In the case that a compute host with multiple NICs can be connectted to
iscsi storage network, we expect the storage traffic only be
transimitted via storage NIC. This need to define an custom iface and set the
iface.net_ifacename value to the storage interface name.
Unfortunately, the current iscsi connector does not support to use a
custom iface. The _validate_iface_transport will change it to default.
After add tcp to the supported_transports, it is possible to use a
custom iface.
Please get the details as below:
1. CMP001 has three NICs: NIC1 is for management network, NIC2 is
for tenant network, and NIC3 is for storage network. NIC3 can access
storage device via layer 2 network. And NIC1 can access storage device
too, but it can only access via layer 3 network. We hope all the storage
traffic only be transimitted via NIC3, because they are in the same
layer 2 network.
2. Get the NIC3 MAC:
fa:16:3e:79:fd:63
3. Add new iscsi iface:
# iscsiadm -m iface --op=new -I tcp.fa:16:3e:79:fd:63
# iscsiadm -m iface -I tcp.fa:16:3e:79:fd:63 --op=update
-n iface.hwaddress -v fa:16:3e:79:fd:63
4. Edit nova.conf, and change iscsi_iface value:
iscsi_iface = tcp.fa:16:3e:79:fd:63
5. Restart nova compute service:
# systemctl restart openstack-nova-compute
Without this change, _validate_iface_transport will change the
custom iface tcp.fa:16:3e:79:fd:63 to default due to it cannot find tcp
transport.
According to this change, we should have a corresponding change to the
nova docs, here is change to nova
https://review.openstack.org/#/c/524443/
Change-Id: I85a62cb6e7f8d8982e97e792d647d38ce5641060
Closes-Bug:#1722432
cfe5b19fad