Distributed transaction question

Question

I'm implementing distributed transactions between two oracle servers and am confused as to how this can work reliably. I see that I must call prepare then commit. If prepare fails then I must call rollback instead of commit. However, if I call prepare on each database and get a success from each one and then call commit the commit can throw an exception. ie:
database 1
database 2
call prepare for database 1 - success
call prepare for database 2 - success
call commit for database 1 - succes
call commit for database 2 - failure
now what? how is this beter than just calling commit not using distributed transactions? I thought that any errors for commit must be given by prepare otherwise what is the purpose of calling prepare?
Calvin R. Smith

Answer 1

As part of the 2PC contract, during the prepare phase, the resource manager
is required to prepare itself in such a way (e.g. acquire necessary locks,
perform necessary integrity constraint validation, do logging etc) that it
can successfully commit or rollback the transaction branch according to the
final commit/rollback call from the transaction manager. If they are unable
to prepare themselves, they are given a chance to report that early to the
transaction manager by returning an error on prepare. For the 2PC protocol
to achieve atomically, it is imperative that all resource managers take this
contract seriously.
While it is impossible to provide a completely fool-proof guarantee,
resource managers should make every effort to honor that. On rare,
catastrophic occasions where they are unable to fulfill the contract, they
are allowed to return a Heuristic error upon the subsequent commit/rollback
call issued by the transaction manager. It should be noted that heuristic
errors are considered very rare events in distributed transactions, as
compared to commit failures for local (non-XA) transactions.
Regards,
Priscilla
Calvin R. Smith <[email protected]> wrote in message
news:3be18096$[email protected]..
I'm implementing distributed transactions between two oracle servers and am
confused as to how this can work reliably. I see that I must call prepare
then commit. If prepare fails then I must call rollback instead of commit.
However, if I call prepare on each database and get a success from each one
and then call commit the commit can throw an exception. ie:
database 1
database 2
call prepare for database 1 - success
call prepare for database 2 - success
call commit for database 1 - succes
call commit for database 2 - failure
now what? how is this beter than just calling commit not using distributed
transactions? I thought that any errors for commit must be given by prepare
otherwise what is the purpose of calling prepare?
Calvin R. Smith