hcq/MaxScale
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fixed typos in documentation.

Browse Source
This commit is contained in:
Markus Makela
2015-04-14 20:36:29 +03:00
parent 253c63e958
commit 786f34cf49
26 changed files with 60 additions and 60 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
Documentation/Tutorials/Replication-Proxy-Binlog-Router-Tutorial.md
View File

@ -10,18 +10,18 @@ The most obvious solution to the requirement for a proxy layer within a replicat
Using an intermediate master does not however solve all the problems and introduces some due to the way replication is implemented. A slave server reads the binary log data and creates a relay log from that binary log. This then provides a source of SQL statements which are executed within the slave in order to make the same changes to the databases on the slaves as were made on the master. If the log_slave_updates option has been enabled new binary log entries are created for the statements executed from the relay log. This means that the data in the binary log of the intermediate master is not a direct copy of the data that was received from the binary log of the real master. The resultant changes to the database will be the same, provided no updates have been performed on the intermediate master that did not originate on the real master, but the steps to achieve those changes may be different. In particular if group commit functionality is used, to allow multiple transactions to commit in parallel, these may well be different on the intermediate master. This can cause a reduction in the parallelism of the commits and a subsequent reduction in the performance of the slave servers.
This re-execution of the SQL statements also adds latency to the intermediate master solution, since the full process of parsing, optimisation and execution must occur for every statement that is replicated from the master to the slaves must be performed in the intermediate master. This latency introduces lag in the replication chain, with a greater delay being introduced from the time a transaction is committed on the master until the data is available on the slaves.
This re-execution of the SQL statements also adds latency to the intermediate master solution, since the full process of parsing, optimization and execution must occur for every statement that is replicated from the master to the slaves must be performed in the intermediate master. This latency introduces lag in the replication chain, with a greater delay being introduced from the time a transaction is committed on the master until the data is available on the slaves.
Use of an intermediate master does improve the process of failover of the master server, since the slaves are only aware of the intermediate master the process of promoting one of the existing slaves to become the new master only involves that slave and the intermediate master. A slave can become the new master as soon as all the changes from the intermediate master have been processed. The intermediate master then needs to be reset to the correct point in the binary log of the new master and replication can continue.
An added complexity that needs to be dealt with is the failure of the intermediate master itself. If this occurs then the same problem as described earlier exists, all slaves must be updated when a new intermediate master is created. If multiple intermediate masters are used there is also a restriction that slaves can not be moved from the failed intermediate master to another intermediate master due to the fact that the binlog on the different intermediate nodes are not gaurenteed to be the same.
An added complexity that needs to be dealt with is the failure of the intermediate master itself. If this occurs then the same problem as described earlier exists, all slaves must be updated when a new intermediate master is created. If multiple intermediate masters are used there is also a restriction that slaves can not be moved from the failed intermediate master to another intermediate master due to the fact that the binlog on the different intermediate nodes are not guaranteed to be the same.
## MaxScale's approach
MaxScale takes a much simpler approach to the process of being a replication proxy. It acts as a slave to the real master and as a master to the slaves in the same way as an intermediate master does, however it does not implement any re-execution of the statements within the binary log. MaxScale creates a local cache of the binary logs it receives from the master and it will serve binary log events to the slaves from this cache of the master's binary log. This means that the slaves will always get binary log events that have a one-to-one correlation to those written by the master. Parallelism in the binary log events of the master is maintained in the events that are observed by the slaves.
In the MaxScale approach the latency that is introduced is mostly the added network latency associated with adding the extra network hop. There is no appreciable processing performed at the MaxScale level, other than for managing the local cache of the binlog files.
In addition every MaxScale that is acting as a proxy of the master will have exactly the same binlog events as the master itself. This means that a slave can be moved between any of the MaxScale server or to the real master without the need to perform any special processing. The result is much simpler behaviour for failure recovery and the ability to have a very simple, redundant proxy layer with slaves free to both between the proxies.
In addition every MaxScale that is acting as a proxy of the master will have exactly the same binlog events as the master itself. This means that a slave can be moved between any of the MaxScale server or to the real master without the need to perform any special processing. The result is much simpler behavior for failure recovery and the ability to have a very simple, redundant proxy layer with slaves free to both between the proxies.
# Configuring MaxScale as a replication proxy
Using MaxScale as a replication proxy is much the same as using MaxScale as a proxy between the clients and the database servers. In this case the master server should be considered as the database backend and the slave servers as the clients of MaxScale.