mysql_common.c:gw_MySQL_get_next_packet didn't handle case where an insert command followed by alter table in the same read buffer. It shouldn't been possible without multi-statement being set.
Added counter clean-up code to readwritesplit closeSession. If operation counters have leaked, that is, due to some error in query routing some counter have been left positive, it is zeroed at session ending. This ensures that there is no garbage in global counter even if individual session wouldn't clean up its counters.
Fixed bug in session command resul handling. In case where backend sent error message the session command cursor wasn't updated properly.
Added check to bref_clear_state, if bref's waiter counter would go negative, decrement to global operation counter is skipped.
Conflicts:
server/core/session.c
server/modules/protocol/mysql_backend.c
server/modules/routing/readwritesplit/readwritesplit.c
Resolved.
Addition of user in topfilter report and general tidyup
Rwsplit handles ERRACT_NEW_CONNECTION by clearing backend reference, removing callbacks and associating backend reference with new backend server. If it succeeds and the router session can continue, handleError returns true. Otherwise false. When ever false is returned it means that session must be closed.
Rwsplit now tolerates backend failures in a way that it searches new backends when monitor, backend, or client operation fails due to backend failure.
dcb.h: define ERRHAND temporarily since changes are still behind that macro
Defined two error handling actions in router.h: ERRACT_NEW_COMMECTION and ERRACT_REPLY_CLIENT.
Failed database is logged at expanding frequence to error and to message log due changes in mysql_mon.c. Added two new members in MONITOR_SERVERS: mon_err_count, and mon_prev_status so that each backend can be treated individually.
Error handling: if mysql_backend.c:dcb_read fails, router's handleError is called instead of closing session.
If mysql_client.c:SESSION_ROUTE_QUERY fails router's handleError is called instead of sending error to client.
readwritesplit.c:select_connect_backend_servers is modified so that in can be called during active router session. When called, it attempts to find one master and maximum number of configured slaves in correct state if necessary.
When handleError needs to replace failed unit it now calls select_connect_backend_servers.
Added a lot of logging to error, trace and message logs which should help the user to handle errors which can't be automatically resolved, like attempt to use nonexisting database.
In order to be able to protect the interface so that mistyped argument
do not endanger MaxScale yet still allow the developer access a router
option has been added that gives the debugcli two modes.
mysqlk_client.c now assumes that mysql packets may arrive in separate pieces.
gw_read_client_event reads client data, and if packet is incomplete it is added to new dcb->dcb_readqueue. When next packet is combined with data in readqueue.
mysql_common.c:gw_MySQL_get_next_packet can now return mysql packet from read buffer although it would span multiple gwbuf buffers.
Supports pulling binlogs from the master, caching to a local file and relaying to a slave.
Only tested with a single slave and a single MaxScale thread.
For each session command a sescmd property is created and added to the end of list of session commands. List is owned by router client session and it includes all session commands from the beginning of router session.
Router maintains an individual session command cursor for each backend. A cursor refers to the first session command which the corresponding backend server haven't yet responded yet.
When response message arrives at any time from backend, first it is checked whether backend's cursor is active. Cursor is active if a session command is routed to backend and the backend haven't responded to it yet. If cursor is active, next it is checked whether the current session command property has been responded by other backend. If both are true, then response message is sent to client as is. If session command response is routed to client already, the arrived response is discarded.
This implementation doesn't guarantee execution order between session commands and queries
if other backend server lags behind in session command execution.
In poll.c : moved processing of EPOLLERR and EPOLLHUP after processing of EPOLLIN and EPOLLOUT.
This ensures that COM_QUIT messages are read and routed forward before signals arrive (from local client/backend).
