Redis master-slave synchronization

Concept

runID every last redis The unique identifier of the instance , I don't know the other side runID with ？ Instead of
offset Replication progress -1 Represents the first copy

The first master-slave synchronization

Sent by from instance pysnc Tell the primary instance to connect for the first time
The primary instance will offset, and runID Send to slave Library
The main instance passes bgsave Generate the full amount RDB, Generate RDB Operation commands during and after are written to repl buffer
take RDB Send to the slave instance
Finish executing from the instance RDB after , The main instance will then repl buffer Send to the slave instance

Recovery process of master-slave network flash outage ( Incremental replication )

There is a ring buffer on both the master instance and the slave instance (repl_back_log)
In the main instance repl_back_log Record where you write
From the example repl_back_log Record where you read
Every time the main instance executes a command, it will go to repl_buffer,repl_back_log Write records in
When the master and slave restore the connection . From the instance, put your own read location (slave_repl_offset) Send to primary instance
The main instance writes according to its location (master_repl_offset), Find the missing data from the instance (master_repl_offset - slave_repl_offset Interval data ) Send to slave instance synchronization

Master slave cascade

In a master-slave mode , If the data is synchronized by the primary instance , The main instance bears great pressure
At this time, some synchronization tasks can be handed over to the first synchronized instance to synchronize to the unsynchronized instance .
Compare the main instance to the boss , From example to employee , The boss should assign tasks to employees one by one
But if the employee is promoted to （ The general manager , President ）, Then the boss assigns the task to it two , It is then allocated to employees , This will reduce the pressure of the boss

matters needing attention

because repl_back_log It's a ring buffer
The ring buffer will start overwriting data after one cycle of writing
If the write speed of the master instance is faster than the read speed of the slave instance , It may cause the data that has not been read from the instance to be overwritten by the newly written operation of the master instance
Therefore, it is important to set the size of the ring buffer reasonably (repl_backlog_size)
Check the following information , Generally set to （ Main instance write command speed * Operation size - Master slave network transmission command speed * Operation size ） * 2