What were you trying to do that didn't work?

When we disable the storage network (or ceph-osds are unreachable), the guests (having a CEPH /dev/vdb attached but not mounted) running on RHEL9.2 EUS hypervisor will stop pinging after more or less 30 seconds.   When re-establishing the connectivity, normal guest operations are resumed.

The storage network is a VLAN attached to a bond where ceph traffic is routed , in this case vlan1002.   If we do a "ifdown vlan1002", 30 seconds after performing that action, the issue is reproduced. 

Is this expected behavior ?   The customer would like to know why this is expected if in RHEL8 / previous ceph versions this was not a problem and would like this issue to be fixed if possible because they have lots of attached volumes that are not mounted and it causes them outages when that happens.

What is the impact of this issue to you?

Guests are not reachable via network when a ceph disk is attached to the guest (but not mounted) and the hypervisor connectivity to the storage backend is severed.

Please provide the package NVR for which the bug is seen:

Latest RHOSP17.1/RHEL9.2 E4S

How reproducible is this bug?:

Never on RHEL8.4 AUS but always on RHEL9.2E4S

Steps to reproduce

1. Create a VM booted on ephemeral local hypervisor disks (not in ceph)
2. Attach a /dev/vdb hosted on ceph (but do not mount it)
3. Kill the network between hypervisor and ceph cluster

Expected results

In RHEL 8.4 AUS, guests would remain up and running , answering to ICMP packets, etc

Actual results

In RHEL 9.2 E4S, guests stops pinging until the storage network connectivity is re-established.