This and other RFCs are available on the web at http://dev.perl.org/rfc/ =head1 TITLE Transaction-enabled variables for Perl6 =head1 VERSION Maintainer: Szabó, Balázs <[EMAIL PROTECTED]> Date: 17 Aug 2000 Last Modified: 13 Sep 2000 Mailing List: [EMAIL PROTECTED] Number: 130 Version: 6 Status: Developing =head1 ABSTRACT Transactions are quite important in a database-enabled application. Professional database systems have transaction-handling inside, but there are only a few laguage out there, what supports transactions in variable level. In perl6, we can support transaction-enabled variables (including objects and tied variables), and we can control transaction-enabled perl modules with that (this include modules that do external I/O also). We can use our perl program to tie several transaction-enabled data sources, and we can use perl to easily maintain the consistency between them. In this RFC we will look at how these variables would look like in perl6. =head1 STATUS The idea and the implementation issuse is mainly clarified: We have now a tidy and simple design. Some questions are still open, these are the following: =over 4 =item * Now we need to get a name for a new keyword name, which is more widely accepted than the current "trans". =item * We need to decide whether we want multi-version concurrency control or not in the core. If we doesn't implement it, all things are remains simple. If we implement this, we will get smaller chance to deadlock, but it makes the core complicated. =item * We need to think about what if 'trans' is used with a non-transaction-enabled tied array or hash. How we can make it effective withour sacrificing either simplicity, speed or memory-requirement. =back =head1 WHAT'S NEW IN VERSION 6 =over4 =item * Added STATUS section =item * Added some Implementation issues =item * Added some more explanation to the ABSTRACT section. =back =head1 DESCRIPTION In short, we have "local" keyword, which changes a value of a variable for only the runtime of the current scope. The transaction-enabled variables should start up like "local", but IF the currenct scope reaches at the end, it then copied into the global one. We need to get a keyword to mark a variable transaction-enabled. I chosed "trans" in this ducument, but other suggestions are welcome. Possible alternatives are: transaction transactional acid atomic onsuccess consistent The final decision will be made by the porters, I use "trans" in this document. Preferred syntax: sub trans_test { my ($self,@params)=@_; trans $self->{value}=$new_value; # ... die "Error occured" if $some_error; function_call(...) # ... } ; Meaning (in semi perl5 syntax): sub trans_test { local $self->{value}=$new_value; # ... die "Error occured" if $ome_erre; function_call(...) # ... global $self->{value}=$self->{value}; }; If we want to gain more control and want to maintain easy syntax, we can use another pragma, which sets up the attributes of the isolation of transaction data. I think the "transaction" pragma could be a good name: use transaction (mode => 'lock', timeout=>6000); Parameters for "use transaction": =over 4 =item mode can be: =over 4 =item simple No blocking, concurrency control (default). In a not-threaded environment this causes minimal overhead, and no locking overhead at all. =item lock Explicitly lock the accessed variables. (shared and exclusive locks used between threads). =item version This is simlar to the postgres' multi-version concurrency control. It requires more memory, but has a less chance to get into deadlock. =back =item timeout Timeout in ms until we wait for the lock. 0 means nonblocking. If the timeout reached, the system throws an exception. =item isolation Transaction isolation level. This can be: =over 4 =item 0 Read uncommitted =item 1 Read committed (default) =item 2 Repeatable read =item 3 Serializable. =back PostgreSQL implements only 1 and 3 AFAIK, so I think we could implment only those levels. Then 0 and 2 will be equal to 1 and 3, but we