- Implement DatabaseWorkerPool::DirectCommitTransaction for synchronous transaction execution (as opposed to asynchronous/enqueued).
- Add MySQL errno 1213 "Deadlock found when trying to get lock; try restarting transaction" handler. If 1213 is called the core will retry to directly execute the transaction a maximum of 5 times.
- Make time format of MaxPingTime consistent between authserver and worldserver. Note that the value is in minutes, update your config files accordingly if needed.
- Remove obsolete atomic uint32 and prevent concurrent calls to vector::size() in DatabaseWorkerPool
--HG--
branch : trunk
- Make the mysql connectionpool shared for async and syncrhonous connections.
- Allow configurable amount of connections for the pool
- Allow configurable amount of delaythreads
Note that delaythreads now only represent in-core threads. Whenever they execute a task they will pick a free connection from the pool instead of using their previously unique assigned connection.
The purpose of this design change is better distribution of SQL requests (no bottlenecks paired with idling) among available resources.
This also prevents a ¨memory waste¨ of preparing async prepared statements on synchronous connections (that were never called) - and vice versa. Now, connections aren´t explicitly async or synchronous, they serve both purposes.
Use at own risk, might cause instabilities.
Don´t forget to update your config files and clear your cmake cache.
--HG--
branch : trunk
- Fix a race condition in KeepAlive() when connections are using mysql context when ping is called.
- Don´t wait for locks to be released on a connection when pinging, this means the connection is not idle and locking is redundant.
Patch by admin@****.net
Fixes issue #4599
--HG--
branch : trunk
- DB Threading model update
* Get rid of ThreadBundleMask and bundled connection
* Implement configurable amount of Synch threads for databasepools
* Use modulus based algorithm to check for free synchronous connections instead of previous ¨get connection by thread key or bundlemask¨ feature
* Locks on mysql context objects are now managed outside the mysql query methods
Fixes issue #4058
Fixes issue #4059
Introduces a ton of more issues. Use at own risk. You were warned. Really.
Don´t forget to update your worldserver.conf
--HG--
branch : trunk
- Rewrite KeepAlive method for DatabaseWorkerPool. Use mysql_ping instead of explicit select queries, and also schedule KeepAlives for asynchronous threads.
NOTE: While the function is implemented and previous keepalive calls were transformed, it´s possible the keepalive call will need to be placed in several other locations in the code. Please leave feedback on whether or not this fixes your timeout issues.
Update issue #4062
--HG--
branch : trunk
- Rewrite Field class to be able to store both binary prepared statement data and data from adhoc query resultsets
- Buffer the data of prepared statements using ResultSet and Field classes and let go of mysql c api structures after PreparedResultSet constructor. Fixes a race condition and thus a possible crash/data corruption (issue pointed out to Derex, basic suggestion by raczman)
- Conform PreparedResultSet and ResultSet to the same design standards, and using Field class as data buffer class for both
* NOTE: This means the fetching methods are uniform again, using ¨Field* fields = result->Fetch();¨ and access to elements trough fields[x].
* NOTE: for access to the correct row in prepared statements, ¨Field* fields = result->Fetch();¨ must ALWAYS be called inside the do { }while(result->NextRow()) loop.
* NOTE: This means that Field::GetString() returns std::string object and Field::GetCString() returns const char* pointer.
Still experimental and all that jazz, not recommended for production servers until feedback is given.
--HG--
branch : trunk
- Declare the datatypes used to determine transaction element data (prepared statement/adhoc query string) on a generic level in SQLOperation.h
- Implement variable SQL element data for SQLQueryHolder class so it can execute both prepared statements and adhoc queries
- Make MySQLConnection::Query for adhoc queries return pointer to type instead of an autopointer, the autopointer is now applied on higher level code just like the function for querying prepared statements
--HG--
branch : trunk
* Example implementation of prepared statements with resultset in RealmList and AuthSocket code (selectively)
* Also correct a few bobo´s from previous commit.
--HG--
branch : trunk
* Rename QueryResult class to ResultSet
* Rename QueryResult_AutoPtr to QueryResult
* Declare ACE refcounted auto pointer for PreparedResultSet class
--HG--
branch : trunk
- Store threadbundlemask internally per database pool and prevent direct access to config file post startup
- Fix threadbundlemask flag checking for ReactorRunnable, WorldRunnable
- Remove CLI threadbundlemask flag, CLI doesn´t need a seperate mysql connection nor thread
- Remove unused Character Database connection from WorldSocketMgr / ReactorRunnable
- Add proper LoginDatabase connection to RA Runnable (soon to be overhauled)
Note: still experimental and not tested for live use
--HG--
branch : trunk
- Add some small documentation on prepared statement defines naming contention (placed in the 3 database implementation header files so it can´t be missed)
- Remove obsolete mutex declaration in DatabaseWorkerPool (was unused as of recently)
--HG--
branch : trunk
- Add basic prepared statement interface (without implementation and thus without testing, without resultset support).
- Based on raczman/Albator´s work on TrinityCore3
Build:
- Add sLog define ¨hack¨ in DatbaseWorkerPool.h to fix some more sLog errors temporarily
FYI: Builds on windows
--HG--
branch : trunk
- Implement deriviate classes of MySQLConnection for every database type (world, realm, characters)
- Make DatabaseWorkerPool templatized and use the above mentioned classes as parameter
- Implementation of the new types in code
(This is in preparation of prepared statement interface)
--HG--
branch : trunk
- Make SQL Transactions actual objects used in code. (Thanks to Derex for the idea)
* Uncommitted transactions will be automatically rolled back and cleaned up using ACE_Refcounted_Auto_Ptr, so no need to call Rollback() in the code.
* Prevents recursive transactions and makes developers aware of transactions going on.
* Gets rid of unneccesary overhead iterating over a concurrent map.
- Some cleanups in affected code, including better usage of transaction control in AH / mail related code to prevent data loss.
*** Experimental, use at own risk, recommended to backup your DBs. ***
--HG--
branch : trunk
leak