Always set timeout on requests

If you don't set a timeout, your request may hang indefinitely.

The timeout parameter is used for connect and read timeouts.  The whole request
may take longer than the timeout; but if the connection takes longer than the
timeout, or if the connection fails to send more data for an interval longer than
the timeout, it will fail.

I chose 5 seconds for the upload_file, which (at least when run on replicate
infrastructure) is always to an endpoint in the same data center.

I chose 10 seconds for downloading URLFile inputs as they may be making a
request over the public internet.

Co-authored-by: Philip Potter <[email protected]>

python/cog/files.py

@@ -15,7 +15,7 @@ def upload_file(fh: io.IOBase, output_file_prefix: Optional[str] = None) -> str:
     if output_file_prefix is not None:
         name = getattr(fh, "name", "output")
         url = output_file_prefix + os.path.basename(name)
-        resp = requests.put(url, files={"file": fh}, timeout=None)
+        resp = requests.put(url, files={"file": fh}, timeout=5)
         resp.raise_for_status()
         return url
 

python/cog/types.py

@@ -336,7 +336,7 @@ def __wrapped__(self) -> Any:
         except AttributeError:
             pass
         url = object.__getattribute__(self, "__url__")
-        resp = requests.get(url, stream=True, timeout=None)
+        resp = requests.get(url, stream=True, timeout=10)
         resp.raise_for_status()
         resp.raw.decode_content = True
         object.__setattr__(self, "__target__", resp.raw)