Fault Tolerant long-lived TCP connections

Thanks. That was my concern too. I've been looking at ethereal outputs of some rvd chat and think i've got a basic understanding of the transactions involved. Of course I don't know what I don't know. I think I could do what you're saying. i.e. queue a copy of each request until I see a response. If no response within a certain timeframe, then choose a new backend server, relogin, and resend the request. If I could pull this off, then i'm safe from client data loss. But the other issue is this. I noticed last weekend was that as soon as the server is shot in the head, the client immediately disconnects. No iRule events happen in between the server death and the client disconnect. So there appears to be no way to "catch" the server death and reattach to another running server before the client gets killed. This wreaks havoc on the applications using RVD. I need to get a network trace of everything involved in this scenario and see if I can figure out what the client is waiting for, but not getting, that is causing it to timeout and disconnect. Let me ask you this: On the TCP connection between the client and the LB, (assuming a ONECONNECT setup), does the LB immediately ACK the TCP packets to the client? Or does the backend (server side) ACK get sent back?

 

 

Thanks again for your help.

 

 

Dave