str | None = None,
+        wait_for_assembly: bool = True,
+        assembly_timeout: int = MAX_ASSEMBLY_WAIT_TIME,
     ) -> str:
         """Upload a file or directory to a futurehouse job bucket.
 
@@ -920,29 +936,47 @@ class RestClient:
             job_name: The name of the futurehouse job to upload to.
             file_path: The local path to the file or directory to upload.
             upload_id: Optional folder name to use for the upload. If not provided, a random UUID will be used.
+            wait_for_assembly: After file chunking, wait for the assembly to be processed.
+            assembly_timeout: Maximum time to wait for assembly in seconds.
 
         Returns:
             The upload ID used for the upload.
 
         Raises:
             FileUploadError: If there's an error uploading the file.
+            RestClientError: If assembly fails or times out.
         """
         file_path = Path(file_path)
         if not file_path.exists():
             raise FileNotFoundError(f"File or directory not found: {file_path}")
 
         upload_id = upload_id or str(uuid.uuid4())
+        uploaded_files: list[str] = []
 
         if file_path.is_dir():
             # Process directory recursively
-            self._upload_directory(job_name, file_path, upload_id)
+            uploaded_files = self._upload_directory(job_name, file_path, upload_id)
         else:
             # Process single file
             self._upload_single_file(job_name, file_path, upload_id)
+            uploaded_files = [file_path.name]
+
+        # Wait for all assemblies if requested and we have files
+        if wait_for_assembly and uploaded_files:
+            success = self._wait_for_all_assemblies_completion(
+                job_name, upload_id, uploaded_files, assembly_timeout
+            )
+            if not success:
+                raise RestClientError(
+                    f"Assembly failed or timed out for one or more files: {uploaded_files}"
+                )
+
         logger.info(f"Successfully uploaded {file_path} to {upload_id}")
         return upload_id
 
-    def _upload_directory(self, job_name: str, dir_path: Path, upload_id: str) -> None:
+    def _upload_directory(
+        self, job_name: str, dir_path: Path, upload_id: str
+    ) -> list[str]:
         """Upload all files in a directory recursively.
 
         Args:
@@ -950,12 +984,17 @@ class RestClient:
             dir_path: The path to the directory to upload.
             upload_id: The upload ID to use.
 
+        Returns:
+            List of uploaded file names.
+
         Raises:
             FileUploadError: If there's an error uploading any file.
         """
         # Skip common directories that shouldn't be uploaded
         if any(ignore in dir_path.parts for ignore in FILE_UPLOAD_IGNORE_PARTS):
-            return
+            return []
+
+        uploaded_files: list[str] = []
 
         try:
             # Upload all files in the directory recursively
@@ -965,24 +1004,27 @@ class RestClient:
                 ):
                     # Use path relative to the original directory as file name
                     rel_path = path.relative_to(dir_path)
+                    file_name = str(rel_path)
                     self._upload_single_file(
                         job_name,
                         path,
                         upload_id,
-                        file_name=str(rel_path),
+                        file_name=file_name,
                     )
+                    uploaded_files.append(file_name)
         except Exception as e:
             raise FileUploadError(f"Error uploading directory {dir_path}: {e}") from e
 
-    @refresh_token_on_auth_error()
+        return uploaded_files
+
     def _upload_single_file(
         self,
         job_name: str,
         file_path: Path,
         upload_id: str,
         file_name: str | None = None,
-    ) -> None:
-        """Upload a single file in chunks.
+    ) -> str | None:
+        """Upload a single file in chunks using parallel uploads.
 
         Args:
             job_name: The key of the crow to upload to.
@@ -990,6 +1032,9 @@ class RestClient:
             upload_id: The upload ID to use.
             file_name: Optional name to use for the file. If not provided, the file's name will be used.
 
+        Returns:
+            The status URL if this was the last chunk, None otherwise.
+
         Raises:
             FileUploadError: If there's an error uploading the file.
         """
@@ -999,17 +1044,103 @@ class RestClient:
         # Skip empty files
         if file_size == 0:
             logger.warning(f"Skipping upload of empty file: {file_path}")
-            return
+            return None
 
         total_chunks = (file_size + self.CHUNK_SIZE - 1) // self.CHUNK_SIZE
 
         logger.info(f"Uploading {file_path} as {file_name} ({total_chunks} chunks)")
 
