Allow usage of unsigned S3 client (#195)

* Disable signing to make requests without AWS credentials

Update S3ClientConfig to pass in the configuration
for allowing unsigned requests.

* Update instructions

Author: dnanuti

CHANGELOG.md

@@ -1,3 +1,9 @@
+## Unreleased
+
+### New features
+* Update S3ClientConfig to pass in the configuration for allowing unsigned requests, under boolean flag `unsigned`.
+
+
 ## v1.2.2 (March 22, 2024)
 
 ### New features

doc/DEVELOPMENT.md

@@ -150,6 +150,8 @@ Using S3ClientConfig you can set up the following parameters for the underlying
   (max number of parts per upload is 10,000, minimum upload part size is 5 MiB).
   Part size must have **values between 5MiB and 5GiB.** Is set by default to **8MiB** (may change in future).
 
+* `unsigned(bool)`: Set to true to disable signing S3 requests.
+
 For example this can be passed in like: 
 ```py
 from s3torchconnector import S3MapDataset, S3ClientConfig
@@ -165,6 +167,9 @@ s3_map_dataset = S3MapDataset.from_prefix(DATASET_URI, region=REGION, s3client_c
 s3_checkpoint = S3Checkpoint(region=REGION, s3client_config=config)
 # Works similarly for Lightning checkpoints.
 s3_lightning_checkpoint = S3LightningCheckpoint(region=REGION, s3client_config=config)
+
+# Disable signing to make requests without AWS credentials
+s3_client = S3Client(region=REGION, s3client_config=S3ClientConfig(unsigned=True))
 ```
 
 **When modifying the default values for these flags, we strongly recommend to run benchmarking to ensure you are not

s3torchconnector/src/s3torchconnector/_s3client/_mock_s3client.py

@@ -37,6 +37,7 @@ def __init__(
             throughput_target_gbps=self.s3client_config.throughput_target_gbps,
             part_size=self.s3client_config.part_size,
             user_agent_prefix=self.user_agent_prefix,
+            unsigned=self.s3client_config.unsigned,
         )
 
     def add_object(self, key: str, data: bytes) -> None:

s3torchconnector/src/s3torchconnector/_s3client/_s3client.py

@@ -77,6 +77,7 @@ def _client_builder(self) -> MountpointS3Client:
             user_agent_prefix=self._user_agent_prefix,
             throughput_target_gbps=self._s3client_config.throughput_target_gbps,
             part_size=self._s3client_config.part_size,
+            unsigned=self._s3client_config.unsigned,
         )
 
     def get_object(

s3torchconnector/src/s3torchconnector/_s3client/s3client_config.py

@@ -19,3 +19,4 @@ class S3ClientConfig:
 
     throughput_target_gbps: float = 10.0
     part_size: int = 8 * 1024 * 1024
+    unsigned: bool = False

s3torchconnector/tst/e2e/test_e2e_s3datasets.py

@@ -9,7 +9,7 @@
 from torch.utils.data.datapipes.datapipe import MapDataPipe
 from torchdata.datapipes.iter import IterableWrapper, IterDataPipe
 
-from s3torchconnector import S3IterableDataset, S3MapDataset
+from s3torchconnector import S3IterableDataset, S3MapDataset, S3ClientConfig
 
 
 def test_s3iterable_dataset_images_10_from_prefix(image_directory):
@@ -100,6 +100,21 @@ def test_dataset_unpickled_iterates(image_directory):
     assert expected == actual
 
 
+def test_unsigned_client():
+    s3_uri = "s3://s3torchconnector-demo/geonet/images/"
+    region = "us-east-1"
+    s3_dataset = S3MapDataset.from_prefix(
+        s3_uri=s3_uri,
+        region=region,
+        transform=lambda obj: obj.read(),
+        s3client_config=S3ClientConfig(unsigned=True),
+    )
+    s3_dataloader = _pytorch_dataloader(s3_dataset)
+    assert s3_dataloader is not None
+    assert isinstance(s3_dataloader.dataset, S3MapDataset)
+    assert len(s3_dataloader) >= 1296
+
+
 def _compare_dataloaders(
     local_dataloader: DataLoader, s3_dataloader: DataLoader, expected_batch_count: int
 ):

s3torchconnector/tst/unit/test_s3_client.py

@@ -107,6 +107,7 @@ def test_s3_client_custom_config(part_size: int, throughput_target_gbps: float):
     )
     assert s3_client._client.part_size == part_size
     assert s3_client._client.throughput_target_gbps == throughput_target_gbps
+    assert s3_client._client.unsigned is False
 
