Gate no-HEAD download path on response_checksum_validation=when_required

Address reviewer feedback (SDKs durability team): HEAD requests are
required by default for full-object checksum validation on downloads.
Only skip the HEAD when the caller has already opted out of response
checksum validation via botocore's `response_checksum_validation =
when_required` config. The prior HEAD + size-based GET logic is restored
as the default path.

This conditional is temporary; it can be removed once S3 supports
returning checksums for ranged GETs (tracking: P320750170).

Also:
- Add TestDownloadWhenRequiredChecksumValidation covering the HEAD-less
  path for non-empty and 0-byte objects (InvalidRange handling)
- Fix ruff import ordering / formatting in time-batch-download.py
  flagged by the Lint CI action

Author: crowecawcaw

.changes/next-release/feature-download-85557.json

@@ -0,0 +1,5 @@
+{
+  "type": "feature",
+  "category": "download",
+  "description": "Skip the HEAD request during downloads when the client is configured with ``response_checksum_validation='when_required'``, reducing latency for small-object transfers. The HEAD request remains in place by default to enable full-object checksum validation until S3 supports returning checksums for ranged GETs."
+}

s3transfer/download.py

@@ -356,15 +356,162 @@ def _submit(
             transfer_future, osutil
         )(osutil, self._transfer_coordinator, io_executor)
 
-        self._submit_first_chunk_request(
-            client,
-            config,
-            osutil,
+        # Skip the HEAD request only when the caller has explicitly opted out
+        # of response checksum validation. Otherwise we need the HEAD to
+        # obtain the full-object ETag/size for checksum validation. This
+        # conditional can be removed once S3 supports returning checksums
+        # for ranged GETs (tracking: P320750170).
+        if client.meta.config.response_checksum_validation == "when_required":
+            self._submit_first_chunk_request(
+                client,
+                config,
+                osutil,
+                request_executor,
+                io_executor,
+                download_output_manager,
+                transfer_future,
+                bandwidth_limiter,
+            )
+            return
+
+        if (
+            transfer_future.meta.size is None
+            or transfer_future.meta.etag is None
+        ):
+            response = client.head_object(
+                Bucket=transfer_future.meta.call_args.bucket,
+                Key=transfer_future.meta.call_args.key,
+                **transfer_future.meta.call_args.extra_args,
+            )
+            transfer_future.meta.provide_transfer_size(
+                response['ContentLength']
+            )
+            transfer_future.meta.provide_object_etag(response.get('ETag'))
+
+        if transfer_future.meta.size < config.multipart_threshold:
+            self._submit_download_request(
+                client,
+                config,
+                osutil,
+                request_executor,
+                io_executor,
+                download_output_manager,
+                transfer_future,
+                bandwidth_limiter,
+            )
+        else:
+            self._submit_ranged_download_request(
+                client,
+                config,
+                osutil,
+                request_executor,
+                io_executor,
+                download_output_manager,
+                transfer_future,
+                bandwidth_limiter,
+            )
+
+    def _submit_download_request(
+        self,
+        client,
+        config,
+        osutil,
+        request_executor,
+        io_executor,
+        download_output_manager,
+        transfer_future,
+        bandwidth_limiter,
+    ):
+        call_args = transfer_future.meta.call_args
+        fileobj = download_output_manager.get_fileobj_for_io_writes(
+            transfer_future
+        )
+        progress_callbacks = get_callbacks(transfer_future, 'progress')
+        get_object_tag = download_output_manager.get_download_task_tag()
+        final_task = download_output_manager.get_final_io_task()
+        self._transfer_coordinator.submit(
             request_executor,
-            io_executor,
-            download_output_manager,
-            transfer_future,
-            bandwidth_limiter,
+            ImmediatelyWriteIOGetObjectTask(
+                transfer_coordinator=self._transfer_coordinator,
+                main_kwargs={
+                    'client': client,
+                    'bucket': call_args.bucket,
+                    'key': call_args.key,
+                    'fileobj': fileobj,
+                    'extra_args': call_args.extra_args,
+                    'callbacks': progress_callbacks,
+                    'max_attempts': config.num_download_attempts,
+                    'download_output_manager': download_output_manager,
+                    'io_chunksize': config.io_chunksize,
+                    'bandwidth_limiter': bandwidth_limiter,
+                },
+                done_callbacks=[final_task],
+            ),
+            tag=get_object_tag,
+        )
+
+    def _submit_ranged_download_request(
+        self,
+        client,
+        config,
+        osutil,
+        request_executor,
+        io_executor,
+        download_output_manager,
+        transfer_future,
+        bandwidth_limiter,
+    ):
+        call_args = transfer_future.meta.call_args
+        progress_callbacks = get_callbacks(transfer_future, 'progress')
+        fileobj = download_output_manager.get_fileobj_for_io_writes(
+            transfer_future
+        )
+        part_size = config.multipart_chunksize
+        num_parts = calculate_num_parts(transfer_future.meta.size, part_size)
+        get_object_tag = download_output_manager.get_download_task_tag()
+        finalize_download_invoker = CountCallbackInvoker(
+            self._get_final_io_task_submission_callback(
+                download_output_manager, io_executor
+            )
+        )
+        for i in range(num_parts):
+            range_parameter = calculate_range_parameter(
+                part_size, i, num_parts
+            )
+            extra_args = {'Range': range_parameter}
+            if transfer_future.meta.etag is not None:
+                extra_args['IfMatch'] = transfer_future.meta.etag
+            extra_args.update(call_args.extra_args)
+            finalize_download_invoker.increment()
+            self._transfer_coordinator.submit(
+                request_executor,
+                GetObjectTask(
+                    transfer_coordinator=self._transfer_coordinator,
+                    main_kwargs={
+                        'client': client,
+                        'bucket': call_args.bucket,
+                        'key': call_args.key,
+                        'fileobj': fileobj,
+                        'extra_args': extra_args,
+                        'callbacks': progress_callbacks,
+                        'max_attempts': config.num_download_attempts,
+                        'start_index': i * part_size,
+                        'download_output_manager': download_output_manager,
+                        'io_chunksize': config.io_chunksize,
+                        'bandwidth_limiter': bandwidth_limiter,
+                    },
+                    done_callbacks=[finalize_download_invoker.decrement],
+                ),
+                tag=get_object_tag,
+            )
+        finalize_download_invoker.finalize()
+
+    def _get_final_io_task_submission_callback(
+        self, download_manager, io_executor
+    ):
+        final_task = download_manager.get_final_io_task()
+        return FunctionContainer(
+            self._transfer_coordinator.submit, io_executor, final_task
         )
 
