Ironic grub UEFI PXE booting fails

[xinliang]

Ironic grub UEFI PXE booting fails when launching bare metal instance

Step to reproduce
=================
1, Deploy ironic uefi grub mode with kolla-ansible
2, Enrol baremetal node following bellow guide
https://docs.openstack.org/ironic/latest/install/configure-pxe.html

openstack baremetal node create --driver ipmi --name $NODE_NAME \
  --driver-info ipmi_username=$IPMI_USER \
  --driver-info ipmi_password=$IPMI_PASSWD \
  --driver-info ipmi_address=$IMPI_ADDR \
  --resource-class baremetal-resource-class --property cpus=$CPU_OUNT \
  --property memory_mb=$RAM_MB --property local_gb=$DISK_GB \
  --property cpu_arch=aarch64 --property capabilities='boot_mode:uefi' \
  --driver-info deploy_kernel=$DEPLOY_VMLINUZ_UUID \
  --driver-info deploy_ramdisk=$DEPLOY_INITRD_UUID \
  --deploy-interface direct

NODE_UUID=$(openstack baremetal node show $NODE_NAME -c uuid| awk '/id/ {print $4}')
openstack baremetal port create $MAC_ADDRESS --node $NODE_UUID --physical-network physnet1

openstack baremetal node manage $NODE_UUID
openstack baremetal node provide $NODE_UUID
3, Launch bare metal instance
openstack server create --image cirros --flavor D05 \
  --key-name mykey --network public1 demo1

looking at serial console, found it stop at pxe booting and fails.

[xinliang (xin3liang)]

￥ docker logs ironic_dnsmasq
dnsmasq-dhcp: DHCPDISCOVER(eth0) 00:18:85:04:00:e2 ignored
dnsmasq-dhcp: DHCPDISCOVER(eth0) 00:18:85:04:00:e2 ignored
dnsmasq-dhcp: DHCPDISCOVER(eth0) 00:18:85:04:00:e2 ignored
dnsmasq: inotify, new or changed file /etc/dnsmasq/dhcp-hostsdir/00:18:85:04:00:e2
dnsmasq-dhcp: read /etc/dnsmasq/dhcp-hostsdir/00:18:85:04:00:e2
dnsmasq-dhcp: DHCPDISCOVER(eth0) 00:18:85:04:00:e2 ignored
dnsmasq-dhcp: DHCPDISCOVER(eth0) 00:18:85:04:00:e2 ignored
dnsmasq-dhcp: DHCPDISCOVER(eth0) 00:18:85:04:00:e2 ignored
dnsmasq-dhcp: DHCPDISCOVER(eth0) 00:18:85:04:00:e2 ignored

$ (ironic-dnsmasq)[root@j12-d05-07 /]# cat /etc/dnsmasq/dhcp-hostsdir/00\:18\:85\:04\:00\:e2
00:18:85:04:00:e2,ignore

Found inspector put node mac to irgnore.

[Mark Goddard (mgoddard)]

Inspector should be ignoring this DHCP request. When using nova & neutron, it is neutron that should provide the DHCP response on the provisioning network.

[xinliang (xin3liang)]

So current kolla-ansible ironic deployment looks like a standalone mode? Has an external dnsmasq DHCP server and not using with neutron dhcp provider. Maybe we should set dhcp provider to none for ironic conductor in this case, which koll-ansible should handle this. If we use neutron then we should not set up a external dnsmasq DHCP server.

[xinliang (xin3liang)]

Even though I set dhcp_provider to none inspector still ignores the node's mac too。

[dhcp]
dhcp_provider = none

[Mark Goddard (mgoddard)]

Kolla ansible can deploy ironic in a standalone or nova-backend mode. If nova and ironic are both enabled, ironic will be configured to work with nova. The ironic dnsmasq server is just for inspection, it is not used during provisioning.

[Mark Goddard (mgoddard)]

I would guess the part you are missing is having network connectivity from the bare metal node's provisioning network through to neutron. Usually this is a flat or VLAN network.

[xinliang (xin3liang)]

Thanks mark, if ironic-dnsmasq is just for inspection , this explains all.
Then I will need to setup neutron flat network and use neutron dhcp for provisioning :)

[xinliang (xin3liang)]

After setup ironic provision flat network and subnet dhcp, provisioning bare metal machine works.