[Storage] R2 Cloud Object Storage integration

docs/source/getting-started/installation.rst

@@ -18,7 +18,7 @@ Install SkyPilot using pip:
   $ # pip install "skypilot[all]"
 
 
-SkyPilot currently supports four cloud providers: AWS, GCP, Azure, and Lambda Cloud.
+SkyPilot currently supports five cloud providers: AWS, GCP, Azure, Lambda Cloud, and Cloudflare(R2, storage only).
 If you only have access to certain clouds, use any combination of
 :code:`"[aws,azure,gcp,lambda]"` (e.g., :code:`"[aws,gcp]"`) to reduce the
 dependencies installed.
@@ -118,6 +118,44 @@ To configure Lambda Cloud access, go to the `API Keys <https://cloud.lambdalabs.
 
 .. _verify-cloud-access:
 
+Cloudflare R2
+~~~~~~~~~~~~~~~~~~
+
+As R2 is S3 compatible, we would need to set up AWS credentials for R2 with :code:`aws configure`. To generate the Access Key ID and Secret Access Key for R2, please follow these `instructions <https://developers.cloudflare.com/r2/data-access/s3-api/tokens/>`_. With the given Key values, you can set up the credentials. If you already configured for AWS, you can skip this step.
+
+.. code-block:: console
+
+  $ # Install boto
+  $ pip install boto3
+
+  $ # Configure your R2 credentials
+  $ aws configure
+
+Then, set the following.
+
+.. code-block:: text
+
+  AWS Access Key ID [None]: <access_key_id>
+  AWS Secret Access Key [None]: <access_key_secret>
+  Default region name [None]: auto
+  Default output format [None]: json
+
+Open :code:`~/.aws/credentials`, and if you did not configure for AWS, change :code:`[default]` to :code:`[r2]`. If you already have configured for AWS and skipped the step above, create another configuration for r2 below your configuration for AWS.
+
+.. code-block:: text
+
+  ...
+
+  [r2]
+  aws_access_key_id = <access_key_id>
+  aws_secret_access_key = <access_key_secret>
+
+Get your Account ID from your R2 dashboard and store it in :code:`~/.cloudflare/accountid`.
+
+.. code-block:: text
+
+  <account_id>
+
 Verifying cloud access
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 

sky/adaptors/cloudflare.py

@@ -0,0 +1,120 @@
+"""Cloudflare cloud adaptors"""
+
+# pylint: disable=import-outside-toplevel
+
+import functools
+import threading
+import os
+
+boto3 = None
+botocore = None
+_session_creation_lock = threading.RLock()
+ACCOUNT_ID_PATH = '~/.cloudflare/accountid'
+R2_PROFILE_NAME = 'r2'
+
+
+def import_package(func):
+
+    @functools.wraps(func)
+    def wrapper(*args, **kwargs):
+        global boto3, botocore
+        if boto3 is None or botocore is None:
+            try:
+                import boto3 as _boto3
+                import botocore as _botocore
+                boto3 = _boto3
+                botocore = _botocore
+            except ImportError:
+                raise ImportError('Fail to import dependencies for Cloudflare.'
+                                  'Try pip install "skypilot[aws]"') from None
+        return func(*args, **kwargs)
+
+    return wrapper
+
+
+# lru_cache() is thread-safe and it will return the same session object
+# for different threads.
+# Reference: https://docs.python.org/3/library/functools.html#functools.lru_cache # pylint: disable=line-too-long
+@functools.lru_cache()
+@import_package
+def session():
+    """Create an AWS session."""
+    # Creating the session object is not thread-safe for boto3,
+    # so we add a reentrant lock to synchronize the session creation.
+    # Reference: https://github.com/boto/boto3/issues/1592
+    # However, the session object itself is thread-safe, so we are
+    # able to use lru_cache() to cache the session object.
+    with _session_creation_lock:
+        return boto3.session.Session(profile_name=R2_PROFILE_NAME)
+
+
+@functools.lru_cache()
+@import_package
+def resource(resource_name: str):
+    """Create an Cloudflare resource.
+
+    Args:
+        resource_name: Cloudflare resource name (e.g., 's3').
+        kwargs: Other options.
+    """
+    # Need to use the resource retrieved from the per-thread session
+    # to avoid thread-safety issues (Directly creating the client
+    # with boto3.resource() is not thread-safe).
+    # Reference: https://stackoverflow.com/a/59635814
+
+    session_ = session()
+    cloudflare_credentials = session_.get_credentials().get_frozen_credentials()
+    endpoint = create_endpoint()
+
+    return session_.resource(
+        resource_name,
+        endpoint_url=endpoint,
+        aws_access_key_id=cloudflare_credentials.access_key,
+        aws_secret_access_key=cloudflare_credentials.secret_key,
+        region_name='auto')
+
+
+@functools.lru_cache()
+def client(service_name: str, region):
+    """Create an CLOUDFLARE client of a certain service.
+
+    Args:
+        service_name: CLOUDFLARE service name (e.g., 's3').
+        kwargs: Other options.
+    """
+    # Need to use the client retrieved from the per-thread session
+    # to avoid thread-safety issues (Directly creating the client
+    # with boto3.client() is not thread-safe).
+    # Reference: https://stackoverflow.com/a/59635814
+
+    session_ = session()
+    cloudflare_credentials = session_.get_credentials().get_frozen_credentials()
+    endpoint = create_endpoint()
+
+    return session_.client(
+        service_name,
+        endpoint_url=endpoint,
+        aws_access_key_id=cloudflare_credentials.access_key,
+        aws_secret_access_key=cloudflare_credentials.secret_key,
+        region_name=region)
+
+
+@import_package
+def botocore_exceptions():
+    """AWS botocore exception."""
+    from botocore import exceptions
+    return exceptions
+
+
+def create_endpoint():
+    """Reads accountid necessary to interact with R2"""
+
+    accountid_path = os.path.expanduser(ACCOUNT_ID_PATH)
+    with open(accountid_path, 'r') as f:
+        lines = f.readlines()
+        accountid = lines[0]
+
+    accountid = accountid.strip()
+    endpoint = 'https://' + accountid + '.r2.cloudflarestorage.com'
+
+    return endpoint

