Enable TNU to use Azure auth for DRS operations #399

.gitignore

@@ -26,6 +26,9 @@ __pycache__/
 /.idea
 /*.iml
 
+# Default virtual environment name used by PyCharm
+/.venv
+
 # JS/node/npm/web dev files
 node_modules
 npm-debug.log

README.md

@@ -40,11 +40,13 @@ pip install --upgrade --no-cache-dir terra-notebook-utils
 
 ## Credentials
 Much of the terra-notebook-utilities functionality requires credentialed access through a Google Cloud Platform account.
-Credentials are already available when running in a Terra notebook environment. Otherwise, credentials may be obtained
-with the command
+Credentials are already available when running in a Terra Google notebook environment.
+Otherwise, Google credentials may be obtained with the command
 ```
 gcloud auth application-default login
 ```
+The terra-notebook-utilities `drs` subcommands (only) run successfully in a Terra Azure
+notebook environment using the Azure default credentials.
 
 ## Usage
 

requirements.txt

@@ -1,3 +1,4 @@
+azure-identity >= 1.12.0, < 2
 google-cloud-storage >= 1.38.0, < 2
 gs-chunked-io >= 0.5.1, < 0.6
 firecloud

terra_notebook_utils/__init__.py

@@ -1,4 +1,6 @@
 import os
+from dataclasses import dataclass
+from enum import Enum
 
 WORKSPACE_NAME = os.environ.get('WORKSPACE_NAME', None)
 WORKSPACE_NAMESPACE = os.environ.get('WORKSPACE_NAMESPACE')  # This env var is set in Terra Cloud Environments
@@ -26,3 +28,20 @@
     DRS_RESOLVER_URL = MARTHA_URL
 else:
     DRS_RESOLVER_URL = DRSHUB_URL
+
+
+class ExecutionEnvironment(Enum):
+    TERRA_WORKSPACE = "TERRA_WORKSPACE",  # Executing in a Terra Workspace (on any supported platform)
+    OTHER = "OTHER"  # Executing outside a Terra Workspace (e.g., local system)
+
+
+class ExecutionPlatform(Enum):
+    AZURE = "AZURE",  # Executing in an Azure compute environment
+    GOOGLE = "GOOGLE",  # Executing in a Google compute environment
+    UNKNOWN = "UNKNOWN"  # Execution platform not identified (e.g., local system)
+
+
+@dataclass
+class ExecutionContext:
+    execution_environment: ExecutionEnvironment
+    execution_platform: ExecutionPlatform

terra_notebook_utils/azure_auth.py

@@ -0,0 +1,70 @@
+"""
+Microsoft Azure identity/auth support.
+
+See:
+https://azuresdkdocs.blob.core.windows.net/$web/python/azure-identity/1.12.0/index.html
+https://learn.microsoft.com/en-us/python/api/azure-identity/azure.identity.defaultazurecredential?view=azure-python
+https://github.com/Azure/azure-sdk-for-python/blob/main/sdk/identity/azure-identity/azure/identity/_credentials/default.py
+"""
+import logging
+import os
+from typing import Optional
+
+from azure.identity import DefaultAzureCredential
+from terra_notebook_utils.logger import logger
+
+
+# Single instance of DefaultAzureCredential that initialized lazily.
+# The instance is treated as threadsafe and reusable.
+# The Azure documentation is silent on thread safety.
+# Based on scanning the code, it appears to be threadsafe.
+# See: https://github.com/Azure/azure-sdk-for-python/blob/main/sdk/identity/
+#              azure-identity/azure/identity/_credentials/default.py
+_AZURE_CREDENTIAL: Optional[DefaultAzureCredential] = None
+
+
+def _set_azure_identity_logging_level(level: int) -> None:
+    """ Set the logging level for modules participating the Azure default credential flow """
+    import azure.identity
+    logging.getLogger(azure.identity._credentials.environment.__name__).setLevel(level)
+    logging.getLogger(azure.identity._credentials.managed_identity.__name__).setLevel(level)
+    logging.getLogger(azure.identity._credentials.chained.__name__).setLevel(level)
+    logging.getLogger("azure.core.pipeline.policies.http_logging_policy").setLevel(level)
+
+
+# Suppress extraneous azure-identity INFO level logging
+_set_azure_identity_logging_level(logging.WARNING)
+
+
+def _get_default_credential() -> DefaultAzureCredential:
+    """
+    Instantiate DefaultAzureCredential lazily if/when needed.
+
+    Note: It would not need to be instantiated this way, as
+    # no exception is raised even if Azure credentials are not configured.
+    :return: Reference to instance of DefaultAzureCredential
+    """
+
+    # Should a more sophisticated Singleton pattern be used instead?
+    global _AZURE_CREDENTIAL
+    if not _AZURE_CREDENTIAL:
+        _AZURE_CREDENTIAL = DefaultAzureCredential()
+    return _AZURE_CREDENTIAL
+
+
+def get_azure_access_token() -> str:
+    """
+    Return an Azure access token.
+
+    raises ClientAuthenticationError
+    """
+    if os.environ.get('TERRA_NOTEBOOK_AZURE_ACCESS_TOKEN'):
+        logger.debug("Using Azure token configured using 'TERRA_NOTEBOOK_AZURE_ACCESS_TOKEN'")
+        token = os.environ['TERRA_NOTEBOOK_AZURE_ACCESS_TOKEN']
+    else:
+        logger.debug("Requesting Azure default credentials token.")
+        token_scope = "https://management.azure.com/.default"
+        azure_token = _get_default_credential().get_token(token_scope)
+        logger.debug("Using Azure default credentials token.")
+        token = azure_token.token
+    return token

