soft poweroff can cause issues in rebooting some (buggy) BIOS's

[Dan Prince]

When using IPA I'm seeing issues on some of my machines (2 of them) which after a soft power off don't come back online. This occurs even though I've configured the machines to POWER ON after a power outage (the normal way you would configure machines that use an external power adapter). These machines both worked great with the previous non-IPA deployment ramdisk and Ironic conductor code.

I've tested the machines manually and soft powering them down (regardless of use of Ironic) requires you manually hit the power button, meaning an external power adapter is pretty much useless.

To be clear the soft power off I'm speaking of is the one in the reboot_and_finish_deploy function in agent_base_vendor.py.

This is clearly a BIOS vendor bug... but the code path here is new and because there is no way to disable the soft power off via Ironic. I'm working through the issue with the hardware vendor but wanted to report this in Ironic as well because the code path here is new (only an IPA deployment tries this). I'm wondering if adding a simple option to control this makes sense for Ironic? Perhaps a soft_poweroff_on_deploy_reboot = false option?

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/241745

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/291829

[OpenStack Infra (hudson-openstack)]

Change abandoned by Dan Prince (<email address hidden>) on branch: master
Review: https://review.openstack.org/241745

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/291829
Committed: https://git.openstack.org/cgit/openstack/ironic/commit/?id=824ad1676bd8032fb4a4eb8ffc7625a376a64371
Submitter: Jenkins
Branch: master

commit 824ad1676bd8032fb4a4eb8ffc7625a376a64371
Author: Lucas Alvares Gomes <email address hidden>
Date: Mon Mar 14 12:20:58 2016 +0000

    Agent: Out-of-band power off on deploy

    The patch is adding a new property called "deploy_forces_oob_reboot"
    for the Agent deployment methodology. If the property is specified in
    the node's driver_info with the value of True, Ironic will call IPA to
    flush the file system and then issue a out-of-band reboot at the end of
    the deployment to the node to boot into the user's image.

    Some hardware/firmware may have problems doing a power off in-band
    (usually faulty hardware/bad firmware, see the bug this commit is fixing)
    therefore we need Ironic to do it directly in the BMC.

    This patch also closes one more consistency gap between the IPA deploy
    model and the now deprecated bash ramdisk one which used the hard reboot
    to finish the deployment by default.

    Depends-On: I5cd1d1b821426e995dc584452494b93ab23917e0
    Closes-Bug: #1512492
    Change-Id: I5e389c5245826f86466aedfcb9730117e24b59db

[Thierry Carrez (ttx)]

This issue was fixed in the openstack/ironic 5.1.0 release.