I have recently hit several bugs during testing refactored recipe parsing.
1. Corrupted netdevice names
If bond set up was left after config_only and nettestslave instance was
killed and restarted then inconsistent device names were read (e.g. in
bond mode 0 all devices had the same MAC address and t_bond0 was
incorrectly identified instead of e.g. eth0)
2. Bonding module is not loaded
If config_only run was performed with independent slaves running
continuously the netconfigs from previous run were used resulting in
inconsistent set of used netdevices (e.g. slave saw a bond device in used
netdevice types and therefore bonding module was not loaded)
This patch set resolves these problems and makes cleanup more robust and
I have tested the patchset on RHEL6.