Hi, On Fri, May 01, 2009 at 02:18:06PM -0600, Caleb Anthony wrote: > Hello, > > In my organization we use ldirectord to monitor and maintain the ipvs > table. It works great, except that we run a bit of a strange setup > that involves nested ipvs (ipvs forwarding to another ipvs), and a > combination of direct routing and IP tunneling. I'm actually in the > process of writing up a small of a testimonial about our > configuration, which I don't think is very common, or maybe ever been > used at all. But, in our endeavors we've come across two limitations > with ldirectord in our environment. > > First, we use the source hash scheduling algorithm to maintain > affinity with real servers. And according to the documentation that I > have seen, the weight setting really means "number of connections" to > the source hash scheduler. This is a problem for us with the fallback > setting, because ldirectord doesn't allow you to set the weight of a > fallback server. Instead, it just defaults to 1. So, in the event of > an emergency, ipvs is only going to allow 1 connection at a time to > our fallback server. Being able to specify the weight of a fallback > server would be a great addition for us. I'll double check, but I think the weight specifies a proportion of conections rather than an absolute number. Could you let me know which documentation you are looking at? > Our second suggestion would be to have the ability to specify multiple > fallback servers, just like you can with real servers. It's hard for > me to explain why we can't live with just one fallback server without > a huge explanation about our setup, but basically we have more than > one fallback server that serves real content, and if a failure > occurred, we have to hope that that one server we have configured is > not only running, but able to handle the sudden load. Also, I'm not > entirely sure if fallback servers are health monitored. I ask because, > if we could specify multiple fallback servers, we would like them to > be monitored just like a real server and added or removed from the > ipvs table as necessary. Fallback servers are not monitored at this time. Implementing Fallback as a second pool of monitored servers would be possible to add to ldirectord, though it would be a non-trivial change. In other words, it would be a reasonable amount of work. But I can't see any disadvantages to your idea as it should be possible to maintain the existing model too. -- Simon Horman VA Linux Systems Japan K.K. Satellite Lab in Sydney, Australia H: www.vergenet.net/~horms/ W: www.valinux.co.jp/en -- To unsubscribe from this list: send the line "unsubscribe lvs-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
