Mysql：FAQ：A.14 Replication

To ensure that catchup can occur for a slave that has been disconnected, you must not remove binary log files from the master that contain information that has not yet been replicated to the slaves. Asynchronous replication can work only if the slave is able to continue reading the binary log from the point where it last read events.

When the slave SQL thread executes an event read from the master, it modifies its own time to the event timestamp. (This is why TIMESTAMP is well replicated.) In the Time column in the output of SHOW PROCESSLIST, the number of seconds displayed for the slave SQL thread is the number of seconds between the timestamp of the last replicated event and the real time of the slave machine. You can use this to determine the date of the last replicated event. Note that if your slave has been disconnected from the master for one hour, and then reconnects, you may immediately see large Time values such as 3600 for the slave SQL thread in SHOW PROCESSLIST. This is because the slave is executing statements that are one hour old. See Section 16.2.2, “Replication Implementation Details”.

1. On the master, execute these statements:

   mysql> FLUSH TABLES WITH READ LOCK;
   mysql> SHOW MASTER STATUS;
   

   Record the replication coordinates (the current binary log file name and position) from the output of the SHOW statement.

2. On the slave, issue the following statement, where the arguments to the MASTER_POS_WAIT() function are the replication coordinate values obtained in the previous step:

   mysql> SELECT MASTER_POS_WAIT(‘log_name‘, log_pos);
   

   The SELECT statement blocks until the slave reaches the specified log file and position. At that point, the slave is in synchrony with the master and the statement returns.

3. On the master, issue the following statement to enable the master to begin processing updates again:

   mysql> UNLOCK TABLES;
   

You should also realize that two-way replication actually does not improve performance very much (if at all) as far as updates are concerned. Each server must do the same number of updates, just as you would have a single server do. The only difference is that there is a little less lock contention because the updates originating on another server are serialized in one slave thread. Even this benefit might be offset by network delays.

To determine how many slaves you can use before the added benefits begin to level out, and how much you can improve performance of your site, you must know your query patterns, and determine empirically by benchmarking the relationship between the throughput for reads and writes on a typical master and a typical slave. The example here shows a rather simplified calculation of what you can get with replication for a hypothetical system. Let reads and writes denote the number of reads and writes per second, respectively.

Let‘s say that system load consists of 10% writes and 90% reads, and we have determined by benchmarking that reads is 1200 - 2 * writes. In other words, the system can do 1,200 reads per second with no writes, the average write is twice as slow as the average read, and the relationship is linear. Suppose that the master and each slave have the same capacity, and that we have one master and N slaves. Then we have for each server (master or slave):

reads = 1200 - 2 * writes

reads = 9 * writes / (N + 1) (reads are split, but writes replicated to all slaves)

9 * writes / (N + 1) + 2 * writes = 1200

writes = 1200 / (2 + 9/(N + 1))

The last equation indicates the maximum number of writes for N slaves, given a maximum possible read rate of 1,200 per second and a ratio of nine reads per write.

This analysis yields the following conclusions:

• If N = 0 (which means we have no replication), our system can handle about 1200/11 = 109 writes per second.
• If N = 1, we get up to 184 writes per second.
• If N = 8, we get up to 400 writes per second.
• If N = 17, we get up to 480 writes per second.
• Eventually, as N approaches infinity (and our budget negative infinity), we can get very close to 600 writes per second, increasing system throughput about 5.5 times. However, with only eight servers, we increase it nearly four times.

These computations assume infinite network bandwidth and neglect several other factors that could be significant on your system. In many cases, you may not be able to perform a computation similar to the one just shown that accurately predicts what will happen on your system if you add N replication slaves. However, answering the following questions should help you decide whether and by how much replication will improve the performance of your system:

• What is the read/write ratio on your system?
• How much more write load can one server handle if you reduce the reads?
• For how many slaves do you have bandwidth available on your network?

To handle the process manually, you should be able to switch from a failed master to a pre-configured slave by altering your application to talk to the new server or by adjusting the DNS for the MySQL server from the failed server to the new server.

For more information and some example solutions, see Section 16.3.7, “Switching Masters During Failover”.

mysql> SHOW VARIABLES LIKE ‘binlog_format‘;

The value shown will be one of STATEMENT, ROW, or MIXED. For MIXED mode, statement-based logging is used by default but replication switches automatically to row-based logging under certain conditions, such as unsafe statements. For information about when this may occur, see Section 5.4.4.3, “Mixed Binary Logging Format”.