9:58 a.m.
On Fri, Jun 9, 2017 at 1:46 PM, Dan Williams <dan.j.williams(a)intel.com> wrote:
On Fri, Jun 9, 2017 at 12:52 PM, Linda Knippers
<linda.knippers(a)hpe.com> wrote:
> If you try to combine the two, it's not clear what state you're getting
> because it's not obvious from ndctl whether it's static or dynamic data.
> I believe that's why you correctly didn't call the state flags
"health"
> data.
True, ok, we do already have the distinction that the "health"
sub-object is live data. So how about a middle ground? For the live
detail flags that indicate the similar data as the static health-state
flags from the NFIT let call them the same name. So it would look like
this:
{
"dev":"nmem0",
"health":{
"flag_failed_arm":true,
}
}
Where a monitor application can see that we started with the DIMM
armed and it went unarmed, but it only ever needs to worry about one
key name for the "armed" state.
...and to be clear the reverse case (failed at boot, but later becomes
armed) would look like this:
{
"dev":"nmem0",
"flag_failed_arm":true,
"health":{
"flag_failed_arm":false,
}
}
Where, we need to emit the updated flag in the live "health" case
because it is now different than the boot state.