[RFE] Soft reboot and NMI implementation for iRMC

[Daisuke Fujita]

Fujitsu PRIMERGY Customer requires to automate their deployment and management tasks everything through openstack api.
However Ironic iRMC driver doesn't allow customer to control Soft Reboot and NMI even though iRMC is capable of doing both Soft Reboot and NMI.

[Yuriy Zveryanskyy (yzveryanskyy)]

Ironic does not support soft reboot at all https://github.com/openstack/ironic/blob/master/ironic/drivers/base.py#L373-L376 It should be a generic feature, not only for iRMC.

[Naohiro Tamura (naohirot)]

Yuriy,

As I discussed this issue with rameshg87 in the ironic channel yesterday, 2015-05-28, I agreed to standardize the soft reboot and NMI interface for long term [1]. Actually I have already submitted the bp [2] and spec.

However I'm going fix this customer problem by vendor passthru for short term as discussed with rameshg87 [1]

[1] http://eavesdrop.openstack.org/irclogs/%23openstack-ironic/%23openstack-ironic.2015-05-28.log.html
[2] https://blueprints.launchpad.net/ironic/+spec/enhance-power-interface-for-soft-reboot-and-nmi
[3] https://review.openstack.org/#/c/186700/

Thanks.

Naohiro

---
http://eavesdrop.openstack.org/irclogs/%23openstack-ironic/%23openstack-ironic.2015-05-28.log.html

naohirot rameshg87: I have a basic question regarding power driver. 07:54
rameshg87 sure 07:54
naohirot rameshg87: currently Ironic reboot is hard reboot. 07:54
naohirot rameshg87: If customer typed "nova reboot", he/she expect it is soft reboot. 07:54
naohirot rameshg87: Have you discussed this inconsistency in Ironic ever? 07:55
rameshg87 naohirot: ah no 07:56
rameshg87 naohirot: how does it work for vms ? 07:56
rameshg87 naohirot: does it issue a soft reboot ? 07:56
rameshg87 naohirot: i think the main question is what happens if the soft reboot gets stuck (instance never reboots) 07:56
naohirot rameshg87: In nova cli, there is a option --hard, "nova reboot --hard" means hard reboot. 07:57
rameshg87 oh okay 07:57
naohirot rameshg87: and in Nova cli, no option means soft reboot. 07:57
rameshg87 naohirot: i always wished we do a soft power off during a reboot, wait for bare metal to be powered off polling the state 07:57
rameshg87 naohirot: wait for x seconds 07:58
rameshg87 naohirot: hard power off baremetal if it doesn't power off in x seconds 07:58
rameshg87 naohirot: and then power on the node 07:58
naohirot naohirot: I got a bug report from iRMC customer regarding this inconsistency. 07:58
rameshg87 in that way we should be able to reliably detect if instance ever rebooted 07:58
rameshg87 oh okay 07:58
rameshg87 naohirot: wdyt ? 07:59
naohirot rameshg87: And iRMC has graceful shutdown scci API, and I'm thinking to implement it in vendor paththrough 07:59
rameshg87 naohirot: i think it was discussed before 08:00
naohirot rameshg87: what does "wdyt" stand for? 08:00
rameshg87 what do you think - wdyt 08:00
rameshg87 :) 08:00
naohirot rameshg87: :) 08:00
naohirot rameshg87: then what was the conclusion at that time? 08:00
rameshg87 naohirot: yeah, may be try pursuing to make this a standard 08:01
rameshg87 naohirot: instead of a vendor passthru 08:01
rameshg87 there is enough interest i guess 08:01
rameshg87 i am +1 on introducing soft reboot in ironic :) 08:01
naohirot rameshg87: does ilo have gracefull shutdown API too? 08:01
rameshg87 yes 08:01
naohirot rameshg87: How about NMI API? 08:02
rameshg87 NMI ? 08:02
naohirot rameshg87: which stands for Non Markable Interrupt, 08:03
rameshg87 naohirot: no idea 08:03
naohirot rameshg87: out customer also would like to issue this API, which allow us to take kernel dump. 08:03
rameshg87 oh okay, got it 08:04
naohirot rameshg87: in order to investigate kernel problem. 08:04
rameshg87 i need to check. a...

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/187082

[Dmitry Tantsur (divius)]

Generic spec: http://specs.openstack.org/openstack/ironic-specs/specs/approved/enhance-power-interface-for-soft-reboot-and-nmi.html

[Jim Rollenhagen (jim-rollenhagen)]

This will be approved when the generic change is done.

[OpenStack Infra (hudson-openstack)]

Change abandoned by Dmitry Tantsur (<email address hidden>) on branch: master
Review: https://review.openstack.org/187082
Reason: This was implemented via new driver interfaces, so I think it can be abandoned.

[Hironori Shiina (shiina-hironori)]

The following patches were posted by associating with the same RFE for the generic implementation[1] and already merged.

https://review.openstack.org/#/c/216743/
https://review.openstack.org/#/c/348732/

We can close this RFE.

[1] https://bugs.launchpad.net/ironic/+bug/1526226

[Ruby Loo (rloo)]

Thanks for the update Hironori. I marked it as a duplicate of #1526226.

[Ruby Loo (rloo)]

I tagged this as rfe-approved, since the dup bug has been approved.