The code relied on last_read for the idle time calculation which caused
the pings that were written to not reset the idle time. This increased the
chance of multiple COM_PING packets being sent to a backend before a reply
was received.
The use of the server state is not transactional across multiple uses of
the function. This means that any assertions on the target state can fail
if the monitor updates the state between target selection and the
assertion.
The slave backend would be closed twice if it would both respond with a
different result and be closed due to a hangup before the master
responded.
Added a test case that reproduced the problem.
It appears that rollback errors are possible outside of
transactions. Since this was not something we expected to see, logging it
as an error allows us to see why this happens in production deployments.
The errors that are ignored by readwritesplit are now stored as the
current close reason in the Backend. This allows the information about the
error to be retained and it can be used later in the error handler to
display the true reason why the connection was closed.
When lazy_connect is enabled and the first query is `SET autocommit=0`, a
master connection can be created. If it is, then the m_current_master
pointer must also be updated.
Also fixed the case where a failure to connect to one slave would cause
the connection attempts to stop too early.
This allows the same verbose information to be logged in the cases where
it is of use. Mostly this information can be used to figure out why a
certain session was closed.
By doing the reconnection only when a new query arrives, we prevent the
excessive reconnecting that is done when a server's actual and monitored
states are in conflict.
If a master failed during an ongoing session command history replay, it
would be treated as if a normal session command failed which would result
in the already executed session command being re-executed on all servers
at the wrong logical position.
To fix this, the history replay must be distinguished from normal session
command execution. When a connection replaying the history fails, the
query routing simply needs to be attempted again.
If session command execution during server reconnection caused a query to
be queued, the query would be put on the tail end of the queue. This would
cause queries to be reordered if the queue wasn't empty. The correct thing
to do would be to put the next pending query back at the front of the
queue.
If a master reconnection occurred after the session command history was
disabled due to the limit being exceeded, a debug assertion would be hit
in prepare_target. This assert makes sure that a connection can be safely
created to the server which means that in release mode builds the session
state would be inconsistent on the new master.
As this is an unrecoverable situation, the session should stop immediately
even if delayed_retry is enabled. Currently the session will continue
until the delayed retry timeout is hit. This happens due to the fact that
the delayed retry mechanism handles all errors in a similar way.
When a query returns a WSREP error, most of the time it is not something
the client application is expecting. To prevent this from affecting the
client, it can be treated the same way a transaction rollback is treated:
ignore the error and try again.
The expected response counter was not decremented if a transaction replay
was started. This caused the connections to hang which in turn caused the
failure of the mxs1507_trx_stress test case.
If the slave's response differs from the master and the slave sent an
error packet, log the contents of the error. This should make it obvious
as to what caused the failure.
The assertion in routeQuery that expects there to be at least one ongoing
query would be triggered if a query was received after a master had failed
but before the session would close. To make sure the internal logic stays
consistent, the error handler should only decrement the expected response
count if the session can continue.
