Prepared statements with a SELECT statement were falsely detected as SELECT queries with session variable modification.
This caused the error message and the failure to route the query.
readwritesplit.c:get_route_target routed query to slave although query type was combined QUERY_TYPE_READ | QUERY_TYPE_MASTER_READ, where the latter is supposed to be routed to master.
RWSplit router handles query processing errors that happened in router by sending parse error reply to client. routeQuery fails only when backend has failed.
query_classifier.cc: set_query_type lost previous query type if the new was more restrictive. Problem was that if query is both READ and SESSION_WRITE and configuration parameter use_sql_variables_in=all was set, routing target became ambiguous. Replaced call to set_query_type with simply adding new type to type (=bit field) and checking unsupported combinations in readwritesplit.c:get_route_target. If such a case is met, a detailed error is written to error log in readwritesplit.c. mysql_client.c sees the error code and sends an error to client. Then mysql_client.c calls router's handleError which ensures that there are enough backend servers so that the session can continue.
readwritesplit.c:route_session_write failed if the last backend on all backends list was not in use. THe situation where not all backends are used by routing session is normal especially if max_slave_connections is not set to 100%. Thus session commands may have failed if user was bit unlucky.
Changed the logic so that the function fails (and session is closed) if routing fails to any such backend which is in use in the session.
Added server references to services instead of using a raw pointers to server instances.
Changed all references to service->database to service->dbref.
readwritesplit.h:Removed invalid macros which assumed that ROUTER_CLIENT_SES->rses_backend_ref always pointed to valid and used backend reference and thus included potential risk of NULL-pointer refernce.
mysql_backend.c and mysql_client.c:avoid executing CHK_PROTOCOL(p) after original DCB has been released the memory.
readwritesplit.c:Replaced RSES_CLEINT_DCB macro with a function which returns client DCB for a given router client session.
session.c:session_free:if session is child of another service (tee in this case), it is the parent which releases child's allocated memory back to the system. This now also includes the child router session.
dcb.h: Added DCB_IS_CLONE macro
tee.c:freeSession:if parent session triggered closing of tee, then child session may not be closed yet. In that case free the child session first and only then free child router session and release child session's memory back to system.
tee.c:routeQuery: only route if child session is ready for routing. Log if session is not ready for routing and set tee session inactive
mysql_client.c:gw_client_close:if DCB is cloned one don't close the protocol because they it is shared with the original DCB.
