null_resource SSH provisioner hangs on interrupted connections
null_resource provisioners (file, remote-exec, chef, puppet) can hang indefinitely — sometimes for hours — when the SSH connection silently drops mid-provision, with no error until a TCP timeout eventually fires. The hang often survives SSH keepalive settings because keepalive only detects a dead peer, not a black-holed session, and some cloud networks silently sever connections a few minutes after instance creation. If apply hangs on a provisioner, check whether it eventually dies with a dial/getsockopt timeout, and retry terraform apply — reruns often succeed because the connection path differs on second attempt. Workaround: replace null_resource SSH provisioning with local-exec running scp/ssh directly, which surfaces real connection errors instead of hanging.