[Gluster-devel] HA translator questions

[Martin Fick]

I am a bit curious about the new HA translator and how it it supposed to work?  
I have looked through the code a bit and this is my naive interpretation of the 
way it is designed:

It appears that the HA translator keeps track of its subvolumes and whether 
they are active or not.  When attempting to dispatch a request, it picks a 
currently active subvolume or fails if none are currently active.  It appears 
that once it has chosen an active subvolume for a request, it can no longer 
fail over to another subvolume for that particular request, is this true?  If 
so, then it is possible (perhaps likely) that during failovers some requests 
will fail before failover happens even if certain subvolumes never go down?  :( 
  Is this correct or am I missing something? 

TIA for any insights,

-Martin