 
 @pytest.mark.parametrize(
@@ -130,3 +131,11 @@ def test_s3_client_invalid_part_size_config(part_size: int):
         )
         # The client is lazily initialized
         assert s3_client._client.part_size == part_size
+
+
+def test_unsigned_s3_client():
+    s3_client = S3Client(
+        region=TEST_REGION,
+        s3client_config=S3ClientConfig(unsigned=True),
+    )
+    assert s3_client._client.unsigned is True

s3torchconnectorclient/python/src/s3torchconnectorclient/_mountpoint_s3_client.pyi

@@ -10,7 +10,7 @@ class MountpointS3Client:
     region: str
     part_size: int
     profile: Optional[str]
-    no_sign_request: bool
+    unsigned: Optional[bool]
     user_agent_prefix: str
     endpoint: str
 
@@ -21,7 +21,7 @@ class MountpointS3Client:
         throughput_target_gbps: float = 10.0,
         part_size: int = 8 * 1024 * 1024,
         profile: Optional[str] = None,
-        no_sign_request: bool = False,
+        unsigned: Optional[bool] = False,
         endpoint: Optional[str] = None,
     ): ...
     def get_object(self, bucket: str, key: str) -> GetObjectStream: ...
@@ -39,6 +39,7 @@ class MockMountpointS3Client:
     region: str
     part_size: int
     user_agent_prefix: str
+    unsigned: bool
 
     def __init__(
         self,
@@ -48,6 +49,7 @@ class MockMountpointS3Client:
         throughput_target_gbps: float = 10.0,
         part_size: int = 8 * 1024 * 1024,
         user_agent_prefix: str = "mock_client",
+        unsigned: bool = False,
     ): ...
     def create_mocked_client(self) -> MountpointS3Client: ...
     def add_object(self, key: str, data: bytes) -> None: ...

s3torchconnectorclient/python/tst/unit/test_mountpoint_s3_client.py

@@ -243,7 +243,7 @@ def test_put_object_with_storage_class():
 # TODO: Add hypothesis setup after aligning on limits
 def test_mountpoint_client_pickles():
     expected_profile = None
-    expected_no_sign_request = False
+    expected_unsigned = False
     expected_region = REGION
     expected_part_size = 5 * 2**20
     expected_throughput_target_gbps = 3.5
@@ -254,7 +254,7 @@ def test_mountpoint_client_pickles():
         part_size=expected_part_size,
         throughput_target_gbps=expected_throughput_target_gbps,
         profile=expected_profile,
-        no_sign_request=expected_no_sign_request,
+        unsigned=expected_unsigned,
     )
     dumped = pickle.dumps(client)
     loaded = pickle.loads(dumped)
@@ -271,7 +271,7 @@ def test_mountpoint_client_pickles():
         == expected_throughput_target_gbps
     )
     assert client.profile == loaded.profile == expected_profile
-    assert client.no_sign_request == loaded.no_sign_request == expected_no_sign_request
+    assert client.unsigned == loaded.unsigned == expected_unsigned
 
 
 @pytest.mark.parametrize(

s3torchconnectorclient/rust/src/mock_client.rs

@@ -26,18 +26,21 @@ pub struct PyMockClient {
     pub(crate) part_size: usize,
     #[pyo3(get)]
     pub(crate) user_agent_prefix: String,
+    #[pyo3(get)]
+    pub(crate) unsigned: bool,
 }
 
 #[pymethods]
 impl PyMockClient {
     #[new]
-    #[pyo3(signature = (region, bucket, throughput_target_gbps = 10.0, part_size = 8 * 1024 * 1024, user_agent_prefix="mock_client".to_string()))]
+    #[pyo3(signature = (region, bucket, throughput_target_gbps = 10.0, part_size = 8 * 1024 * 1024, user_agent_prefix="mock_client".to_string(), unsigned=false))]
     pub fn new(
         region: String,
         bucket: String,
         throughput_target_gbps: f64,
         part_size: usize,
         user_agent_prefix: String,
+        unsigned: bool,
     ) -> PyMockClient {
         let unordered_list_seed: Option<u64> = None;
         let config = MockClientConfig { bucket, part_size, unordered_list_seed };
@@ -48,7 +51,8 @@ impl PyMockClient {
             region,
             throughput_target_gbps,
             part_size,
-            user_agent_prefix
+            user_agent_prefix,
+            unsigned
         }
     }
 
@@ -59,7 +63,7 @@ impl PyMockClient {
             self.throughput_target_gbps,
             self.part_size,
             None,
-            false,
+            self.unsigned,
             self.mock_client.clone(),
             None,
         )