Re: [Nut-upsuser] confusion about shutdown

[Jim Klimov via Nut-upsuser]

> So I'll have a look at this file being removed on upsd startup, or in our
rc.d scripts for now.

Note: it *is* removed upon `upsmon` startup (any successful one, except for
sending signals `upsmon -c CMD` or querying this file `upsmon -K`).
Also checked that this is a rare path not hardcoded into binaries ;) and
only coming from the config file. Some older scripts and doc examples that
referenced `/etc/killpower` were changed by that PR to use `upsmon -K`
(which assumes the binary and its SO requirements are available) or the
`@POWERDOWNFLAG@` substitution from `configure`'d default instead of exact
string.

Jim


On Fri, Jan 6, 2023 at 2:40 PM Greg Troxel  wrote:

> On 1/5/23 3:04 PM, Jim Klimov wrote:
> > https://github.com/networkupstools/nut/pull/1762
> >  (and maybe some of
> > other recent PRs) updated quite a bit here and there, including a
> > configure-time option for default POWERDOWNFLAG value (using legacy
> > default still).
> >
> > Distros now have easier time to put it into a tmpfs of their choice that
> > they know will remain mounted.
>
> Thanks - that is helpful.
>
> In general, OS shutdown is tricky business before we add in killpower,
> and with killpower with no delay it is almost impossibly tricky.   I'd
> be a bit scared of assuming ramdisks are still there.
>
> It looks like the FreeBSD port assumes offdelay is long enough (and that
> the UPS will have an offdelay).  It may be that essentially all units
> really do have a delay, and this is ok.
>
> In NetBSD, ramdisks w/o device nodes are unmounted just before swap is
> deconfigured, which is before the place where critical filesystems would
> be made ro.
>
> So I'll have a look at this file being removed on upsd startup, or in
> our rc.d scripts for now.
>
>
___
Nut-upsuser mailing list
Nut-upsuser@alioth-lists.debian.net
https://alioth-lists.debian.net/cgi-bin/mailman/listinfo/nut-upsuser

Re: [Nut-upsuser] confusion about shutdown

[Greg Troxel]

On 1/5/23 3:04 PM, Jim Klimov wrote:
https://github.com/networkupstools/nut/pull/1762 
 (and maybe some of 
other recent PRs) updated quite a bit here and there, including a 
configure-time option for default POWERDOWNFLAG value (using legacy 
default still).


Distros now have easier time to put it into a tmpfs of their choice that 
they know will remain mounted.


Thanks - that is helpful.

In general, OS shutdown is tricky business before we add in killpower, 
and with killpower with no delay it is almost impossibly tricky.   I'd 
be a bit scared of assuming ramdisks are still there.


It looks like the FreeBSD port assumes offdelay is long enough (and that 
the UPS will have an offdelay).  It may be that essentially all units 
really do have a delay, and this is ok.


In NetBSD, ramdisks w/o device nodes are unmounted just before swap is 
deconfigured, which is before the place where critical filesystems would 
be made ro.


So I'll have a look at this file being removed on upsd startup, or in 
our rc.d scripts for now.



___
Nut-upsuser mailing list
Nut-upsuser@alioth-lists.debian.net
https://alioth-lists.debian.net/cgi-bin/mailman/listinfo/nut-upsuser

Re: [Nut-upsuser] confusion about shutdown

[Jim Klimov via Nut-upsuser]

https://github.com/networkupstools/nut/pull/1762 (and maybe some of other
recent PRs) updated quite a bit here and there, including a configure-time
option for default POWERDOWNFLAG value (using legacy default still).

Distros now have easier time to put it into a tmpfs of their choice that
they know will remain mounted.

