>
> In this case, I was getting message build-up even with when the
> subscriber used a queue size of one. It seems that the teleop message
> where small enough to accumulate in the TCP kernel or HW buffer and
> not being discarded while the callback was running. This can be easily
> identified by looking at the message timestamp. In our case, the
> message age was increasing (obviously, you can only evaluate that on
> machines with synchronised clocks).
>

If you have a way to reproduce this, please file a ticket -- roscpp should
be draining the sockets as quickly as possible whether or not your callbacks
take a long time.  They should be totally decoupled.

Josh