     def _submit_first_chunk_request(

scripts/performance/time-batch-download.py

@@ -2,10 +2,12 @@
 """Direct timing of batch downloads without shell wrapper."""
 
 import argparse
-import tempfile
 import shutil
+import tempfile
 import time
+
 from botocore.session import get_session
+
 from s3transfer.manager import TransferManager
 
 
@@ -20,13 +22,13 @@ def main():
     parser.add_argument('--file-size', type=int, required=True)
     parser.add_argument('--s3-bucket', required=True)
     args = parser.parse_args()
-    
+
     session = get_session()
     client = session.create_client('s3')
-    
+
     tempdir = tempfile.mkdtemp()
     s3_keys = []
-    
+
     try:
         # Upload files
         print(f"Uploading {args.file_count} files...")
@@ -37,7 +39,7 @@ def main():
                 s3_key = f"perf_test_{i}"
                 manager.upload(file_path, args.s3_bucket, s3_key)
                 s3_keys.append(s3_key)
-        
+
         # Download files
         print(f"Downloading {args.file_count} files...")
         start_time = time.time()
@@ -46,9 +48,9 @@ def main():
                 download_path = f"{tempdir}/download_{i}"
                 manager.download(args.s3_bucket, s3_key, download_path)
         duration = time.time() - start_time
-        
+
         print(f"Download duration: {duration:.2f} seconds")
-        
+
         # Cleanup
         for s3_key in s3_keys:
             client.delete_object(Bucket=args.s3_bucket, Key=s3_key)