Jim

On Thu, Jan 5, 2023, 19:03 Greg Troxel  wrote:

> Charles Lepple  writes:
>
> > On Dec 30, 2022, at 7:42 PM, Greg Troxel  wrote:
> >>
> >> maybe me, maybe docs.
> >>
> >> I have never had auto shutdown set up on netbsd.  Mostly I monitor to
> >> mqtt.  But now I'm trying and have multiple confusions.
> >>
> >> 1) 'upsdrvctl shutdown' vs 'upscmd shutdown.return'
> >>
> >> upsdrvctl is about stopping the driver, but lowbatt shutdown wants to
> >> send a command.  I don't get the upsmon example file:
> >
> > "upsdrvctl stop" is for stopping the driver, but "upsdrvctl shutdown"
> > runs "$DRIVER -k" to kill power to the load on the UPS (which is
> > actually a second, non-long-running instance of the driver that
> > doesn't talk to upsd, after the rest of the system is ready to stop).
> >
> > That being said, upscmd needs to talk to upsd, which needs a running
> > driver - IMHO this is only workable if you trust your
> > delay-before-shutdown in the UPS. The "upsdrvctl shutdown" case can
> > often be run after everything has been remounted read-only, at which
> > point it is okay if there is no delay before the power gets pulled.
>
> Thanks.  I am editing docs for a soon PR.  Checking my understanding:
>
> psdrvctl stop will not affect the UPS, and upsdrvctl shutdown will spin
> up $driver -k, assuming that the previous long-running driver is no
> longer running.
>
> And thus one does not use upscmd to do a shutdown, because it needs
> upsd/driver running and in getting to fs being ro, those will have been
> stopped.
>
> And, one can either:
>
>   do shutdown when upsd/driver/upsmon is stopped and hope delay is long
>   enough
>
> or
>
>   have another script that runs essentially last, after the fs are ro,
>   so even if no delay it's ok
>
> > Hopefully explained by the above. Also, you don't have to check for the
> file directly - you can use "upsmon -K" as in the following excerpt from
> the rc.d/nut script in FreeBSD's NUT port:
> >
> > nut_poststop() {
> > if ${nut_prefix}/sbin/upsdrvctl stop && checkyesno nut_upsshut;
> then
> > if ${nut_prefix}/sbin/upsmon -K; then
> > ${nut_prefix}/sbin/upsdrvctl shutdown
> > fi
> > fi
> >
> > }
>
> got it about -K
>
> This assumes delay, or is that somehow post remount ro?
>
> >> 2) LB config.   On a Best Fortress, I can set lowbatt runtime, but only
> >> 3 digits, even though I want 1800s.   Is this likely a Fortress
> >> limitation, or a bug?  I will read the source...
> >
> > Looks like it might be a NUT bug. There should be room for four digits
> > in the protocol, but the protocol appears to be minutes. NUT shouldn't
> > limit the seconds field to three characters.
>
> Thanks, will read the printed protocol spec I have someplace.
>
> >> 2A) seems like nut should be able to have time-based config to start
> >> shutdown, all of batt%, seconds remaining, and seconds on battery, built
> >> in.
> >
> > Others have covered solutions in greater depth (I think Roger's config
> > guide handles most of these), but IMHO things like "seconds on
> > battery" aren't as easy as they look (unless you don't care about the
> > intersection with seconds remaining).
>
> Sure, I want:
>
>   if on batt for 2min, start shutdown
>
> and I'm fine with
>
>   if LB, start shutdown
>
> also firing.  Basically one UPS has a desktop (piggy, not important to
> stay up, important not to break) and an RPI/switch/wifi (good to keep
> running).
>
> I have queued the config examples pdf for reading.
>
> >> 3) seems like shutdown should remove /etc/killpower but maybe upsmon
> >> removes it at boot.  should be explained more, probably, guessing it is
> >> at boot since fs is ro
> >
> > probably should be removed at boot (or better yet, placed on a RAM-based
> filesystem), but this is not my area of expertise.
>
> /etc is ~never RAM-based, so we need to move to /var/run or equiv, but I
> think removing it at startup is easy and maybe already done.  I will
> look.
>
> ___
> Nut-upsuser mailing list
> Nut-upsuser@alioth-lists.debian.net
> https://alioth-lists.debian.net/cgi-bin/mailman/listinfo/nut-upsuser
>
___
Nut-upsuser mailing list
Nut-upsuser@alioth-lists.debian.net
https://alioth-lists.debian.net/cgi-bin/mailman/listinfo/nut-upsuser

Re: [Nut-upsuser] confusion about shutdown

[Greg Troxel]

Charles Lepple  writes:

