redfishpower: drop native "cycle" support

[chu11]

Consider if redfishpower should drop native "cycle" support.

Unlike power on, we cannot "wait until on" because with power cycle b/c it is racy. You don't know if the "on" you got is before or after the cycle happened.

Normally no big deal, this is the nature of things, and we support it anyways, just like with ipmipower.

BUT with parent/hierarchy support via #81, this is more problematic. pm --on parentNode,childNode can work, pm --off parentNode,childNode can work, but we just have to outright just not allow pm --cycle parentNode,childNode b/c we can't trust if the parent is actually on after it has been cycled.

Note that "cycle" can still be supported at the powerman level, which will simply define cycle as "off" followed by "on".

[garlick]

Unlike power on, we cannot "wait until on" because with power cycle b/c it is racy. You don't know if the "on" you got is before or after the cycle happened.

I'm not sure I follow. What is the sequence of events that is racy?

[chu11]

I'm not sure I follow. What is the sequence of events that is racy?

• you send power cycle to node, get response
• node starts up process to power cycle itself, this can sometimes take a second to start
• you send power status to node, get "on"

is the status "on" before the power cycle began (i.e. before it was off) or after it was "off" and it is now back "on"?

Remember I'm referring to "native" power cycle of redfish. Powerman may define a "cycle" as an "off" followed by an "on".

For example in redfishpower-supermicro.dev

		expect "redfishpower> "
		send "setonpath redfish/v1/Systems/1/Actions/ComputerSystem.Reset {\"ResetType\":\"On\"}\n"
		expect "redfishpower> "
		send "setoffpath redfish/v1/Systems/1/Actions/ComputerSystem.Reset {\"ResetType\":\"ForceOff\"}\n"
		expect "redfishpower> "
		send "setcyclepath redfish/v1/Systems/1/Actions/ComputerSystem.Reset {\"ResetType\":\"ForceRestart\"}\n"

b/c native "cycle" is implemented, we do

	script cycle_ranged {
		send "cycle %s\n"
		expect "redfishpower> "
	}

vs in cray-windom, native cycle isn't supported so

	script cycle_ranged {
		send "off %s\n"
		expect "redfishpower> "
		send "on %s\n"
		expect "redfishpower> "
	}

Edit: FWIW, I now see that we made cycle_ranged off and then on in ipmipower.dev, b/c of this exact same reason. ipmipower still supports cycle natively, b/c why not, that's in a different project. But since redfishpower lives inside powerman, the need to support it natively is not as strong.

[garlick]

Oh sorry, I didn't realize you meant the native redfish protocol. I thought you meant in the redfish program.

Gotcha! Seems fine to just do this in the device script then (off, delay, on).

Thanks for taking the time to explain!

[chu11]

yeah, i think in device scripts we should just do off, delay, on. But if we do that, we don't need all the cycle stuff in redfishpower anymore (i.e. "setcyclepath", "cycle %s", etc.). Then all future code can be cleaner too, like in my hierarchy support.

Issue is we've released device files with "cycle" in them before. Should redfishpower still support all the cycle stuff (setcyclepath, etc.) for backwards compatibility?