On 10/19/11 1:57 AM, Benny Lofgren wrote:
On 2011-10-19 10.23, Paul de Weerd wrote:
| I think your methodology is fllawed. think of the situations when you
| have power loss, then shutdown is started and then power is back.
| or situations where you starting machine after blackout and then there
| is a blackout again...
| With good ups you at least have 'switch off after some time is gone"
| option.
You can build a lot of logic in the way you do this, including 'switch
off after some time is gone'. Really the only thing you get with
'good ups' is an indication of how long your battery is going to last,
which might even resemble something close to reality if you're lucky.
Don't get me wrong: 'proper' upses have a lot of benefits, but that's
mostly related to the ease of doing this controlled power down in case
of blackouts.
Well, Gregory is right in a way. The one flaw there is with my "poor man's
UPS watchdog" is that there is no way to get the server going again if
power is restored after the script decides to shut the server down but
before the UPS runs out of battery juice and actually shuts off the
power to the server.
In that case, when power is restored the server will never have had its
power cycled, so can never turn back on again even if you set its bios to
boot when power is applied regardless of its state before power outage.
What we can do in that case is to "almost" power it off, that is, shut
down all services, get down to single user mode, unmount all volumes
except for root which is remounted read-only, and then just wait.
Either the server will eventually die, in which case it will boot back up
as good as new when power is next applied, or the power will get back on
in which case the script can detect that and simply do a reboot.
Regards,
/Benny
I like Benny's concept - go down to single user mode and wait... simple.
It is also a good solution for a remote setup.
Mehma