> On Dec 30, 2022, at 7:42 PM, Greg Troxel  wrote:
>> 
>> maybe me, maybe docs.
>> 
>> I have never had auto shutdown set up on netbsd.  Mostly I monitor to
>> mqtt.  But now I'm trying and have multiple confusions.
>> 
>> 1) 'upsdrvctl shutdown' vs 'upscmd shutdown.return'
>> 
>> upsdrvctl is about stopping the driver, but lowbatt shutdown wants to
>> send a command.  I don't get the upsmon example file:
>
> "upsdrvctl stop" is for stopping the driver, but "upsdrvctl shutdown"
> runs "$DRIVER -k" to kill power to the load on the UPS (which is
> actually a second, non-long-running instance of the driver that
> doesn't talk to upsd, after the rest of the system is ready to stop).
>
> That being said, upscmd needs to talk to upsd, which needs a running
> driver - IMHO this is only workable if you trust your
> delay-before-shutdown in the UPS. The "upsdrvctl shutdown" case can
> often be run after everything has been remounted read-only, at which
> point it is okay if there is no delay before the power gets pulled.

Thanks.  I am editing docs for a soon PR.  Checking my understanding:

psdrvctl stop will not affect the UPS, and upsdrvctl shutdown will spin
up $driver -k, assuming that the previous long-running driver is no
longer running.

And thus one does not use upscmd to do a shutdown, because it needs
upsd/driver running and in getting to fs being ro, those will have been
stopped.

And, one can either:

  do shutdown when upsd/driver/upsmon is stopped and hope delay is long
  enough

or

  have another script that runs essentially last, after the fs are ro,
  so even if no delay it's ok

> Hopefully explained by the above. Also, you don't have to check for the file 
> directly - you can use "upsmon -K" as in the following excerpt from the 
> rc.d/nut script in FreeBSD's NUT port:
>
> nut_poststop() {
> if ${nut_prefix}/sbin/upsdrvctl stop && checkyesno nut_upsshut; then
> if ${nut_prefix}/sbin/upsmon -K; then
> ${nut_prefix}/sbin/upsdrvctl shutdown
> fi
> fi
>
> }

got it about -K

This assumes delay, or is that somehow post remount ro?

>> 2) LB config.   On a Best Fortress, I can set lowbatt runtime, but only
>> 3 digits, even though I want 1800s.   Is this likely a Fortress
>> limitation, or a bug?  I will read the source...
>
> Looks like it might be a NUT bug. There should be room for four digits
> in the protocol, but the protocol appears to be minutes. NUT shouldn't
> limit the seconds field to three characters.

Thanks, will read the printed protocol spec I have someplace.

>> 2A) seems like nut should be able to have time-based config to start
>> shutdown, all of batt%, seconds remaining, and seconds on battery, built
>> in.
>
> Others have covered solutions in greater depth (I think Roger's config
> guide handles most of these), but IMHO things like "seconds on
> battery" aren't as easy as they look (unless you don't care about the
> intersection with seconds remaining).

Sure, I want:

  if on batt for 2min, start shutdown

and I'm fine with

  if LB, start shutdown

also firing.  Basically one UPS has a desktop (piggy, not important to
stay up, important not to break) and an RPI/switch/wifi (good to keep
running).

I have queued the config examples pdf for reading.

>> 3) seems like shutdown should remove /etc/killpower but maybe upsmon
>> removes it at boot.  should be explained more, probably, guessing it is
>> at boot since fs is ro
>
> probably should be removed at boot (or better yet, placed on a RAM-based 
> filesystem), but this is not my area of expertise.

/etc is ~never RAM-based, so we need to move to /var/run or equiv, but I
think removing it at startup is easy and maybe already done.  I will
look.

___
Nut-upsuser mailing list
Nut-upsuser@alioth-lists.debian.net
https://alioth-lists.debian.net/cgi-bin/mailman/listinfo/nut-upsuser

Re: [Nut-upsuser] confusion about shutdown

[Charles Lepple via Nut-upsuser]

On Dec 30, 2022, at 7:42 PM, Greg Troxel  wrote:
> 
> maybe me, maybe docs.
> 
> I have never had auto shutdown set up on netbsd.  Mostly I monitor to
> mqtt.  But now I'm trying and have multiple confusions.
> 
> 1) 'upsdrvctl shutdown' vs 'upscmd shutdown.return'
> 
> upsdrvctl is about stopping the driver, but lowbatt shutdown wants to
> send a command.  I don't get the upsmon example file:

"upsdrvctl stop" is for stopping the driver, but "upsdrvctl shutdown" runs 
"$DRIVER -k" to kill power to the load on the UPS (which is actually a second, 
non-long-running instance of the driver that doesn't talk to upsd, after the 
rest of the system is ready to stop).

