Posted by: Eric Hansen
Central, management, single point of failure, SPOF
I’ve always been one to dislike central management systems. Monitoring systems are nice. If there wasn’t a central system to maintain servers it would make life a lot more busy. However, if you have a cluster and you decide to manage all the slaves, as well as the master, via control panel only, there’s one major flaw about this. That is…
…what if that single point fails? If you have SSH enabled on the server, that’s great. Just SSH Into the server, restart the service or figure out why it failed, and go from there. But, on the other hand, if you disable SSH, there’s no way to access the server short of physically going to it, which isn’t possible if your data center is located half-way across the world.
There is a reason why people tend to not implement single-points of failure in their infrastructure. Not saying there’s no purposes to it, but it should be used with caution. While yes, having a back door open to the server isn’t exactly the smartest choice, there’s always ways around it. I think if you do go with a plan such as single point, you should make sure that multiple point is just not feasible. Perhaps there’s too much eavesdropping going on throughout the network, or even the protocol/system/etc… you are doing this for doesn’t support it.