Page MenuHomePhabricator

relocate/reimage cloudvirt1008 with 10G interfaces
Closed, ResolvedPublic

Description

This host only holds a canary instance; it can be moved to eqiad1-r any time.

This includes renaming the host from labvirt1008 to cloudvirt1008.

  • disable puppet in labvirt1008
  • drain labvirt1008
  • update RAID configuration to create spare disks
  • re-rack host in 10G rack
  • - connect primary 10G port and enable PXE (right now its connected but PXE is not enabled)
  • - connect secondary 10G port and setup switch port
  • - assign back to @RobH for followup after the above is set
  • merge puppet patch to rename, get the new debian installer working and disable notifications (rename hieradata/hosts/labvirt1009yaml to cloudvirt1009.yaml and add "profile::base::notifications: disabled" temporarily)
  • merge dns patch to add the new FQDNs (partial, the old mgmt names still remains)
  • re-image
  • merge puppet patch to re-enable notifications (remove "profile::base::notifications")
  • netbox update https://netbox.wikimedia.org/dcim/devices/1452/
  • update docs https://wikitech.wikimedia.org/wiki/Portal:Cloud_VPS/Admin/Deployments
  • physical relabeling and switch port description T220443
  • done

Event Timeline

Andrew updated the task description. (Show Details)

Mentioned in SAL (#wikimedia-operations) [2019-02-22T14:03:51Z] <moritzm> removed labvirt1008 from debmonitor (T216661)

attempting to silence some alerts (I think a result of an expiring downtime) I restarted this and it had a hard time coming up, complaining about a failure to mount drives. I told it to skip the mount and now that it's up it looks fine but this needs more investigation.

the next step is moving this to a new rack for 10G connections (either 2, 4 or 7 in row B) so I'm tagging dc-ops. You can hand it back to me for the rename/rebuild once it's in place.

jbond triaged this task as Medium priority.Mar 4 2019, 7:45 PM

Change 499480 had a related patch set uploaded (by Andrew Bogott; owner: Andrew Bogott):
[operations/puppet@production] installer: prepare to rebuild labvirt1008 as cloudvirt1009

https://gerrit.wikimedia.org/r/499480

Change 499480 merged by Andrew Bogott:
[operations/puppet@production] installer: prepare to rebuild labvirt1008 as cloudvirt1009

https://gerrit.wikimedia.org/r/499480

Mentioned in SAL (#wikimedia-operations) [2019-04-02T07:52:57Z] <moritzm> removed labvirt1008 from debmonitor (T216661)

Change 500800 had a related patch set uploaded (by Andrew Bogott; owner: Andrew Bogott):
[operations/dns@master] Rename labvirt1008 to cloudvirt1008

https://gerrit.wikimedia.org/r/500800

Change 500800 merged by Andrew Bogott:
[operations/dns@master] Rename labvirt1008 to cloudvirt1008

https://gerrit.wikimedia.org/r/500800

So when attempting to setup PXE on this, the network device 2 (ie the 10G interface) isn't showing as a bootable device. The identical system (cloudvirt1009) shows the bootable device referenced, so it seems this system didnt get PXE enabled/saved in bios.

RobH renamed this task from cloudVPS: drain and rebuild labvirt1008 as cloudvirt1008 to relocate/reimage cloudvirt1009 with 10G interfaces.Apr 4 2019, 6:31 PM
RobH renamed this task from relocate/reimage cloudvirt1009 with 10G interfaces to relocate/reimage cloudvirt1008 with 10G interfaces.
RobH assigned this task to Cmjohnson.
RobH updated the task description. (Show Details)
Cmjohnson subscribed.

Assigning back to robh NIC has been enabled to PXE, second cable has been run, port description updated and added to cloud-virt-instance-trunk

I'll describe the issue I'm seeing, and how I've troubleshot it, so far to no avail:

cloudvirt1008 will PXE boot, and does so successfully off its 10G interface. If I login to the https://cloudvirt1008.mgmt.eqiad.wmnet mgmt interface, I can see the MAC addresses for the network devices:

10G NIC port 1 (of 2) MAC : f0:92:1c:05:bc:d8
10G NIC port 2 (of 2) MAC : f0:92:1c:05:bc:dc

I boot the system into the one time boot of 'network device 1' and it shows it attempting to network boot with MAC address f0:92:1c:05:bc:d8. We can even see it hit the DHCP server:

Apr  5 18:40:23 install1002 dhcpd: labvirt1009.eqiad.wmnet: host unknown.
Apr  5 18:40:23 install1002 dhcpd: DHCPDISCOVER from f0:92:1c:05:bc:d8 via 10.64.20.2: network 10.64.20.0/24: no free leases
Apr  5 18:40:23 install1002 dhcpd: DHCPDISCOVER from f0:92:1c:05:bc:d8 via 10.64.20.3: network 10.64.20.0/24: no free leases
Apr  5 18:40:27 install1002 dhcpd: DHCPDISCOVER from f0:92:1c:05:bc:d8 via 10.64.20.2: network 10.64.20.0/24: no free leases
Apr  5 18:40:27 install1002 dhcpd: DHCPDISCOVER from f0:92:1c:05:bc:d8 via 10.64.20.3: network 10.64.20.0/24: no free leases

The first line is the kicker, it points out a bad entry trying to reference labvirt1009.eqiad.wmnet. Looking in the lease file shows the mistake, someone seems to have introduced a copy/paste error. (I'm not bothering to run the git log to find blame because who cares.) Fix following shortly.

Change 501666 had a related patch set uploaded (by RobH; owner: RobH):
[operations/puppet@production] cloudvirt1008 dhcp lease file correction

https://gerrit.wikimedia.org/r/501666

Change 501666 merged by RobH:
[operations/puppet@production] cloudvirt1008 dhcp lease file correction

https://gerrit.wikimedia.org/r/501666

RobH removed RobH as the assignee of this task.Apr 10 2019, 10:17 PM