terra_notebook_utils/drs.py

@@ -18,6 +18,7 @@
 from terra_notebook_utils.blobstore.progress import Indicator
 from terra_notebook_utils.blobstore import Blob, copy_client, BlobNotFoundError
 from terra_notebook_utils.logger import logger
+from terra_notebook_utils.terra_auth import get_terra_access_token
 
 
 DRSInfo = namedtuple("DRSInfo", "credentials access_url bucket_name key name size updated checksums")
@@ -46,7 +47,7 @@ def enable_requester_pays(workspace_name: Optional[str]=WORKSPACE_NAME,
     rawls_url = (f"https://rawls.dsde-{TERRA_DEPLOYMENT_ENV}.broadinstitute.org/api/workspaces/"
                  f"{workspace_namespace}/{encoded_workspace}/enableRequesterPaysForLinkedServiceAccounts")
     logger.info("Enabling requester pays for your workspace. This will only take a few seconds...")
-    access_token = gs.get_access_token()
+    access_token = get_terra_access_token()
 
     headers = {
         'authorization': f"Bearer {access_token}",
@@ -61,7 +62,7 @@ def enable_requester_pays(workspace_name: Optional[str]=WORKSPACE_NAME,
 
 def get_drs(drs_url: str, fields: List[str]) -> Response:
     """Request DRS information from DRS Resolver."""
-    access_token = gs.get_access_token()
+    access_token = get_terra_access_token()
 
     headers = {
         'authorization': f"Bearer {access_token}",

terra_notebook_utils/logger.py

@@ -1,4 +1,4 @@
-import json
 import logging
+from logging import Logger
 
-logger = logging.getLogger(__name__)
+logger: Logger = logging.getLogger(__name__)

terra_notebook_utils/terra_auth.py

@@ -0,0 +1,70 @@
+"""
+Support for auth with Terra backend services.
+"""
+from azure.core.exceptions import ClientAuthenticationError
+from google.auth.exceptions import DefaultCredentialsError
+
+from terra_notebook_utils import azure_auth, gs, ExecutionPlatform
+from terra_notebook_utils.logger import logger
+from terra_notebook_utils.utils import get_execution_context
+
+
+class AuthenticationError(Exception):
+    pass
+
+
+class TerraAuthTokenProvider:
+    """
+    Provides auth bearer tokens suitable for use with Terra backend services.
+    """
+    def __init__(self):
+        self.execution_context = get_execution_context()
+
+    @staticmethod
+    def _identify_valid_access_token() -> str:
+        """
+        Try to obtain an auth bearer token suitable for use with Terra backend services
+        from the Terra supported auth providers. First try Google, then try Azure.
+        Return the first successfully obtained token, otherwise raise AuthenticationError.
+
+        :return: auth bearer token suitable for use with Terra backend services
+        :raises: AuthenticationError
+        """
+        try:
+            logger.debug("Attempting to obtain a Google access token to use with Terra backend services.")
+            google_token = gs.get_access_token()
+            logger.debug("Using Google access token to use with Terra backend services.")
+            return google_token
+        except DefaultCredentialsError as ex:
+            logger.debug("Failed to obtain a Google access token to use with Terra backend services.", exc_info=ex)
+
+        try:
+            logger.debug("Attempting to obtain a Azure access token to use with Terra backend services.")
+            azure_token = azure_auth.get_azure_access_token()
+            logger.debug("Using Azure access token to use with Terra backend services.")
+            return azure_token
+        except ClientAuthenticationError as ex:
+            logger.debug("Failed to obtain a Azure access token to use with Terra backend services.", exc_info=ex)
+
+        raise AuthenticationError("Failed to obtain a Google or Azure token to auth with Terra backend services.")
+
+    def get_terra_access_token(self) -> str:
+        if self.execution_context.execution_platform == ExecutionPlatform.GOOGLE:
+            logger.debug("Using Google default credentials to auth with Terra services.")
+            return gs.get_access_token()
+        elif self.execution_context.execution_platform == ExecutionPlatform.AZURE:
+            logger.debug("Using Azure default credentials to auth with Terra services.")
+            return azure_auth.get_azure_access_token()
+        else:
+            return self._identify_valid_access_token()
+
+
+# Single instance of TerraAuthTokenProvider.
+TERRA_AUTH_TOKEN_PROVIDER = TerraAuthTokenProvider()
+
+
+def get_terra_access_token() -> str:
+    """ Return an auth bearer token suitable for use with Terra backend services.
+    :raises: AuthenticationError
+    """
+    return TERRA_AUTH_TOKEN_PROVIDER.get_terra_access_token()

