Comment 8 for bug 1916482

I do not experience this issue when Using Openstack V and Ceph O, we have been using l7 mode for years now but we also separate the storage and management traffic into different network bridges with different MTU sizes. I'm don't think that a TCP keepalive packet will mess with HTTP chunking these are on different OSI levels in general, I would much rather believe you could run into a timeout in which the next chunk is expected or premature port closures. Apart from other feature such as HTTP keepalive and pipelining.
You are using glance 2.1 API with show_image_direct_url set to true I assume