OK, I have seen the previous discussion about timeout, in that case, timeout is used for the reliability protection of message transmission, eg. once the packet is dropped, the reciever will enable the the timeout behavior,. so you can specify timeout action as a sucessor of packet loss action in the state machine.
But in my case, the timeout action is periodically happened, it used to trigger state sync operation to the whole cluster periodically , so it is always enabled as long as the node is not faulty.
The following is my current description about the cylic timer function for a dedicate node.
TimerProc(n) == /\ NodeState[n] = "Work"
/\ ( IF Tick = MAX_TIME
/\Tick' = 0
/\ BroadcastSyncMessage(n) \*when the cylic timer is out, node send broadcast message to the whole cluster and reset the timer
Tick' = Tick +1 \* the independent tick elapse ++
/\ UNCHANGED vars
\ / TimerProc(n)
\/ ... \* other service behavior
is that OK? I am not so content with the formula above, because I don't need to module the real time action of the system, so I want to know if the Tick variable could be eliminated?
I am also afraid that since the TimerProc() is always enabled, will it leads to state space explosion of the whole system?
在 2020年6月15日星期一 UTC+8下午2:25:05，Stephan Merz写道：
I am not really sure that I understand your question. Of course, TLA+ is meant to model systems whose behavior "never ends", and as long as the state space is finite, TLC can verify properties of such systems.
There have been previous discussions on modeling timers and timeouts in this Google group, e.g. . Perhaps you find some hints there, otherwise feel free to come back, perhaps with a simplified example that explains the issue.
I am specifying a distributed system consists of a cluster of nodes.
Each node's service state will be syncronized to the whole cluster periodically.
In the implementtation layer, there is a cycle timer in each node and generate syncronization message periodically.
What really confused me is how to specify this mechanism abstractly.
It is easy to simulate the timeout method under the pre-condition that the node is in working state, but the result is that the state machine will always driven by the timeout mechanism and never end.
This scenario is somewhat like the so called stuttering state if the system had been in convergeny and without other valid trigger.
So, what is the principle to get rid of the stuttering cycle timer's influence in buliding the model?
You received this message because you are subscribed to the Google Groups "tlaplus" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tla...@googlegroups.com
To view this discussion on the web visit https://groups.google.com/d/msgid/tlaplus/c3bcf614-06bd-4990-853a-2c30c224b31eo%40googlegroups.com