We are using agent_ipmitool with an Ironic Python Agent ramdisk to deploy images using a standalone Ironic (kilo) deployment. Around 60-70% of the time, when we are performing a deployment using 'ironic node-set-provision-state <node_uuid> active', the image is written to disk and reboots successfully. However, sometimes after the image deployment is finished, and the node is rebooted, we get 'deploy failed' and the following error:
Error rebooting node <node_uuid>. Error: Failed to set node power state to power on.
Even from this 'deploy failed' state, manually powering on the node and booting to local disk shows that the image has been written successfully. The attached log states "Not going to change_node_power_state because current state = requested state = 'power off'", but then immediately fails with an Asynchronous exception where it fails to turn the power back on.
It seems that Conductor isn't giving the node enough time to power off before attempting to power it on again, so the state transition somehow fails.
I should mention that the environment Ironic is running on is Suse Linux Enterprise Server (SLES) 12.