feat: resume interrupted transfers

CHANGELOG.md

@@ -17,6 +17,10 @@ raise *"Multiple suitable storage providers found"* for any HTTP(S) URL.
   mtime read from `Content-Length` and `Last-Modified` response headers. Servers that do not
   support `HEAD` requests are handled gracefully (size and mtime default to 0). No checksum
   is available for generic URLs.
+- Resumable downloads: interrupted transfers are continued from where they left off using
+  HTTP `Range` requests (`206 Partial Content`). Servers that do not support range requests
+  fall back to a full re-download. Partial files are preserved across retries for
+  connection/timeout errors, and discarded on checksum mismatches or other errors.
 
 ### Removed
 

README.md

@@ -19,6 +19,7 @@ A Snakemake storage plugin for downloading files via HTTP with local caching, ch
 - **Checksum verification**: Automatically verifies checksums (from Zenodo API, data.pypsa.org manifests, or GCS object metadata)
 - **Rate limit handling**: Automatically respects Zenodo's rate limits using `X-RateLimit-*` headers with exponential backoff retry
 - **Concurrent download control**: Limits simultaneous downloads to prevent overwhelming servers
+- **Resumable downloads**: Interrupted transfers resume from where they left off using HTTP range requests
 - **Progress bars**: Shows download progress with tqdm
 - **Immutable URLs**: Returns mtime=0 for Zenodo and data.pypsa.org (persistent URLs); uses actual mtime for GCS and generic HTTP
 - **Environment variable support**: Configure via environment variables for CI/CD workflows
@@ -154,6 +155,7 @@ The plugin automatically:
 - Uses `X-RateLimit-Reset` to calculate wait time
 - Retries failed requests with exponential backoff (up to 5 attempts)
 - Handles transient errors: HTTP errors, timeouts, checksum mismatches, and network issues
+- Resumes interrupted downloads using `Range` requests where supported by the server
 
 ## URL Handling
 

src/snakemake_storage_plugin_cached_http/__init__.py

@@ -264,19 +264,23 @@ def _get_rate_limit_wait_time(self, headers: httpx.Headers) -> float | None:
         return wait_seconds
 
     @asynccontextmanager
-    async def httpr(self, method: str, url: str):
+    async def httpr(self, method: str, url: str, headers: dict | None = None):
         """
         HTTP request wrapper with rate limiting and exception logging.
 
         Args:
             method: HTTP method (e.g., "get", "post")
             url: URL to request
+            headers: Optional additional HTTP headers
 
         Yields:
             httpx.Response object
         """
         try:
-            async with self.client() as client, client.stream(method, url) as response:
+            async with (
+                self.client() as client,
+                client.stream(method, url, headers=headers) as response,
+            ):
                 wait_time = self._get_rate_limit_wait_time(response.headers)
                 if wait_time is not None:
                     logger.info(
@@ -728,20 +732,33 @@ async def managed_retrieve(self):
                     return
 
         try:
+            # Check for existing partial file to resume
+            offset = local_path.stat().st_size if local_path.exists() else 0
+            headers = {"Range": f"bytes={offset}-"} if offset > 0 else None
+
             # Download using a get request, rate limit errors are detected and raise
             # WorkflowError to trigger a retry
-            async with self.provider.httpr("get", query) as response:
-                if response.status_code != 200:
+            async with self.provider.httpr("get", query, headers=headers) as response:
+                if response.status_code == 206:
+                    # Server supports resume - append to existing partial file
+                    mode = "ab"
+                    logger.info(f"Resuming {filename} from byte {offset}")
+                elif response.status_code == 200:
+                    # Server doesn't support Range - discard partial and restart
+                    mode = "wb"
+                    offset = 0
+                else:
                     raise WorkflowError(
                         f"Failed to download: HTTP {response.status_code} ({query})"
                     )
 
-                total_size = int(response.headers.get("content-length", 0))
+                total_size = int(response.headers.get("content-length", 0)) + offset
 
