Delay S3 resource construction until object requested (#111)

Constructing a S3 handle takes up memory and time.  We retrieve the backend whenever we initialize a tile, but we want to delay the expensive work until the data is requested.

The perf issue is sufficiently bad that we want to do a release, so bumping to 4.0.1.

Test plan: Able to quickly load an experiment where all the tiles are located in the cloud.

setup.py

@@ -7,7 +7,7 @@
 
 setuptools.setup(
     name="slicedimage",
-    version="4.0.0",
+    version="4.0.1",
     description="Library to access sliced imaging data",
     author="Tony Tung",
     author_email="[email protected]",

slicedimage/backends/_s3.py

@@ -15,38 +15,42 @@ class S3Backend(Backend):
     CONFIG_UNSIGNED_REQUESTS_KEY = "unsigned-requests"
 
     def __init__(self, baseurl, s3_config):
-        unsigned_requests = s3_config.get(S3Backend.CONFIG_UNSIGNED_REQUESTS_KEY, False)
-
-        if unsigned_requests:
-            resource_config = Config(signature_version=UNSIGNED)
-        else:
-            resource_config = None
-
         parsed = urllib.parse.urlparse(baseurl)
         assert parsed[0].lower() == "s3"
-        session = boto3.session.Session()
-        s3 = session.resource("s3", config=resource_config)
-        self._bucket = s3.Bucket(parsed[1])
 
+        self._bucket = parsed[1]
         if parsed[2][0] == "/":
             self._basepath = PurePosixPath(parsed[2][1:])
         else:
             self._basepath = PurePosixPath(parsed[2])
+        self._s3_config = s3_config
 
     def read_contextmanager(self, name, checksum_sha256=None):
         key = str(self._basepath / name)
-        print(key)
-        return _S3ContextManager(self._bucket.Object(key), checksum_sha256)
+        return _S3ContextManager(self._bucket, key, checksum_sha256, self._s3_config)
 
 
 class _S3ContextManager:
-    def __init__(self, s3_obj, checksum_sha256):
-        self.s3_obj = s3_obj
+    def __init__(self, s3_bucket, s3_key, checksum_sha256, s3_config):
+        self.s3_bucket = s3_bucket
+        self.s3_key = s3_key
         self.checksum_sha256 = checksum_sha256
+        self.s3_config = s3_config
 
     def __enter__(self):
+        unsigned_requests = self.s3_config.get(S3Backend.CONFIG_UNSIGNED_REQUESTS_KEY, False)
+
+        if unsigned_requests:
+            resource_config = Config(signature_version=UNSIGNED)
+        else:
+            resource_config = None
+
+        session = boto3.session.Session()
+        s3 = session.resource("s3", config=resource_config)
+        bucket = s3.Bucket(self.s3_bucket)
+        s3_obj = bucket.Object(self.s3_key)
         self.buffer = BytesIO()
-        self.s3_obj.download_fileobj(self.buffer)
+        s3_obj.download_fileobj(self.buffer)
         self.buffer.seek(0)
         verify_checksum(self.buffer, self.checksum_sha256)
         return self.buffer.__enter__()

slicedimage/io/_base.py

@@ -329,6 +329,7 @@ def _parse_collection(parse_method, baseurl, backend_config):
     """
     def parse(name_relative_path_or_url_tuple):
         name, relative_path_or_url = name_relative_path_or_url_tuple
+
         partition = parse_method(relative_path_or_url, baseurl, backend_config)
         partition._name_or_url = relative_path_or_url