Ironic nova driver fails to setup initial state correctly

[Robert Collins]

2014-06-05 04:04:54.552 28915 ERROR ironic.nova.virt.ironic.driver [req-66403d15-5f7e-4a59-8d3d-ba9d6e654fb5 None] Failed to request Ironic to provision instance ef3421ef-e7b3-4203-811c-dad052b9badf: RPC do_node_deploy called for cfa5c267-3a7c-4973-bdcf-80a139a947ea, but provision state is already deploy failed. (HTTP 500)

This happened because the node wasn't 'properly' cleaned after the last instance_uuid was removed from it. Seems to me that the ironic nova driver should not make any assumptions - just its instance_uuid atomically, and then reset all the state, and finally proceed to set the state it wants for deployment.

[Dmitry Tantsur (divius)]

Just a guess: if spawn() fails, driver calls _cleanup_deploy which removes instance_uuid, but nothing else.

[aeva black (tenbrae)]

I suspect there are other situations which can result in this condition, too. Moving to instance_info should alleviate this problem.

https://github.com/openstack/ironic-specs/blob/master/specs/juno/add-node-instance-info.rst

[Michael Davies (mrda)]

Devananda thinks this change might (https://review.openstack.org/#/c/101864/) might resolve this bug. I'll try and verify this.

[Dmitry Tantsur (divius)]

Hi Michael, any updates on this bug? I'm moving it Nova since driver is now there.

[Jim Rollenhagen (jim-rollenhagen)]

This has been fixed for a while; we only expose resources available to a node in AVAILABLE/NONE provision state and with no instance uuid: https://github.com/openstack/nova/blob/master/nova/virt/ironic/driver.py#L301-L318