mira, smithi, and gibba are automatically reimaged by FOG when locked using teuthology-lock
.
Prior to 2018, we didn't have FOG and testnodes would need to be reimaged every so often using cobbler when the OS would get too mangled to re-use. The dhcp server is configured to point smithi and mira to the FOG PXE server. All other machine types can use Cobbler.
Cobbler is configured using ceph-cm-ansible and primarily to install on /dev/sda
on a machine with a single NIC. If a machine has a more complex configuration than that, the Manual method below is recommended.
For machines that typically use FOG, you will need to tell the DHCP server to point whatever machine you're reimaging to Cobbler's PXE server instead of FOG's. You can do this with set-next-server-local.sh. e.g.,
dgalloway@p50 ~ ()$ set-next-server-local.sh smithi071 cobbler + ssh store01.front.sepia.ceph.com 'sudo /usr/local/sbin/set-next-server.sh smithi071 cobbler && sudo service dhcpd restart' + dhcpconfig=/etc/dhcp/dhcpd.front.conf ++ date +%s + timestamp=1656427483 + cobblerip=172.21.0.11 + cobblerfilename=/pxelinux.0 + fogip=172.21.0.72 + fogfilename=/undionly.kpxe ++ sed -n '/host smithi071-front/,/}/p' /etc/dhcp/dhcpd.front.conf ++ grep 'hardware ethernet' ++ awk '{ print $3 }' ++ tr -d ';' + macaddr=0c:c4:7a:8f:d7:65 ++ sed -n '/host smithi071-front/,/}/p' /etc/dhcp/dhcpd.front.conf ++ grep fixed-address ++ awk '{ print $2 }' ++ tr -d ';' + ipaddr=172.21.15.71 ++ grep -n smithi071 /etc/dhcp/dhcpd.front.conf ++ cut -d : -f1 + linenum=2111 + '[' -z 0c:c4:7a:8f:d7:65 ']' + '[' -z 172.21.15.71 ']' + '[' -z 2111 ']' + cp /etc/dhcp/dhcpd.front.conf /etc/dhcp/dhcpd.front.conf_1656427483.bak + sed -i '/host smithi071-front {/,/}/d' /etc/dhcp/dhcpd.front.conf + '[' cobbler == cobbler ']' + sed -i '2111 i \ host smithi071-front {\n\ hardware ethernet 0c:c4:7a:8f:d7:65;\n\ fixed-address 172.21.15.71;\n\ next-server 172.21.0.11;\n\ filename "/pxelinux.0";\n\ }' /etc/dhcp/dhcpd.front.conf + dhcpd -q -t -cf /etc/dhcp/dhcpd.front.conf + '[' 0 '!=' 0 ']' + rm /etc/dhcp/dhcpd.front.conf_1656427483.bak Redirecting to /bin/systemctl restart dhcpd.service
Be sure to set it back to FOG when you're done!
If you don't already have a Cobbler Web UI account, request one by filing a ticket.
DE:AD:BE:EF:AF:AF
or something else you know is bogus (this makes sure the PXE server doesn't recognize the machine you're reimaging after it reboots so that you get the Cobbler Menu)target.ipmi.sepia.ceph.com
)initrd=
and vmlinuz=
parameters.inst.repo=http://172.21.0.11/cobbler/ks_mirror/DISTRO
(e.g., inst.repo=http://172.21.0.11/cobbler/ks_mirror/CentOS-7.5-x86_64
)You will need to have sudo access on the cobbler host. This is typically reserved for Infra Admins.
ssh cobbler.front.sepia.ceph.com # To see a list of available Profiles/Distros, sudo cobbler profile list # To list the systems sudo cobbler system list # After you've decided on a profile, sudo cobbler system edit --name $SHORT_HOSTNAME --profile=$PROFILE_NAME --netboot-enabled=1 # Example: sudo cobbler system edit --name smithi001 --profile=CentOS-7.2-x86_64 --netboot-enabled=1 # Example for reimaging a large number of testnodes at once: for host in smithi{001..030}; do sudo cobbler system edit --name $host --profile=CentOS-7.2-x86_64 --netboot-enabled=1; ipmitool -I lanplus -U inktank -P XXXXX -H $host.ipmi.sepia.ceph.com chassis power cycle; done
Now power cycle the testnode.