+        status_url = None
+
         try:
-            with open(file_path, "rb") as f:
-                for chunk_index in range(total_chunks):
-                    # Read the chunk from the file
-                    f.seek(chunk_index * self.CHUNK_SIZE)
+            status_url = self._upload_chunks_parallel(
+                job_name,
+                file_path,
+                file_name,
+                upload_id,
+                total_chunks,
+            )
+
+            logger.info(f"Successfully uploaded {file_name}")
+        except Exception as e:
+            logger.exception(f"Error uploading file {file_path}")
+            raise FileUploadError(f"Error uploading file {file_path}: {e}") from e
+        return status_url
+
+    def _upload_chunks_parallel(
+        self,
+        job_name: str,
+        file_path: Path,
+        file_name: str,
+        upload_id: str,
+        total_chunks: int,
+    ) -> str | None:
+        """Upload all chunks in parallel batches, including the final chunk.
+
+        Args:
+            job_name: The key of the crow to upload to.
+            file_path: The path to the file to upload.
+            file_name: The name to use for the file.
+            upload_id: The upload ID to use.
+            total_chunks: Total number of chunks.
+
+        Returns:
+            The status URL from the final chunk response, or None if no chunks.
+
+        Raises:
+            FileUploadError: If there's an error uploading any chunk.
+        """
+        if total_chunks <= 0:
+            return None
+
+        if total_chunks > 1:
+            num_regular_chunks = total_chunks - 1
+            for batch_start in range(0, num_regular_chunks, self.MAX_CONCURRENT_CHUNKS):
+                batch_end = min(
+                    batch_start + self.MAX_CONCURRENT_CHUNKS, num_regular_chunks
+                )
+
+                # Upload chunks in this batch concurrently
+                with ThreadPoolExecutor(
+                    max_workers=self.MAX_CONCURRENT_CHUNKS
+                ) as executor:
+                    futures = {
+                        executor.submit(
+                            self._upload_single_chunk,
+                            job_name,
+                            file_path,
+                            file_name,
+                            upload_id,
+                            chunk_index,
+                            total_chunks,
+                        ): chunk_index
+                        for chunk_index in range(batch_start, batch_end)
+                    }
+
+                    for future in as_completed(futures):
+                        chunk_index = futures[future]
+                        try:
+                            future.result()
+                            logger.debug(
+                                f"Uploaded chunk {chunk_index + 1}/{total_chunks} of {file_name}"
+                            )
+                        except Exception as e:
+                            logger.error(f"Error uploading chunk {chunk_index}: {e}")
+                            raise FileUploadError(
+                                f"Error uploading chunk {chunk_index} of {file_name}: {e}"
+                            ) from e
+
+        # Upload the final chunk with retry logic
+        final_chunk_index = total_chunks - 1
+        retries = 0
+        max_retries = 3
+        retry_delay = 2.0
+
+        while retries < max_retries:
+            try:
+                with open(file_path, "rb") as f:
+                    # Read the final chunk from the file
+                    f.seek(final_chunk_index * self.CHUNK_SIZE)
                     chunk_data = f.read(self.CHUNK_SIZE)
 
                     # Prepare and send the chunk
@@ -1028,40 +1159,108 @@ class RestClient:
                             }
                             data = {
                                 "file_name": file_name,
-                                "chunk_index": chunk_index,
+                                "chunk_index": final_chunk_index,
                                 "total_chunks": total_chunks,
                                 "upload_id": upload_id,
                             }
 
-                            # Send the chunk
+                            # Send the final chunk
                             response = self.multipart_client.post(
                                 f"/v0.1/crows/{job_name}/upload-chunk",
                                 files=files,
                                 data=data,
                             )
+
+                            # Handle missing chunks (status 409)
+                            if response.status_code == codes.CONFLICT:
+                                retries += 1
+                                if retries < max_retries:
+                                    logger.warning(
+                                        f"Missing chunks detected for {file_name}, retrying in {retry_delay}s... (attempt {retries}/{max_retries})"
+                                    )
+                                    time.sleep(retry_delay)
+                                    continue
+
                             response.raise_for_status()
+                            response_data = response.json()
+                            status_url = response_data.get("status_url")
 
-                        # Call progress callback if provided
+                            logger.debug(
+                                f"Uploaded final chunk {final_chunk_index + 1}/{total_chunks} of {file_name}"
+                            )
+                            return status_url
 
