The Perl DBI transaction mechanism is based on explicit manipulation of auto-commit mode:
Turn on the RaiseError
attribute if it’s not enabled
and disable PrintError
if it’s
on. You want errors to raise exceptions without printing anything,
and leaving PrintError
enabled can interfere with
failure detection in some cases.
Disable the
AutoCommit
attribute so that a commit will be done only when you say
so.
Execute the statements that make up the transaction within
an
eval
block so
that errors raise an exception and terminate the block. The last
thing in the block should be a call to
commit()
,
which commits the transaction if all its statements completed
successfully.
After the eval
executes,
check the
$@
variable. If
$@
contains the empty string,
the transaction succeeded. Otherwise, the eval
will have failed due to the
occurrence of some error and $@
will contain an error message. Invoke
rollback()
to cancel the transaction. If you want to display an error
message, print $@
before
calling rollback()
.
The following code shows how to follow those steps to perform our sample transaction:
# set error-handling and auto-commit attributes correctly $dbh->{RaiseError} = 1; # raise exception if an error occurs $dbh->{PrintError} = 0; # don't print an error message $dbh->{AutoCommit} = 0; # disable auto-commit eval { # move some money from one person to the other $dbh->do ("UPDATE money SET amt = amt - 6 WHERE name = 'Eve'"); $dbh->do ("UPDATE money SET amt = amt + 6 WHERE name = 'Ida'"); # all statements succeeded; commit transaction $dbh->commit (); }; if ($@) # an error occurred { print "Transaction failed, rolling back. Error was: $@ "; # roll back within eval to prevent rollback # failure from terminating the script eval { $dbh->rollback (); }; }
The code shown does not save the current values of the
error-handling and auto-commit attributes before executing the
transaction or restore them afterward. If you save and restore them,
your transaction-handling code becomes more general because it does
not affect other parts of your program that might use different
attribute values, but more lines of code are required. To make
transaction processing easier (while avoiding repetition of the extra
code if you execute multiple transactions), create a couple of
convenience functions to handle the processing that occurs before and
after the eval
:
sub transaction_init { my $dbh = shift; my $attr_ref = {}; # create hash in which to save attributes $attr_ref->{RaiseError} = $dbh->{RaiseError}; $attr_ref->{PrintError} = $dbh->{PrintError}; $attr_ref->{AutoCommit} = $dbh->{AutoCommit}; $dbh->{RaiseError} = 1; # raise exception if an error occurs $dbh->{PrintError} = 0; # don't print an error message $dbh->{AutoCommit} = 0; # disable auto-commit return ($attr_ref); # return attributes to caller } sub transaction_finish { my ($dbh, $attr_ref, $error) = @_; if ($error) # an error occurred { print "Transaction failed, rolling back. Error was: $error "; # roll back within eval to prevent rollback # failure from terminating the script eval { $dbh->rollback (); }; } # restore error-handling and auto-commit attributes $dbh->{AutoCommit} = $attr_ref->{AutoCommit}; $dbh->{PrintError} = $attr_ref->{PrintError}; $dbh->{RaiseError} = $attr_ref->{RaiseError}; }
By using those two functions, our sample transaction can be simplified considerably:
$ref = transaction_init ($dbh); eval { # move some money from one person to the other $dbh->do ("UPDATE money SET amt = amt - 6 WHERE name = 'Eve'"); $dbh->do ("UPDATE money SET amt = amt + 6 WHERE name = 'Ida'"); # all statements succeeded; commit transaction $dbh->commit (); }; transaction_finish ($dbh, $ref, $@);
In Perl DBI, an alternative to manipulating the AutoCommit
attribute manually is to begin a
transaction by invoking begin_work()
. This method disables
AutoCommit
and causes it to be
enabled again automatically when you invoke commit()
or rollback()
later.
3.145.35.247