Distributed Transactions - Transaction Processing - page 299

Thus, each participant of a distributed transaction stores permanently at least the
identifiers of the coordinating server and the parent transaction.
Algorithm 13.1 Procedure process- SQL -request .T 0 ;R/, executed at s 0
if no transaction identifier T 0 was given then
create a new transaction identifier T 0
log .n;hT 0 ;Bi/
insert record .T 0 ;;; forward-rolling;n/to the active-transaction table
end if
transform R into an equivalent statement with arguments R 0 ;:::;R k , where each R i touches
onlydatastoredatservers i , i D 1;:::;k (s 0 being the coordinator)
execute R 0 locally in the name of T 0
for i D 1;:::;k do
send a request .s 0 ;T 0 ;R i / to s i to execute R i for T 0
if s i is not yet in the list of participants of T 0 then
add s i to the list of participants of T 0
log .n;hT 0 ; add-participant ;s i i/
end if
end for
wait for the requests to be serviced
compute the result of R from the results received for R 0 ;:::;R k
return the result of R with the transaction identifier T 0
Algorithm 13.2 Procedure service-remote- SQL -request .s 0 ;T 0 ;R i /, executed at s i
if the local active-transaction table contains no record for a subtransaction of T 0 coordinated by
server s 0 then
create a new transaction identifier T i
log .n;hT i ;s 0 ;T 0 ;Bi/
insert record .T i ;s 0 ;T 0 ;;; forward-rolling;n/to the active-transaction table
else
let T i be the active subtransaction of T 0 coordinated by s 0
end if
execute the SQL statement R i in the name of T i
return the result of R i
Example 13.1 Consider the transaction generated by the following fragment of an
application program:
exec sql update r set V = 2 V ;
exec sql select sum (V ) into : new_sum from r ;
exec sql commit .
The execution of this transaction on a centralized database system was considered
in Example 1.1 .