sky/cloud_stores.py

@@ -12,7 +12,7 @@
 
 from sky.clouds import gcp
 from sky.data import data_utils
-from sky.adaptors import aws
+from sky.adaptors import aws, cloudflare
 
 
 class CloudStorage:
@@ -146,6 +146,63 @@ def make_sync_file_command(self, source: str, destination: str) -> str:
         return ' && '.join(all_commands)
 
 
+class R2CloudStorage(CloudStorage):
+    """Cloudflare Cloud Storage."""
+
+    # List of commands to install AWS CLI
+    _GET_AWSCLI = [
+        'aws --version >/dev/null 2>&1 || pip3 install awscli',
+    ]
+
+    def is_directory(self, url: str) -> bool:
+        """Returns whether R2 'url' is a directory.
+
+        In cloud object stores, a "directory" refers to a regular object whose
+        name is a prefix of other objects.
+        """
+        r2 = cloudflare.resource('s3')
+        bucket_name, path = data_utils.split_r2_path(url)
+        bucket = r2.Bucket(bucket_name)
+
+        num_objects = 0
+        for obj in bucket.objects.filter(Prefix=path):
+            num_objects += 1
+            if obj.key == path:
+                return False
+            # If there are more than 1 object in filter, then it is a directory
+            if num_objects == 3:
+                return True
+
+        # A directory with few or no items
+        return True
+
+    def make_sync_dir_command(self, source: str, destination: str) -> str:
+        """Downloads using AWS CLI."""
+        # AWS Sync by default uses 10 threads to upload files to the bucket.
+        # To increase parallelism, modify max_concurrent_requests in your
+        # aws config file (Default path: ~/.aws/config).
+        endpoint_url = cloudflare.create_endpoint()
+        download_via_awscli = ('aws s3 sync --no-follow-symlinks '
+                               f'{source} s3://{destination} '
+                               f'--endpoint {endpoint_url} '
+                               f'--profile={cloudflare.R2_PROFILE_NAME}')
+
+        all_commands = list(self._GET_AWSCLI)
+        all_commands.append(download_via_awscli)
+        return ' && '.join(all_commands)
+
+    def make_sync_file_command(self, source: str, destination: str) -> str:
+        """Downloads a file using AWS CLI."""
+        endpoint_url = cloudflare.create_endpoint()
+        download_via_awscli = (f'aws s3 cp s3://{source} {destination} '
+                               f'--endpoint {endpoint_url} '
+                               f'--profile={cloudflare.R2_PROFILE_NAME}')
+
+        all_commands = list(self._GET_AWSCLI)
+        all_commands.append(download_via_awscli)
+        return ' && '.join(all_commands)
+
+
 def get_storage_from_path(url: str) -> CloudStorage:
     """Returns a CloudStorage by identifying the scheme:// in a URL."""
     result = urllib.parse.urlsplit(url)
@@ -159,4 +216,5 @@ def get_storage_from_path(url: str) -> CloudStorage:
 _REGISTRY = {
     'gs': GcsCloudStorage(),
     's3': S3CloudStorage(),
+    'r2': R2CloudStorage()
 }

sky/data/data_transfer.py

@@ -13,6 +13,7 @@
 
 TODO:
 - All combinations of Azure Transfer
+- All combinations of R2 Transfer
 - GCS -> S3
 """
 import json
@@ -116,6 +117,21 @@ def s3_to_gcs(s3_bucket_name: str, gs_bucket_name: str) -> None:
         f'Transfer finished in {(time.time() - start) / 60:.2f} minutes.')
 