                 # Download to local path with progress bar
-                with local_path.open(mode="wb") as f:
+                with local_path.open(mode=mode) as f:
                     with tqdm(
                         total=total_size,
+                        initial=offset,
                         unit="B",
                         unit_scale=True,
                         desc=filename,
@@ -758,7 +775,11 @@ async def managed_retrieve(self):
             if self.provider.cache:
                 self.provider.cache.put(query, local_path)
 
+        except (TimeoutError, ConnectionError, httpx.TransportError):
+            # Mid-transfer interruption - keep partial file for resume on next retry
+            raise
         except:
+            # Any other error (wrong checksum, HTTP error, unexpected) - delete and restart
             if local_path.exists():
                 local_path.unlink()
             raise

tests/test_download.py

@@ -8,11 +8,14 @@
 import logging
 import os
 import time
+from contextlib import asynccontextmanager
+from unittest.mock import MagicMock
 from urllib.parse import urlparse
 
 import pytest
 
 from snakemake_storage_plugin_cached_http import (
+    FileMetadata,
     StorageObject,
     StorageProvider,
     StorageProviderSettings,
@@ -339,3 +342,84 @@ async def test_cache_staleness_for_mutable_sources(
     downloaded_content = cached_path.read_bytes()
     assert b'"stale": true' not in downloaded_content
     assert downloaded_content == original_content
+
+
+def make_mock_httpr(content: bytes, fail_at: int | None):
+    """
+    Factory for a mock httpr context manager simulating a range-capable HTTP server.
+
+    Args:
+        content: The full file content to serve.
+        fail_at: If set, drop the connection after serving this many bytes on the
+                 first (non-Range) request, simulating a mid-transfer interruption.
+                 If None, serve the full content without interruption.
+    """
+
+    @asynccontextmanager
+    async def mock_httpr(method, request_url, headers=None):
+        range_header = headers.get("Range") if headers else None
+        mock_httpr.received_range_headers.append(range_header)
+
+        response = MagicMock()
+        if range_header is None:
+            response.status_code = 200
+            chunk = content
+            drop_at = fail_at
+        else:
+            response.status_code = 206
+            offset = int(range_header.removeprefix("bytes=").removesuffix("-"))
+            chunk = content[offset:]
+            drop_at = None
+
+        async def aiter_bytes(chunk_size=8192):
+            if drop_at is None:
+                yield chunk
+            else:
+                yield chunk[:drop_at]
+                raise ConnectionError("peer closed connection")
+
+        response.aiter_bytes = aiter_bytes
+        response.headers = {"content-length": str(len(chunk))}
+
+        yield response
+
+    mock_httpr.received_range_headers = []
+
+    return mock_httpr
+
+
+@pytest.mark.asyncio
+async def test_resume_on_partial_file(storage_provider, tmp_path):
+    """Test that downloads resume from partial files using HTTP Range requests."""
+    url = TEST_CONFIGS["zenodo"]["url"]
+    full_content = b'{"port": "test", "lat": 1.0, "lon": 2.0}'
+
+    fail_at = 10
+    mock_httpr = make_mock_httpr(full_content, fail_at=fail_at)
+    storage_provider.httpr = mock_httpr
+    storage_provider.cache = None
+
+    obj = StorageObject(
+        query=url,
+        keep_local=False,
+        retrieve=True,
+        provider=storage_provider,
+    )
+
+    local_path = tmp_path / "resume_test" / "file.json"
+    local_path.parent.mkdir(parents=True, exist_ok=True)
+    obj.local_path = lambda: local_path
+
+    async def mock_get_metadata(url):
+        return FileMetadata(checksum=None, size=0, mtime=0)
+
+    obj.provider.get_metadata = mock_get_metadata
+
+    await obj.managed_retrieve()
+
+    assert len(mock_httpr.received_range_headers) == 2
+    assert mock_httpr.received_range_headers[0] is None  # first attempt: no Range header
+    assert mock_httpr.received_range_headers[1] == f"bytes={fail_at}-"
+    assert local_path.read_bytes() == full_content
+
+