« back to all changes in this revision
Viewing changes to plugin/transaction_log/transaction_log.cc
- Committer: Jay Pipes
- Date: 2010-03-31 19:04:12 UTC
- mto: (1441.1.3 driz-rabbitmq-fix) (1405.5.5 replication-write-buffers)
- mto: This revision was merged to the branch mainline in revision 1443.
- Revision ID: jpipes@serialcoder-20100331190412-slp0ubgopsfjxld8
* Fixes drizzled's atomics:
- fetch_and_add() was actually add_and_fetch() - fixed to have both methods correct
- compare_and_swap() was incorrect for all traits classes. Fixed to return a bool
true only when the supplied value is actually swapped
- fixes increment() and decrement() methods and operator+=() in outer atomics class
template to call proper add_and_fetch() methods on traits classes
- Now that above are fixed, removed the hacks in Query_id and TransactionLog to
have query ID and the new transactoin ID start properly at 1.
* Transaction messages sent over replication stream now use
a real transaction ID, managed by drizzled::TransactionServices. Previously,
the Query_id was being used, resulting in SELECT statements incrementing the
transaction ID.
* Added a test case to ensure that DDL ops are given a transaction ID and SELECT
ops do not increment the transaction ID.
The transaction ID will be paired with a channel ID to become the global
transaction identifier. ReplicationServices will manage the pairing of
channel and transaction ID and understand how far a particular subscriber
node has applied.
- fetch_and_add() was actually add_and_fetch() - fixed to have both methods correct
- compare_and_swap() was incorrect for all traits classes. Fixed to return a bool
true only when the supplied value is actually swapped
- fixes increment() and decrement() methods and operator+=() in outer atomics class
template to call proper add_and_fetch() methods on traits classes
- Now that above are fixed, removed the hacks in Query_id and TransactionLog to
have query ID and the new transactoin ID start properly at 1.
* Transaction messages sent over replication stream now use
a real transaction ID, managed by drizzled::TransactionServices. Previously,
the Query_id was being used, resulting in SELECT statements incrementing the
transaction ID.
* Added a test case to ensure that DDL ops are given a transaction ID and SELECT
ops do not increment the transaction ID.
The transaction ID will be paired with a channel ID to become the global
transaction identifier. ReplicationServices will manage the pairing of
channel and transaction ID and understand how far a particular subscriber
node has applied.
added
removed
plugin/transaction_log/transaction_log.cc