terra_notebook_utils/utils.py

@@ -1,11 +1,14 @@
 import json
+import os
 import threading
 from functools import lru_cache
 from concurrent.futures import ThreadPoolExecutor, Future, as_completed
 from typing import Any, Callable, Dict, Optional, Iterable, Set
 
 import jmespath
 
+from terra_notebook_utils import ExecutionEnvironment, ExecutionPlatform, ExecutionContext
+
 
 class _AsyncContextManager:
     """Context manager for asynchronous execution. Wait on exit for all jobs to complete."""
@@ -63,3 +66,25 @@ def is_notebook() -> bool:
         return "ZMQInteractiveShell" == get_ipython().__class__.__name__  # type: ignore
     except NameError:
         return False
+
+
+@lru_cache()
+def get_execution_context() -> ExecutionContext:
+    """
+    Identify information about the context in which terra-notebook-utils is executing.
+    TODO Improve the information available and algorithm to identify these values accurately!
+    """
+    # Workaround current insufficient information by assuming
+    # the execution environment is Terra, as that is the most
+    # common and important case.
+    # execution_environment = ExecutionEnvironment.OTHER
+    execution_environment = ExecutionEnvironment.TERRA_WORKSPACE
+    execution_platform = ExecutionPlatform.UNKNOWN
+    workspace_bucket = os.environ.get('WORKSPACE_BUCKET', None)
+    if workspace_bucket and workspace_bucket.startswith("gs://"):
+        execution_platform = ExecutionPlatform.GOOGLE
+    else:
+        # Workaround current insufficient information by assuming
+        # the execution platform is not Google then it is Azure.
+        execution_platform = ExecutionPlatform.AZURE
+    return ExecutionContext(execution_environment, execution_platform)

tests/test_azure_auth.py

@@ -0,0 +1,54 @@
+#!/usr/bin/env python
+"""
+Test auth for Terra running on Microsoft Azure.
+Ideally tests should be run in a Terra Azure workspace.
+Unit test this module by mocking as needed to run in a generic context.
+"""
+import os
+import sys
+import unittest
+from unittest import mock
+from unittest.mock import patch
+from collections import namedtuple
+
+
+pkg_root = os.path.abspath(os.path.join(os.path.dirname(__file__), '..'))  # noqa
+sys.path.insert(0, pkg_root)  # noqa
+
+# from tests import config  # initialize the test environment
+from tests.infra.testmode import testmode
+from terra_notebook_utils import azure_auth
+
+
+@testmode("workspace_access")
+class TestAzureAuth(unittest.TestCase):
+    def test_get_azure_access_token(self):
+        """
+         See this article for info about mock patch for environment variables:
+         https://adamj.eu/tech/2020/10/13/how-to-mock-environment-variables-with-pythons-unittest/
+        """
+        dummy_token_value = "@@DUMMY_AZURE_ACCESS_TOKEN@@"
+
+        # Test token that is explicitly provided by env var "TERRA_NOTEBOOK_AZURE_ACCESS_TOKEN"
+        with self.subTest("Azure token provided as env var"):
+            env_vars = {"TERRA_NOTEBOOK_AZURE_ACCESS_TOKEN": dummy_token_value}
+            with mock.patch.dict(os.environ, env_vars, clear=True):
+                self.assertEqual(dummy_token_value, azure_auth.get_azure_access_token())
+
+        # Test getting Azure default credentials token
+        with self.subTest("Get Azure default credentials access token"):
+            AzureToken = namedtuple("AzureToken", "token")
+            azure_token = AzureToken(dummy_token_value)
+            with patch.object(azure_auth.DefaultAzureCredential, "get_token", return_value=azure_token):
+                self.assertEqual(dummy_token_value, azure_auth.get_azure_access_token())
+
+        # Verify that the default credential is managed as a singleton
+        with self.subTest("Verify DefaultAzureCredential singleton"):
+            azure_auth._AZURE_CREDENTIAL = None
+            first_value = azure_auth._get_default_credential()
+            second_value = azure_auth._get_default_credential()
+            self.assertIs(first_value, second_value)
+
+
+if __name__ == '__main__':
+    unittest.main()