Message-based load balancing problem

Got some inspiration from this https://devcentral.f5.com/Forums/tabid/53/afv/topic/aff/5/aft/2161558/Default.aspx So I tested by removing some TCP::collect from the irules. The delayed response problem disappeared after I commented the TCP::collect in rules handling SERVER_CONNECTED event. However I realized it was not message-based load balancing but connection based load balancing. I did add "mblb" profile to the virtual server. Obviously the irules I wrote to extract individual messages didn't work. But why?

Can someone explain to me when I should use TCP::collect? I found some examples call TCP::collect almost in every event declaration, which I don't understand. I personally think I only need to call it in CLIENT_ACCEPTED and CLIENT_DATA.

Below are my current irules:

when CLIENT_ACCEPTED {

set client_closed 0

TCP::collect

}

periodically checks if the client connection closed, if so, also closes associated server connection.

when CLIENT_CLOSED {

set client_closed 1

}

Triggered when a connection has been established with the target node

when SERVER_CONNECTED {

after 1000 -periodic if {$client_closed} {TCP::close}

TCP::collect

}

when CLIENT_DATA {

binary scan [TCP::payload] II head rlen

if {($head & 0x3) == 1} {

if {[TCP::payload length] < [expr {8 + $rlen}]} {

TCP::collect [expr {$rlen + 8 - [TCP::payload length]}]

return

}

}

TCP::release [expr {$rlen + 8}]

TCP::notify request

TCP::collect

}

when USER_REQUEST {

pool my_pool

}