could keep the place for a future implementation. See the postgreSQL documentation for the details. =back =head2 Two phase commit Two phase commit is the common way to deal with distributed transactions. Perl need an interface to objects and tied variables to deal with these to become a reliable transaction-handler. You can choose to implement these features in your object and your tied variable. If you don't do that, perl will give you a rough default. At the end of the transaction, 2 different thing can happen: rollback or commit. When rollback occured, all the transaction variables must be rolled back. In commit, a two-phase commit procedure has been started. The first phase is preparing to the commit: check the resources, allocates resources to the commit, flushes caches, etc. After that it can decide wheter you can do a commit or not. If all participants send "yes", then the commit phase begins: the coordinator sends "commit" messages to the participants, and the transaction finishes. If any of the participants in the "prepare" phase sends a false value, then the whole transaction need to be rolled back. How it looks like in perl? You have objects. Objects can be transaction-enabled, and if you want that, you need to define the following functions as callbacks: COMMIT, ROLLBACK, PREPARE, BEGIN_TRANSACTION. If you have a tied variable, then you can define callbacks for this: TIE_COMMIT, TIE_ROLLBACK, TIE_PREPARE, TIE_BEGIN_TRANSACTION. These can be used to extend an object or a tied variable to transaction-safe. If you don't define PREPARE or TIE_PREPARE, then it will be only a one phase commit. If you don't define COMMIT (or TIE_COMMIT) and ROLLBACK (or TIE_ROLLBACK), then perl will do the simple "copy back the old value on rollback" mechanism, which works well in cases when no multithreading and no special handling is necessary for the data. If you don't define BEGIN_TRANSACTION or TIE_BEGIN_TRANSACTION, then no special initialization performed on "trans" call. =head2 Tie interface Adding transaction-enabled property of a tied variable is not straightforward. Imagine you have been tied a hash into a (not transaction-enabled) dbm file. When you fetch, you need to put a shared lock (or version-control) the dbm file or key, when you read, you need to put an exclusive lock, and when the transaction ends, you need to release the lock. For this reason, we can add two callback: TIE_COMMIT and TIE_ROLLBACK. If we don't want to use locking, or want to do an advanced transaction-management, we can provide a transaction-id to the callbacks. This can be done with a new package global variable (which is localized in every call), the name can be $Package::TRANSACTION_ID. We could use a new parameter, but it is not is not so neat, because some of the callbacks (PUSH, POP, UNSHIFT, PRINT, PRINTF, etc) are expecting LISTs as an attribute, and this can cause unnecessary rewrite of the tie interface. Following is the description of the modifications of the tie interface: =over 4 =item New package global $Package::TRANSACTION_ID will be a unique identifier of the current transaction (if any). =item New Callbacks Some new callbacks required: =over 4 =item TIE_PREPARE $this This is the first part of the 2 way commit transaction. This must return true if the variable is prepared for the COMMIT, false otherwise. If this callback is not defined, then the variable lose the right to abort the transaction, and perl implicitly returns 1 in this cases. =item TIE_COMMIT $this If it is defined, then it is called after TIE_PREPARE returns 1 for all the transaction-enabled variables in the current scope. This must be used to commit the transaction. =item TIE_ROLLBACK $this If it is defined, then it is called at the end of a failed transaction. If NOT defined, then STORE will be called with the old value of the variable. =item TIE_BEGIN_TRANSACTION $this,$parent_transaction_id It is called by the system, if the program execution finds a "trans $this" in the program. This