That being said, upscmd needs to talk to upsd, which needs a running driver - 
IMHO this is only workable if you trust your delay-before-shutdown in the UPS. 
The "upsdrvctl shutdown" case can often be run after everything has been 
remounted read-only, at which point it is okay if there is no delay before the 
power gets pulled.

> 
>  # POWERDOWNFLAG - Flag file for forcing UPS shutdown on the primary system
>  #
>  # upsmon will create a file with this name in primary mode when it's time
>  # to shut down the load.  You should check for this file's existence in
>  # your shutdown scripts and run 'upsdrvctl shutdown' if it exists, to tell
>  # the UPS(es) to power off.
> 
Hopefully explained by the above. Also, you don't have to check for the file 
directly - you can use "upsmon -K" as in the following excerpt from the 
rc.d/nut script in FreeBSD's NUT port:

nut_poststop() {
if ${nut_prefix}/sbin/upsdrvctl stop && checkyesno nut_upsshut; then
if ${nut_prefix}/sbin/upsmon -K; then
${nut_prefix}/sbin/upsdrvctl shutdown
fi
fi

}

> 
> 2) LB config.   On a Best Fortress, I can set lowbatt runtime, but only
> 3 digits, even though I want 1800s.   Is this likely a Fortress
> limitation, or a bug?  I will read the source...

Looks like it might be a NUT bug. There should be room for four digits in the 
protocol, but the protocol appears to be minutes. NUT shouldn't limit the 
seconds field to three characters.

> 2A) seems like nut should be able to have time-based config to start
> shutdown, all of batt%, seconds remaining, and seconds on battery, built
> in.

Others have covered solutions in greater depth (I think Roger's config guide 
handles most of these), but IMHO things like "seconds on battery" aren't as 
easy as they look (unless you don't care about the intersection with seconds 
remaining).

> 3) seems like shutdown should remove /etc/killpower but maybe upsmon
> removes it at boot.  should be explained more, probably, guessing it is
> at boot since fs is ro

probably should be removed at boot (or better yet, placed on a RAM-based 
filesystem), but this is not my area of expertise.
> 
> 4) not clear where I should edit in the instant command, as it doesn't
> belong when shutting down upsd, but yet it has to be before.  This is
> complicated.

probably covered by the nut_poststop() example above, but shout if you still 
have questions about that.

- Charles
___
Nut-upsuser mailing list
Nut-upsuser@alioth-lists.debian.net
https://alioth-lists.debian.net/cgi-bin/mailman/listinfo/nut-upsuser

[Nut-upsuser] confusion about shutdown

[Greg Troxel]

maybe me, maybe docs.

I have never had auto shutdown set up on netbsd.  Mostly I monitor to
mqtt.  But now I'm trying and have multiple confusions.

1) 'upsdrvctl shutdown' vs 'upscmd shutdown.return'

upsdrvctl is about stopping the driver, but lowbatt shutdown wants to
send a command.  I don't get the upsmon example file:

  # POWERDOWNFLAG - Flag file for forcing UPS shutdown on the primary system
  #
  # upsmon will create a file with this name in primary mode when it's time
  # to shut down the load.  You should check for this file's existence in
  # your shutdown scripts and run 'upsdrvctl shutdown' if it exists, to tell
  # the UPS(es) to power off.


2) LB config.   On a Best Fortress, I can set lowbatt runtime, but only
3 digits, even though I want 1800s.   Is this likely a Fortress
limitation, or a bug?  I will read the source...

2A) seems like nut should be able to have time-based config to start
shutdown, all of batt%, seconds remaining, and seconds on battery, built
in.

3) seems like shutdown should remove /etc/killpower but maybe upsmon
removes it at boot.  should be explained more, probably, guessing it is
at boot since fs is ro

4) not clear where I should edit in the instant command, as it doesn't
belong when shutting down upsd, but yet it has to be before.  This is
complicated.

Any working code on NetBSD?

___
Nut-upsuser mailing list
Nut-upsuser@alioth-lists.debian.net
https://alioth-lists.debian.net/cgi-bin/mailman/listinfo/nut-upsuser