[infinispan-dev] heuristic transactions & failure recovery

[Mircea Markus]

>
> Yes, but this is allowed in 2PC.  This leaves the tx in a state of 
> STATUS_UNKNOWN, and it is up to the transaction manager to initiate a 
> recovery *if* the resources are XA compliant and support recovery.
okay. So shall we support recovery? I'm thinking that there are many 
scenarios/clients that simply wouldn't live with the fact that out tx 
might not be atomic (nothing unusual so far) and there's no way in which 
these atomicity failures are (a) reported and (b) solved.
This does not have to be the default, as our clients have been living 
with this for years in JBossCache; I'm referring to infinispan: this 
might be an extra for us.