-                        logger.debug(
-                            f"Uploaded chunk {chunk_index + 1}/{total_chunks} of {file_name}"
-                        )
+            except Exception as e:
+                if retries >= max_retries - 1:
+                    raise FileUploadError(
+                        f"Error uploading final chunk of {file_name}: {e}"
+                    ) from e
+                retries += 1
+                logger.warning(
+                    f"Error uploading final chunk of {file_name}, retrying in {retry_delay}s... (attempt {retries}/{max_retries}): {e}"
+                )
+                time.sleep(retry_delay)
 
-            logger.info(f"Successfully uploaded {file_name}")
-        except HTTPStatusError as e:
-            if e.response.status_code in AUTH_ERRORS_TO_RETRY_ON:
-                raise AuthError(
-                    e.response.status_code,
-                    f"Authentication failed: {e}",
-                    request=e.request,
-                    response=e.response,
-                ) from e
-            raise
-        except Exception as e:
-            logger.exception(f"Error uploading file {file_path}")
-            raise FileUploadError(f"Error uploading file {file_path}: {e}") from e
+        raise FileUploadError(
+            f"Failed to upload final chunk of {file_name} after {max_retries} retries"
+        )
+
+    def _upload_single_chunk(
+        self,
+        job_name: str,
+        file_path: Path,
+        file_name: str,
+        upload_id: str,
+        chunk_index: int,
+        total_chunks: int,
+    ) -> None:
+        """Upload a single chunk.
+
+        Args:
+            job_name: The key of the crow to upload to.
+            file_path: The path to the file to upload.
+            file_name: The name to use for the file.
+            upload_id: The upload ID to use.
+            chunk_index: The index of this chunk.
+            total_chunks: Total number of chunks.
+
+        Raises:
+            Exception: If there's an error uploading the chunk.
+        """
+        with open(file_path, "rb") as f:
+            # Read the chunk from the file
+            f.seek(chunk_index * self.CHUNK_SIZE)
+            chunk_data = f.read(self.CHUNK_SIZE)
+
+            # Prepare and send the chunk
+            with tempfile.NamedTemporaryFile() as temp_file:
+                temp_file.write(chunk_data)
+                temp_file.flush()
+
+                # Create form data
+                with open(temp_file.name, "rb") as chunk_file_obj:
+                    files = {
+                        "chunk": (
+                            file_name,
+                            chunk_file_obj,
+                            "application/octet-stream",
+                        )
+                    }
+                    data = {
+                        "file_name": file_name,
+                        "chunk_index": chunk_index,
+                        "total_chunks": total_chunks,
+                        "upload_id": upload_id,
+                    }
+
+                    # Send the chunk
+                    response = self.multipart_client.post(
+                        f"/v0.1/crows/{job_name}/upload-chunk",
+                        files=files,
+                        data=data,
+                    )
+                    response.raise_for_status()
 
-    @refresh_token_on_auth_error()
     @retry(
         stop=stop_after_attempt(MAX_RETRY_ATTEMPTS),
         wait=wait_exponential(multiplier=RETRY_MULTIPLIER, max=MAX_RETRY_WAIT),
@@ -1098,13 +1297,6 @@ class RestClient:
             response.raise_for_status()
             return response.json()
         except HTTPStatusError as e:
-            if e.response.status_code in AUTH_ERRORS_TO_RETRY_ON:
-                raise AuthError(
-                    e.response.status_code,
-                    f"Authentication failed: {e}",
-                    request=e.request,
-                    response=e.response,
-                ) from e
             logger.exception(
                 f"Error listing files for job {job_name}, trajectory {trajectory_id}, upload_id {upload_id}: {e.response.text}"
             )
@@ -1117,7 +1309,6 @@ class RestClient:
             )
             raise RestClientError(f"Error listing files: {e!s}") from e
 
-    @refresh_token_on_auth_error()
     @retry(
         stop=stop_after_attempt(MAX_RETRY_ATTEMPTS),
         wait=wait_exponential(multiplier=RETRY_MULTIPLIER, max=MAX_RETRY_WAIT),
@@ -1165,13 +1356,6 @@ class RestClient:
 
             logger.info(f"File {file_path} downloaded to {destination_path}")
         except HTTPStatusError as e:
-            if e.response.status_code in AUTH_ERRORS_TO_RETRY_ON:
-                raise AuthError(
-                    e.response.status_code,
-                    f"Authentication failed: {e}",
-                    request=e.request,
-                    response=e.response,
-                ) from e
             logger.exception(
                 f"Error downloading file {file_path} for job {job_name}, trajectory_id {trajectory_id}: {e.response.text}"
             )