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

[Arjen de Korte]

Citeren Marco Chiappero <marco op absence.it>:

>> When you read the corresponding values from the summary page, the  
>> values are reversed however. If we would parse the summary page  
>> between these alarms (be it right after parsing an alarm or just  
>> before receiving the next), with every new incoming alarm, the  
>> previous status would be reset again. So if we want to keep the  
>> alarms as they are, we should disregard the values set by the  
>> summary page. However, since the above alarm bits are never cleared  
>> by an alarm message after the test, this would mean that pressing  
>> this test button once would effectively permanently disable the  
>> outlet.
> Well, the test is supposed to perform a shutdown, so it's assumed  
> that NUT is restarted. I see no problem in this behaviour.

There is no guarantee that the machine where the driver is running on  
is also powered by the UPS. Most systems administrators like NUT  
because it allows you to centralize UPS monitoring from a single  
machine. Allowing the above would mean that each NMC would require a  
dedicated machine to run the driver on. This pretty much defeats the  
whole purpose of NUT.

> I compiled your code yesterday evening, everything looks fine but I  
> noticed that every 180 secs a new subscription is made: is there a  
> reason for this?

The only reason could be that the driver is not getting alarm  
messages. The driver should periodically receive empty alarm messages  
from the NMC. If it doesn't, it assumes the connection is somehow lost  
and subscribes again. Do you see the empty alarm messages if you run  
the driver with -DD enabled?

> And I see that +20-25% more traffic generated.

This could be the case since we poll both the summary and get_object  
pages for each pollinterval. If this interval is short, you may notice  
an increase compared to the previous code non-subscribed mode. The  
only reason for adding the NSM subscribed mode, is to allow increasing  
the pollinterval. If you keep it at something like 5 seconds, there is  
no point in NSM subscriptions.

> Thinking a little bit, I would just read alarm messages and the  
> get_object page: when everything is fine extended data is updated  
> once every pollinterval and the relevant one immediately by the  
> alarm messages, when the UPS goes on battery we can scan the summary  
> page too, frequently.

We're only reading the summary page because for some NMC models it  
contains info the get_object page doesn't (environmental info mostly,  
but some other things as well). Previously we would poll the summary  
page only one out of ten calls. This is sufficient if the pollinterval  
is short, but now that we poll much less frequently (pollinterval =  
64), we should poll both at the same time.

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