if you turn on qmp for local instances (eg: for hot-plugging locally)
right now this assumes a unique name (which is a timestamp that is appended to non-custom instance names)
i suppose the quick/easy fix is to allow the custom instance name w/the knowledge that you can just use one or use a tmp id of some kind
if rconfig.QMP {
// really horribly hack but we don't really know the pid until
// after we launch so use the last 4 of id assigned to
// instance
ts := strings.Split(rconfig.InstanceName, "-")
rts := strings.Split(ts[1], ".")[0]
last := "4" + rts[len(rts)-4:] // no 0 and <65k, hack
args = append(args, "-qmp tcp:localhost:"+last+",server,wait=off")
}
at least for onprem targets another method might be to generate a id beforehand and not overload the instance name - I don't know how much we rely on the name for other providers (such as gcp) - I think in the past those were overloaded as well and used but most providers allow the concept of a tag so shouldn't have to do that there either
if you turn on qmp for local instances (eg: for hot-plugging locally)
right now this assumes a unique name (which is a timestamp that is appended to non-custom instance names)
i suppose the quick/easy fix is to allow the custom instance name w/the knowledge that you can just use one or use a tmp id of some kind
at least for onprem targets another method might be to generate a id beforehand and not overload the instance name - I don't know how much we rely on the name for other providers (such as gcp) - I think in the past those were overloaded as well and used but most providers allow the concept of a tag so shouldn't have to do that there either