[patch]thread safety connection pool for all backend

[junzhang.jn@…]

Django uses the same connection for all threads. It breaks any database when using transaction or in multithread web server.
this patch is implemented to avoid it.

like ticket [463] but for all.

• use AutoReleaseCursor return connection object to pool when cursor deleted.

• use connections dict protect threaded active connection.

• db backend must implement a get_check_sql function to provide a sql which can check a connection's health.

ado_mssql:

def get_check_sql():
    return "select top 1 1 from sysfiles"

oracle:

def get_check_sql():
    return "select 1 from dual";

other....

sorry for my very very poor english.

[junzhang.jn@…]

delete AutoReleaseCursor , it breaks transaction in multithread when 'threadsafety > 1'.

[junzhang.jn@…]

update the connection pool in magic-removal branch. only the next two diffs needed.

pool.py.2:

• fix check_thread close not thread safety connetion. now thread safety < 2 connection will not run check thread. so you must use zhe connection in the web server which has thread pool.
• fix cannot stop the check_thread in manage.py shell. use atexit.register function for this problem.

django.db.__init__.py.diff:

• add new setting USE_DBPOOL.

and now ,you must implement get_check_sql function in django.db.bachend.xxxx.base.

sqlite3:

def get_check_sql():
    return "SELECT name FROM sqlite_master WHERE type='table' limit 1"

ado_mssql:

def get_check_sql():
    return "select top 1 1 from sysfiles"

postgresql:

def get_check_sql():
    return "SELECT c.relname FROM pg_catalog.pg_class limit 1"

[eugene@…]

Superseded by #1442

[(none)]

Milestone Version 1.0 deleted