is intended to initialize the transaction. This callback is special, because if you use the "trans" or "local" keyword in this subroutine, this will not last only the end of the sub, but the end of the transaction! $parent_transaction_id can be used to track the sub-transaction relations. This is the ID of the transaction which contains this transaction. undef if this is the master transaction (has no parent). If a subtransaction is terminated (and the exception is catched), then the proper rollback callbacks are called, but the parent transaction can be successful! =back =back If a package used in "tie" has one of the above callbacks, then perl _must_ emulate the transaction in every call, so a simple FETCH in non-transaction enironment must be the sequence of TIE_BEGIN_TRANSACTION, FETCH, TIE_PREPARE ? TIE_COMMIT : TIE_ROLLBACK and a simple STORE must be: TIE_BEGIN_TRANSACTION, STORE, TIE_PREPARE ? TIE_COMMIT : TIE_ROLLBACK. =head2 Object interface Object interface is similar to the tied interface: you will need callbacks: PREPARE, COMMIT, ROLLBACK and BEGIN_TRANSACTION. These will do the same as described in the Tie interface. The $Package::TRANSACTION_ID will be set in this case also. Note, if you declare an object as "trans", this means that this is localized for the runtime of the transaction and that PREPARE, COMMIT, ROLLBACK will be called at the end of the block of the declaration. It doesn't mean that all the data structure under that is transaction safe. It cannot be guaranteed, and you need to explicitly declare them as "trans" variables. =head1 TRANSACTION-ENABLED TIED VARIABLE EXAMPLE This is an example of a transaction-enabled tie interface. The following package can be tied to any variable, and can be used as a persistent, transaction-enabled data. Usage: tie $scalar, "Transaction::ScalarFile", $filename; sub my_transaction { trans $scalar; $scalar="Perl" x 1024; ... }; The data in the file referred by $filename can be accessed, modified as $scalar. $scalar can be used in a transaction, supports subtransactions, and supports two-phase commits and locks the accessed file with flock(), so it can be used in multithreaded and multiprocess environment. Here is the code: package Transaction::ScalarFile; use transaction (mode => 'lock', timeout => 30); use strict; use Fcntl qw( :flock ); # constant declaration sub FILENAME { 0; }; sub FILEHANDLE { 1; }; sub VALUE { 2; }; sub LOCKED { 3; }; sub PARENT_TRANS { 4; }; sub TEMP_FILENAME { 5; }; sub TEMP_FILEHANDLE { 6; }; sub TIE_BEGIN_TRANSACTION { my ($s,$parent)=@_; trans $s->[VALUE]; # The value is transaction-enabled local $s->[PARENT_TRANS]=$parent; }; sub TIESCALAR { my ($class,$filename)=@_; my $s=[$filename]; bless $s,ref($class) || $class; $s; }; sub FETCH { my ($s)=@_; return $s->[VALUE] if defined $s->[VALUE]; $s->open or return undef; flock $s->[FILEHANDLE], LOCK_SH; local $/=undef; $s->[LOCKED]=1; return $s->[VALUE]= < $s->[FILEHANDLE] >; }; sub STORE { my ($s,$value)=@_; if ($s->[LOCKED]<=1) { $s->open or return undef; flock $s->[FILEHANDLE], LOCK_EX; $s->[LOCKED]=2; }; $s->[VALUE]=$value; }; sub TIE_PREPARE { my ($s)=@_; # If it is a subtransaction: do nothing return 1 if $s->[PARENT_TRANS]; # If the value is not modified: do nothing return 1 if $s->[LOCKED]<2; # Transaction failed if I cannot make the temp file $s->create_and_lock_temp_file or return 0; $!=0; # reset ERRNO # Writes the new value to the tempfile print $s->[TEMP_FILEHANDLE], $s->[VALUE]; # Transaction failed if error occured unlink($s->[TEMP_FILENAME]),return 0 if $!; return 1; }; sub TIE_COMMIT { my ($s)=@_; return if $s->[PARENT_TRANS]; # This is the weakest point of the transaction, we cannot make those two # operations atomic ... rename $s->[FILENAME], $s->[FILENAME].".old.$$"; rename $s->[TEMP_FILENAME],$s->[FILENAME]; unlink $s->[FILENAME].".old.