Re: [ANN] netdev call - Jan 30th

[Jakub Kicinski <kuba@kernel.org>]

On Mon, 29 Jan 2024 11:20:57 -0800 Jakub Kicinski wrote:
> Hi,
> 
> The bi-weekly netdev call at https://bbb.lwn.net/b/jak-wkr-seg-hjn
> is scheduled tomorrow at 8:30 am (PT) / 5:30 pm (~EU).
> 
> We'll certainly go over CI updates, but please feel free to suggest
> other topics!
> 
> In terms of the review rotation - this week's reviewer is nVidia.

Meeting notes:

 * Tour of the test runner UIs:
   * System status and test summary:
     https://netdev.bots.linux.dev/status.html
   * Example of patchwork reporting (see the netdev/contest check):
     https://patchwork.kernel.org/project/netdevbpf/patch/20240127175033.9640-1-linus.luessing@c0d3.blue/
   * Clicking takes us to the list of tests run as part of the branch
     report:
     https://netdev.bots.linux.dev/contest.html?pw-n=0&branch=net-next-2024-01-29--21-00
   * Last but not least the UI for flaky tests:
     https://netdev.bots.linux.dev/flakes.html

 * Mojatatu is running TCD on their end, so the remote executor thing
   is working fine. Reminder that the data formats here:
   https://docs.google.com/document/d/1TPlOOvv0GaopC3fzW-wiq8TYpl7rh8Vl_mmal0uFeJc/
    * Pedro: TCD executor code is at https://github.com/p4tc-dev/tc-executor

 * What to do about slow tests? Split into a new group? Skip? Export a
   variable to let tests know that the perf is low?
    * Going with the export for now - KSFT_MACHINE_SLOW=yes Tests can
      either adjust their “acceptance criteria” down, or report XFAIL
    * Jesse: some of the perf tests can be tuned down with things like netem
    * Petr: we want them to run on HW as well, so they are kept simple
    * Matthieu: MPTCP selftests auto-detect kernel slow downs
      (kmemleak, lockdep, kasan, prove_locking, etc. by looking at
      kallsyms). We hesitated to add a check for Qemu without KVM but
      we recently modified the selftests to require less resources
    * Willem: how do we deal with latency / timing sensitive tests?
    * Jakub: same approach as perf tests for now, look for the env var

 * HW dependent tests get skipped for veth (ethtool, l3 stats)
    * Petr: goal was to skip cleanly on veth and fail when the env
      misses any tooling etc.
    * Jakub: makes sense but we do have drivers/net/ and could separate
      the tests since they have no chance of running 
    * Petr: with recent lib.sh changes it should be doable 

 * The NIPA repo has moved, Netronome has transferred the ownership to
   linux-netdev so we’ll use https://github.com/linux-netdev/nipa going
   forward.