Comment 4 for bug 1568650

So the problem is just that as the proxy is reading from the backend object node - once it gets to the end it doesn't really know that the EOF coming up out of the socket means the object-server closed the connection prematurely. I guess the ResummingGetter stuff only works if the failure from the backend is initiated by the proxy (i.e. a timeout) if the socket just dies proxy thinks that's the end of the object (even if it's < content-length)

here's a download of an object using my slow_download script where I kill -9'd the servicing object server:

Sep 1 21:56:24 saio proxy-server: 127.0.0.1 127.0.0.1 01/Sep/2016/21/56/24 GET /v1/AUTH_test/test/delete.me HTTP/1.0 200 - python-swiftclient-3.0.1.dev72 AUTH_tk4f571562b... - 88739429 - tx527790b89a084b478e5e7-0057c8a388 - 127.9797 - - 1472766856.030810118 1472766984.010467052 0

And here's the same download using swift command line client:

Sep 1 21:58:48 saio proxy-server: 127.0.0.1 127.0.0.1 01/Sep/2016/21/58/48 GET /v1/AUTH_test/test/delete.me HTTP/1.0 200 - python-swiftclient-3.0.1.dev72 AUTH_tk4f571562b... - 183797100 - tx4b89ab1148bf495ca4f99-0057c8a496 - 2.7687 - - 1472767126.060199022 1472767128.828860998 0

It's possible this all gets easier to fix now that we're tracking backend bytes [2]

1. https://gist.github.com/clayg/2458a4a7e1451c75fbc5a63fcae11635
2. https://review.openstack.org/#/c/363316/