[Nut-upsdev] MGE NMC and NutShutdownModule (and other stuff)

[Arjen de Korte]

Citeren Marco Chiappero <marco op absence.it>:

>> The problem is, that this flag will never be cleared by an alarm message.
> I can't see the problem, it's just for testing purpose, while  
> setting things up. Once done NUT must be restarted, if that schema  
> is used.

It's either we don't use the test buttons, or tell people they have to  
restart the driver when they use them. The first isn't needed in NUT  
in the first place (since we have other commands to test the setup)  
and the second poses a risk during normal operation. Maybe small, but  
it is there. People might accidentally lock up the driver (if they  
forget to restart it after pressing the test buttons) or we might lose  
critical alarms. Whatever we do, it isn't going to be transparent. In  
that case, we should opt for the 'least surprises' option and mention  
that the test buttons don't work.

> Need only to be mentioned in the man page.

Users don't read the documentation or the man pages. You know that  
first hand (no pun intended). :-)

> Alarm messages can't get lost on TCP (unless explicitly dropped).

Sure they do. With the earlier versions of the NMC 66102 you have to  
expect loss of the TCP connection anytime. This isn't a problem if you  
also listen to the UDP broadcasts, but we can't use those because we  
have no way to verify the authenticity of alarms. Anybody would be  
able to spoof them.

>> Could it be that the card is rebooting? If you're running multiple  
>> clients in parallel, this might trigger an NMC reset.
> Ok, I didn't noticed it at first: the NMC keeps resetting all the  
> subscription connections when using other clients along with NUT.

You can subscribe only run one NSM client per host (I told you that).

> However it started working like this after the netxml-ups upgrade.
> Had to stop NUT. I never experienced such issue before.

Your earlier changes would never deal with the loss of NSM  
subscription and instead just continue with the polled status.

> Ok, found... the netxml-ups driver keeps asking the get_object page  
> really often.

It will do that for every alarm message it receives. If nothing  
happens, this will be once every minute or so. But if it keeps getting  
disconnected, this will happen much more frequently.

>> How do you know the subscription is successful without running the  
>> driver in debug mode? Currently, the driver won't inform you about  
>> that.
> Network traffic sniffing (tcpdump in particular).

That only tells you what the NMC is telling, not if the driver  
understands it. The packets could just as well be to another client  
that is left running. This is why I asked to run the driver in debug  
mode. I don't care what the NMC is sending, we need to know if the  
driver understands it or not. You can only tell by running it in debug  
mode.

Best regards, Arjen
-- 
Please keep list traffic on the list