 
+def s3_to_r2(s3_bucket_name: str, r2_bucket_name: str) -> None:
+    """Creates a one-time transfer from Amazon S3 to Google Cloud Storage.
+
+    Can be viewed from: https://console.cloud.google.com/transfer/cloud
+    it will block until the transfer is complete.
+
+    Args:
+      s3_bucket_name: str; Name of the Amazon S3 Bucket
+      r2_bucket_name: str; Name of the Cloudflare R2 Bucket
+    """
+    raise NotImplementedError('Moving data directly from clouds to R2 is '
+                              'currently not supported. Please specify '
+                              'a local source for the storage object.')
+
+
 def gcs_to_s3(gs_bucket_name: str, s3_bucket_name: str) -> None:
     """Creates a one-time transfer from Google Cloud Storage to Amazon S3.
 
@@ -129,6 +145,48 @@ def gcs_to_s3(gs_bucket_name: str, s3_bucket_name: str) -> None:
     subprocess.call(sync_command, shell=True)
 
 
+def gcs_to_r2(gs_bucket_name: str, r2_bucket_name: str) -> None:
+    """Creates a one-time transfer from Google Cloud Storage to Amazon S3.
+
+     Args:
+      gs_bucket_name: str; Name of the Google Cloud Storage Bucket
+      r2_bucket_name: str; Name of the Cloudflare R2 Bucket
+    """
+    raise NotImplementedError('Moving data directly from clouds to R2 is '
+                              'currently not supported. Please specify '
+                              'a local source for the storage object.')
+
+
+def r2_to_gcs(r2_bucket_name: str, gs_bucket_name: str) -> None:
+    """Creates a one-time transfer from Cloudflare R2 to Google Cloud Storage.
+
+    Can be viewed from: https://console.cloud.google.com/transfer/cloud
+    it will block until the transfer is complete.
+
+    Args:
+      r2_bucket_name: str; Name of the Cloudflare R2 Bucket
+      gs_bucket_name: str; Name of the Google Cloud Storage Bucket
+    """
+    raise NotImplementedError('Moving data directly from R2 to clouds is '
+                              'currently not supported. Please specify '
+                              'a local source for the storage object.')
+
+
+def r2_to_s3(r2_bucket_name: str, s3_bucket_name: str) -> None:
+    """Creates a one-time transfer from Amazon S3 to Google Cloud Storage.
+
+    Can be viewed from: https://console.cloud.google.com/transfer/cloud
+    it will block until the transfer is complete.
+
+    Args:
+      r2_bucket_name: str; Name of the Cloudflare R2 Bucket\
+      s3_bucket_name: str; Name of the Amazon S3 Bucket
+    """
+    raise NotImplementedError('Moving data directly from R2 to clouds is '
+                              'currently not supported. Please specify '
+                              'a local source for the storage object.')
+
+
 def _add_bucket_iam_member(bucket_name: str, role: str, member: str) -> None:
     storage_client = gcp.storage_client()
     bucket = storage_client.bucket(bucket_name)

sky/data/data_utils.py

@@ -8,7 +8,7 @@
 
 from sky import exceptions
 from sky import sky_logging
-from sky.adaptors import aws, gcp
+from sky.adaptors import aws, gcp, cloudflare
 from sky.utils import ux_utils
 
 Client = Any
@@ -40,6 +40,18 @@ def split_gcs_path(gcs_path: str) -> Tuple[str, str]:
     return bucket, key
 
 
+def split_r2_path(r2_path: str) -> Tuple[str, str]:
+    """Splits R2 Path into Bucket name and Relative Path to Bucket
+
+    Args:
+      r2_path: str; R2 Path, e.g. r2://imagenet/train/
+    """
+    path_parts = r2_path.replace('r2://', '').split('/')
+    bucket = path_parts.pop(0)
+    key = '/'.join(path_parts)
+    return bucket, key
+
+
 def create_s3_client(region: str = 'us-east-2') -> Client:
     """Helper method that connects to Boto3 client for S3 Bucket
 
@@ -73,6 +85,26 @@ def verify_gcs_bucket(name: str) -> bool:
         return False
 
 
+def create_r2_client(region: str = 'auto') -> Client:
+    """Helper method that connects to Boto3 client for R2 Bucket
+
+    Args:
+      region: str; Region for CLOUDFLARE R2 is set to auto
+    """
+    return cloudflare.client('s3', region)
+
+
+def verify_r2_bucket(name: str) -> bool:
+    """Helper method that checks if the R2 bucket exists
+
+    Args:
+      name: str; Name of R2 Bucket (without r2:// prefix)
+    """
+    r2 = cloudflare.resource('s3')
+    bucket = r2.Bucket(name)
+    return bucket in r2.buckets.all()
+
+
 def is_cloud_store_url(url):
     result = urllib.parse.urlsplit(url)
     # '' means non-cloud URLs.
@@ -118,7 +150,7 @@ def parallel_upload(source_path_list: List[str],
                     max_concurrent_uploads: Optional[int] = None) -> None:
     """Helper function to run parallel uploads for a list of paths.
 
-    Used by S3Store and GCSStore to run rsync commands in parallel by
+    Used by S3Store, GCSStore, and R2Store to run rsync commands in parallel by
     providing appropriate command generators.
 
     Args: