I think in some cases, the participants register state model after they connect to the cluster. In this case, before the participant can register state model, controller can detect the new instance and fire the appropriate transitions which might put the participant in error state. Instead what we do currently is keep the transition in pending state and when the state model is register we invoke the transition handler.
Does it make sense? On Fri, Jul 19, 2013 at 8:48 AM, Santiago Perez <[email protected]>wrote: > not really, it would produce the same result as today, the node would > remain on OFFLINE state > > > On Fri, Jul 19, 2013 at 12:43 PM, kishore g <[email protected]> wrote: > > > So a change for controller to fire transitions only when participant has > > registered a state model should work right. It wont explicitly put the > node > > in error state but avoids sending the transition. > > > > Putting into ERROR state is little extreme. > > > > > > > > > > On Fri, Jul 19, 2013 at 8:27 AM, Santiago Perez <[email protected] > > >wrote: > > > > > We use configuration to mark nodes to support different statemodels. so > > it > > > can be the case that a pre-existing ideal state had node A assigned to > > > resource B which used statemodel C but the node now only supports > > > statemodel D. This is a human error and we want to make it explicit > with > > an > > > error state rather than hiding it behind a transition never starting > > > (specially because we have transitions that take up to 10hs). > > > > > > Thanks > > > Santi > > > > > > > > > On Fri, Jul 19, 2013 at 11:45 AM, kishore g <[email protected]> > wrote: > > > > > > > it would probably be better to change the controller to not send any > > > > transition message until a statemodel is registered. > > > > > > > > Controller can treat it as this node is disabled for this state > model. > > > > > > > > Can you give more info on when would this happen in your usecase. Is > > > there > > > > a case where you connect but then fail to register statemodel? > > > > > > > > thanks, > > > > Kishore G > > > > > > > > On Thu, Jul 18, 2013 at 1:14 PM, Santiago Perez < > [email protected] > > > > >wrote: > > > > > > > > > By default, if a transition is sent to a participant and it has no > > > state > > > > > model registered it will wait until such model is registered. It > > would > > > be > > > > > nice if based on some attribute of the state model def, one could > > force > > > > the > > > > > node to go into error state if such situation arises. > > > > > > > > > > What do you think? > > > > > > > > > > > > > > >
