On Tue, 30 Dec 2025 09:28:50 -0800 Stephen Hemminger wrote:

It is worth testing for the case where netem is used as a leaf qdisc.
I worry that this could cause the parent qdisc to get accounting wrong.
I.e if HTB calls netem and netem queues 2 packets, the qlen in HTB
would be incorrect.

Indeed, presumably backlog should be corrected with
qdisc_tree_reduce_backlog() ? Either way it'd be great to add / adjust 
some test cases to confirm backlogs are 0 after the traffic stopped.