Bug 1928312

Created attachment 1756726 [details]
lateset sosreport

Description of problem:

Customer is attempting to deploy a RHV 4.4 HE environment, and after installing  "dnf remove cockpit-ovirt-dashboard -y" and rebooting the customer loses network access:


Version-Release number of selected component (if applicable):

NAME="Red Hat Enterprise Linux"
VERSION="8.3 (Ootpa)"
ovirt-hosted-engine-setup-2.4.9-2.el8ev.noarch              
ovirt-host-4.4.1-4.el8ev.x86_64                             
vdsm-4.40.40-1.el8ev.x86_64
lldpad-1.0.1-13.git036e314.el8.x86_64

How reproducible:
Always

Steps to Reproduce:

1. Host
Network Adapter Intel
driver: ice
version: 0.8.2-k
firmware-version: 2.10 0x80004341 1.2789.0

2. LLDP is enabled for all interfaces on this Juniper ex4550 switch 

3. Install/update dnf packages dashboard per instructions.

https://access.redhat.com/documentation/en-us/red_hat_virtualization/4.4/html/installing_red_hat_virtualization_as_a_self-hosted_engine_using_the_cockpit_web_interface/installing_the_self-hosted_engine_deployment_host_she_cockpit_deploy#Installing_Cockpit_on_Linux_Hosts_SHE_deployment_host

4.2.1. Enabling the Red Hat Enterprise Linux host Repositories

Ensure that all packages currently installed are up to date:

# dnf upgrade --nobest
Reboot the machine.

3. Go to messages log verify:

~~~
Feb  9 22:34:44 rhev02 lldpad[2891]: get_dcbx_hw: enp67s0f1 nlconnect failed abort get ieee, Unspecific failure
Feb  9 22:34:44 rhev02 lldpad[2891]: set_ieee_hw: enp67s0f0 nlconnect failed abort hardware set, Unspecific failure
Feb  9 22:34:44 rhev02 lldpad[2891]: set_ieee_hw: enp67s0f0 nlconnect failed abort hardware set, Unspecific failure
Feb  9 22:34:44 rhev02 lldpad[2891]: set_ieee_hw: enp67s0f0 nlconnect failed abort hardware set, Unspecific failure
~~~
and Repeating Error - 
~~~
Non-Contiguous TCs - Disabling DCB
~~~

4. Proceed with host installation step to install dashboard

4.2.2. Installing Cockpit on Red Hat Enterprise Linux hosts
Install the dashboard packages:

# dnf install cockpit-ovirt-dashboard

5. Test ssh connection to host.

6. Verify lldpad service.

* lldpad.service - Link Layer Discovery Protocol Agent Daemon.
   Loaded: loaded (/usr/lib/systemd/system/lldpad.service; enabled; vendor preset: enabled)
   Active: active (running) since Thu 2021-02-11 16:49:56 CST; 17h ago
 Main PID: 2883 (lldpad)
    Tasks: 1 (limit: 3298746)
   Memory: 6.6M
   CGroup: /system.slice/lldpad.service
           `-2883 /usr/sbin/lldpad -t

Feb 11 16:57:04 rhev02 lldpad[2883]: l2_packet_send - send: Network is down
Feb 11 16:57:21 rhev02 lldpad[2883]: recvfrom(Event interface): No buffer space available
Feb 11 16:57:33 rhev02 lldpad[2883]: l2_packet_send - send: Network is down
Feb 11 16:57:33 rhev02 lldpad[2883]: recvfrom(Event interface): No buffer space Unit lvm2-activation-early.service could not be found.
Unit lvm2-activation.service could not be found.
available
Feb 11 16:57:58 rhev02 lldpad[2883]: l2_packet_send - send: Network is down
Feb 12 10:40:20 rhev02 lldpad[2883]: l2_packet_send - send: Network is down
Feb 12 10:41:07 rhev02 lldpad[2883]: recvfrom(Event interface): No buffer space available
Feb 12 10:41:28 rhev02 lldpad[2883]: l2_packet_send - send: Network is down
Feb 12 10:41:28 rhev02 lldpad[2883]: recvfrom(Event interface): No buffer space available
Feb 12 10:41:56 rhev02 lldpad[2883]: l2_packet_send - send: Network is down

Actual results:

Customer lost communication to the host. And seemed hit a known issue with lldpad in RHE7 https://bugzilla.redhat.com/show_bug.cgi?id=1554110. The only way to get communication to work to resume deployment is to do the following:

[1] Applied runtime disabling at the interface but does not survive a reboot and defaults back to 'rx'.

#lldptool set-lldp –i enp67s0f0 adminStatus=disasbled

[2] Customer disabled LLDP at the switch for both interfaces using the Intel NIC.

And is not seeing the non-contiguous error after reboot or lldpad errors.


Expected results:

It seems there were some changes with the lldpad in RHEL 8. In RHEL 7 I was able to disable but no more in RHEL 8 goes back to adminStatus=rx. There is like a strict dependency now between vdsmd and lldpad.

Would it be practical to still have this feature? 
What if the customer could not disable lldp?
Is there a purpose now in RHEL 8 not to disable?
Or, is there a method for this that I missed?

I realize also there might be a compatibility issue with Juniper switch as well. But my main concern is why we can not disable this.


Additional info: