Question

Currently, I’m just serving files like this:

# view callable
def export(request):
    response = Response(content_type='application/csv')
    # use datetime in filename to avoid collisions
    f = open('/temp/XML_Export_%s.xml' % datetime.now(), 'r')
        # this is where I usually put stuff in the file
    response.app_iter = f
    response.headers['Content-Disposition'] = ("attachment; filename=Export.xml")
    return response

The problem with this is that I can’t close or, even better, delete the file after the response has been returned. The file gets orphaned. I can think of some hacky ways around this, but I’m hoping there’s a standard way out there somewhere. Any help would be awesome.

Answer 1

You do not want to set a file pointer as the app_iter. This will cause the WSGI server to read the file line by line (same as for line in file), which is typically not the most efficient way to control a file upload (imagine one character per line). Pyramid’s supported way of serving files is via pyramid.response.FileResponse. You can create one of these by passing a file object.

response = FileResponse('/some/path/to/a/file.txt')
response.headers['Content-Disposition'] = ...

Another option is to pass a file pointer to app_iter but wrap it in the pyramid.response.FileIter object, which will use a sane block size to avoid just reading the file line by line.

The WSGI specification has strict requirements that response iterators which contain a close method will be invoked at the end of the response. Thus setting response.app_iter = open(...) should not cause any memory leaks. Both FileResponse and FileIter also support a close method and will thus be cleaned up as expected.

As a minor update to this answer I thought I’d explain why FileResponse takes a file path and not a file pointer. The WSGI protocol provides servers an optional ability to provide an optimized mechanism for serving static files via environ['wsgi.file_wrapper']. FileResponse will automatically handle this if your WSGI server has provided that support. With this in mind, you find it to be a win to save your data to a tmpfile on a ramdisk and providing the FileResponse with the full path, instead of trying to pass a file pointer to FileIter.

http://docs.pylonsproject.org/projects/pyramid/en/1.4-branch/api/response.html#pyramid.response.FileResponse