Dave-
Simple answer, what you are asking is not available.
ACE can do 3 things to connections on an rserver when its probe fails.
Default: ACE no longer sends new connections to it, but leaves the existing connections on it. The idea here is we don't know if the server missed a probe because it was overloaded, so we don't want to kill off all the connections by default.
Failaction purge: This is a configuration done under the serverfarm. When the probe fails, ACE sends a reset to both the client and the server for each connection.
Failaction reassign: This is a configuration done under the serverfarm. When the probe fails, re-assign the back end of the connection to another Rserver. This is only viable for L2 flows, and for devices like firewalls which share connection state between eachother. We simply remap the connection, there is no new TCP handshake on the backend. In that manner, the next packet a client sends to the vip for an existing moved conneciton would be send to the new rserver. Hence, if the packet went to an rserver that was not previously aware of this connection, it would simply send a reset. There are actually quite a few limitations, you can find them in the guide for creating a failaction under a serverfarm.
The reason we don't offer creating new backend sessions is that there are too many possible issues that crop up.
Every connection moved would be handled in software to setup the 3 way handshake on the backend, hence, high cpu for a bit to set them up.
What if the probe is flapping and we have to keep reassigning connections? That will cause issues for both CPU and servers.
Servers are not accustom to taking a large burst all at once (say the rserver that failed had 50k live connections, and you only had 3 rservers. When the probe failed, ACE would have to open 25k connections to each of the other servers instantly. Most servers don't handle that sort of burst well.)
Regards,
Chris Higgins
