Re: Want to know more about the load-balancing

[Jan Kneschke]

Let's say we have:

2x [ app ] -> 1x [ proxy ] -> 2x [ mysqld ]

* 2 application-servers with a local proxy
* 1 proxy as load balancers
* 2 mysqld servers we want to load balance across

This the most basic setup for the proxy as load balancers.

As there is only one load-balancer, it is a SPOF (Single Point of Failure).

2x [ app ] -> 2x [ proxy ] -> 2x [ mysqld ]

Adding a second proxy solves the problem. But it open the question how to does the app access the second proxy in case the first one fails:

1) VRRP and move the virtual IP to the new server
2) if the app uses JDBC let it fail-over itself (if proxy1 is not reachable, it picks proxy2)
3) put a proxy next to the application and let it detect that the LB-proxies are down and failover

2x [ app -> proxy ] -> 2x [ proxy ] -> 2x [ mysqld ]

This way the app can stay unchanged and just connects to a local proxy. It will know which of the proxies is up by tracking the connect() calls for success or failure and send the requests to the proxy which is up.

$ mysql-proxy \
--proxy-backend-addresses=10.0.0.2:3306 \
--proxy-backend-addresses=10.0.0.3:3306

The proxy is tracking the availability of its backends for the normal load balancing. If one of the backends is unavailable, it is take out of the rotation for a while and check back again later. This is transparent to the clients.

Using VRRP saves to setup the proxies on the app servers and means less latency as only one proxy has to be passed.

Jan