$$"; flock $s->[FILEHANDLE],LOCK_UN; flock $s->[TEMP_FILEHANDLE],LOCK_UN; $s->[LOCKED]=0; $s->[TEMP_FILEHANDLE] = $s->[TEMP_FILENAME] = undef; }; sub TIE_ROLLBACK { my ($s)=@_; return if $s->[PARENT_TRANS]; # We don't care if it has parent trans. flock $s->[TEMP_FIELHANDLE], LOCK_UN; flock $s->[FILEHANDLE], LOCK_UN; unlink $s->[TEMP_FILENAME]; $s->[TEMP_FILEHANDLE] = $s->[TEMP_FILENAME] = undef; }; sub open { my ($s)=@_; return $s->[FILEHANDLE]=new FileHandle("<".$s->[FILENAME]); }; sub create_and_lock_temp_file { my ($s)=@_; $s->[TEMP_FILENAME]=$s->[FILENAME].".trans.$$"; $s->[TEMP_FILEHANDLE]=new FileHandle(">".$s->[TEMP_FILENAME) or return 0; flock $s->[TEMP_FILEHANDLE],LOCK_EX; }; =head1 IMPLEMENTATION =head2 Transaction handling methods =over 4 =item simple This is the default method. This needs no magic, implementation is straightforward: When you use "trans", then the transaction-enabled variable is duplicated automatically in the memory (with a copy-on-write method). IF the transaction succeeded, this will copied back to the original. =item lock We need to maintain locks (mutexes) on variables. We assume this will be used in threaded applications. When we use "trans", then perl will put a shared lock on the variable. When we read the variable, we also put a shared lock to that. When we write the variable, we check if it is already locked, and if we locked that already or no exclusive locks present, then write to the value, and lock that with LOCK_EX. If other exclusive lock present on the variable, then we need to wait for the releasing. When the "trans" content ends, we frees the shared (or exclusive lock). If the content ends with a die then we puts the original value back if we have locked it with exclusive lock. =item version It is the mechanism of making multiple versioned copies of the variable every time somebody access this. This needs tiestamping, and postgreSQL-like concurrency control. I don't know more details. =back =head2 Use 'trans' for non-transaction-enabled tied vars and object If we use 'trans' keyword for a value which is a tied variable or an object, but which doesn't implement the transaction-interface, our transaction-safe environment is not guaranteed to be consistent anymore. We cannot make the system 100% transaction-safe anymore. All we can do is to emulate the transactional behaviour with our current tools. If we take a look at the TIE interface, then we can emulate the transaction behaviour only with STORE and FETCH. It is not a problem with a simple scalar, but it is a problem if we think about a tied array or a tied hash, or simply we throw a fatal exception if someone try this. We must think about it. One point is clear: the transaction is as weak as the weakest transaction-enabled variable in it no matter how we emulate the transaction-behaviour. =head1 CHANGES IN PREVIOUS VERSIONS =head2 Version 5 =over 4 =item * TRANSACTION-ENABLED TIED VARIABLE EXAMPLE added =item * BEGIN_TRANSACTION and TIE_BEGIN_TRANSACTION added =item * TIE* renamed to TIE_* =item * CHANGES moved to the end of the RFC =item * Added alternatives to "trans" =back =head2 Version 4 =over 4 =item * TIE interface callbacks are renamed to TIE_COMMIT, TIE_ROLLBACK =item * 2 phase commit described =item * PREPARE, TIE_PREPARE added to support two phase commits =item * Object interface described =item * "safe" renamed to "trans" =back =head2 Version 3 =over 4 =item * Added a tie interface change request: COMMIT and ROLLBACK, and new global =item * Fixed some Formatting error of this pod. =item * 'use varlock' renamed to 'use transaction'. =back =head2 Version 2 =over 4 =item * Detailed implementation description =item * Add a new pragma 'varlock' for controlling the concurrency control. =back =head1 REFERENCES PostgreSQL Multi-version concurrency control http://www.postgresql.org/docs/postgres/mvcc.htm Two phase commit: (Google found that :-) http://oradoc.photo.net/ora8doc/DOC/server803/A54653_01/ds_ch3.htm RFC 19: Rename the C<local> operator RFC 119: object neutral error handling via exceptions perldoc perlthread: the perl5 threading interface perldoc perltie: the perl5 tie interface