oxidecomputer/propolis

Panic count metrics could possibly be combined for easier analysis

Opened this issue · 0 comments

Propolis maintains two metrics counting panics in the guest, and distinguishes between guest- and host-handled panics as different timeseries. Here:

struct PvPanicGuestHandled {

and here:

struct PvPanicHostHandled {

We may want to collapse these into one timeseries counting total panics, with a field to indicate who handled it (guest or host). But it's not clear from the PVPANIC documentation whether these two are mutually exclusive. If not, two timeseries would be needed. But if so, we may want to collapse them.