Backup Parent

Dec 23, 2009 at 5:45 PM

Hi Keith,

I was wondering if it were possible to specify a backup parent in case the primary one fails to respond.  If so, could you tell me what I need to have in the config file.

Thanks (and Merry Christmas),


Dec 23, 2009 at 6:19 PM

There isn't any functionality to specify a backup parent. In production, we have multiple tiers of parents. We use the same parent name for all the application servers and control which parent a server connects to via rules in the DNS. This accomplishes what you would like.

Jun 15, 2011 at 6:03 PM

Following up on this thread Keith ... If you are using load balancing DNS rules to have multiple Parent nodes, and the children register with their Parent on startup, how do all the Parent nodes know about all Children?  It seems like only the Parent that DNS sent the request to would know about the Child that registered.  As such, if you lost that specific Parent, you would lose all of the Children registered with it even though you still have other Parents able to handle the requests.

Jun 15, 2011 at 10:13 PM

In the Bing/MSN data center, I have switched from using the DNS for load balancing to using a VIP with the load balancer. In a larger data center where there are thousands of servers, I have one secondary Wsp router for every 400 servers. All of these secondary Wsp routers are load balanced behind a VIP. They in turn connect to one server which is the primary Wsp router for the data center. All of the primary Wsp routers connect to an origin Wsp router which connects everything together across the world.

The app servers in each data center connect to the VIP for their data center. They know the DNS name for the VIP from the Wsp config file. It is not relevent which secondary Wsp router a server actually connects to. Wsp routes events based upon subscriptions and the physical topology of how the servers are connected is not material.

If a secondary Wsp router dies, all its children will re-connect to the VIP. The load balancer will resolve their request to one of the other servers behind the VIP. They connect and never know that they actually have a new parent.

BTW, events in our data center which may take a handful of hops to go from a publisher in one part of the world to a subscriber around the world, is all sub-second latency.