[Pvfs2-developers] state machine ending

[Phil Carns]

Sam Lang wrote:
> 
> On Jul 24, 2006, at 2:07 PM, Pete Wyckoff wrote:
> 
>> walt at clemson.edu wrote on Mon, 24 Jul 2006 14:47 -0400:
>>
>>> 3) the state machine language has a keyword that lets you state  that 
>>> the
>>> next state is "terminate" which could be used to finish the state
>>> machine, but right now the code has logic that detects this and  
>>> prints a
>>> warning that you should NOT use this.  I don't know why we created  this
>>> method and then made it illegal.  Go figure!
>>
>>
>> I prefer #3 by a large margin over the other options.  It's
>> explicit, not racy, doesn't over-use the return value for a
>> conceptually different event, and doesn't rely on modifying
>> parent state.  Wonder why it was disabled.
>>
> I think terminate ended up being more of a placeholder...we still use  
> it in quite a few of our state machines, but the last state action is  
> expected to call server_state_machine_complete which prevents the SM  
> from transitioning to the 'terminate' state.  I've thought of it as  an 
> assert:  if you get here you're forgetting something.

Just for historical reference, the terminate state was never really used 
in pvfs2- it wasn't that it was disabled or illegal.  It was just that 
the completion functions destroyed the sm before it got to that point. 
I think I added the safety check later on because we had a few "runaway 
state machine" bugs where sms weren't calling their completion functions 
on the way out.  The terminate transition ended up being a good place to 
catch that sort of problem since the pvfs2 state machines that worked 
properly were never supposed to hit it.

I think nowadays a better way to track down that sort of problem would 
be with a state machine dumper utility (like the one Derek Horton 
started but we never got around to polishing up) because something like 
that could point out runaway state machines even if they get lost 
somewhere before termination.  I don't know if you have any interest in 
something like that, Walt, but you might want to have a look at what 
those patches did to see if there is a cleaner way to implement 
something along those lines with your new infrastructure changes.

Anyway, I agree with all of the comments from Walt, Sam, and Pete- it 
sounds like this stuff will be a lot cleaner.


-Phil