mlnx infiniband mechanism error on neutron

[Federico Pinca]

HI

i think there are some missing packages on ubuntu openstack zed containers that prevent the mlnx_infiniband mechanism to work and start correctly

i've not fully investigated, but with first test it seems that, on the following containers:
neutron_eswitchd
neutron_sriov_agent
neutron_server
is not present libzmq5 as apt package

and on this container:
neutron_server
is missing the pip package networking-mlnx

Ubuntu 22.04.2 LTS
5.15.0-72-generic #79-Ubuntu SMP x86_64
Docker version 23.0.6, build ef23cbc

images from:
quay.io/openstack.kolla/neutron-server:zed-ubuntu-jammy

[Michal Nasiadka (mnasiadka)]

Since we have no option to test it out upstream - would you be willing to provide a patch?

[Federico Pinca (clash00)]

i'll try, i'll create a kolla dev env to build container , meanwhile i patched the container manually, in my deployment, and i'll try if sriov + infiniband works correctly in neutron with that added pkgs.

if it's ok i'll work on the patch, i think we must add on dockerfile templates the apt and pip pagkages, correct?

the errors on the container was, for
neutron_server "mlnx_infiniband mechanism not found", and the container continue restarting, after installing that pip pkgs the container seems fine
for
neutron_eswitchd "libmzq.so.5 not found" and the container crashed/restarting, after installing the apt pkgs the container seems fine

bye

[Federico Pinca (clash00)]

sorry , libzmq.so.5 not libmzq

[Federico Pinca (clash00)]

hi
to reproduce the bugs
deploy openstack with kolla-ansible

main settings to reproduce:
openstack_tag: zed-ubuntu-jammy
kolla_source_version: stable/zed
kolla_base_distro: ubuntu
enable_openstack_core: 'yes'
enable_neutron_sriov: "yes"
enable_neutron_mlnx: "yes"

also if you don't have an infiniband/ethernet mellanox with sriov enabled interface, you can see the container restarting with errors not related to the devices but for missing pkgs and lib

[Launchpad Janitor (janitor)]

[Expired for kolla because there has been no activity for 60 days.]