According to the manual, the slave-net-timeout specifies how long time to wait
for data from the master before considering the connection dead. From my
experience this actually means that the slave will reconnect if it hasn't
received a full event in that period of time.
E.g. setting slave_net_timeout to 180 seconds, then an event that takes 240
seconds to transfer will never make it through. The replication will make it
75% through and then start all over.
Does anyone know whether that is a bug in the implementation, or whether it is
the documentation that could be more clear?