We hit this issue in Rackspace where Ethernet Interface went down, and came back, but cloud uploads completely stalled. The app server process had to be restarted in order to resume cloud uploads. I was able to reproduce this scenario in my local environment using the following scenario:

1. Reduce the https.globalAgent.maxSockets to 1 for connection contention, and to monitor the uploads.
2. Wrote a node program to upload files to cloud in parallel. I started with 10 parallel uploads.
3. Uploads work one by one, without any issue.
4. Restart the program again, pull out the network cable after couple of uploads. Now the uploads stall.
5. Running a netstat showed an ESTABLISHED connection to the Rackspace IP. Ethernet is down, but the connection is always in this ESTABLISHED state. The connection did not break down even after 30 minutes.
6. Brought the network interface up. But cloud uploads never recover.

I am using linux mint here.

Any help would be appreciated.