Ask Your Question

luowei's profile - activity

2020-01-07 06:48:17 -0500 received badge  Notable Question (source)
2020-01-07 06:48:17 -0500 received badge  Famous Question (source)
2018-12-06 13:02:29 -0500 answered a question Can't create instances larger 1TB Memory

For Queens built on Ubuntu 18.04, here is a hack that worked for me. After applying, all VMs will have 46 hardware physical bits needed for >1TB RAM. I was able to create 2TB RAM VMs (tested with Windows 10, Ubuntu, CentOS,and cirros ;) VMs).

For big memory VMs, using Queens built on Unbuntu 18.04
On the compute nodes:

1) check that there is a "hw physical bits" type
   # kvm -M ? | grep hpb
   pc-i440fx-bionic-hpb Ubuntu 18.04 PC (i440FX + PIIX, +host-phys-bits=true, 1996)
   pc-q35-bionic-hpb    Ubuntu 18.04 PC (Q35 + ICH9, +host-phys-bits=true, 2009)

2) # cd /usr/lib/python2.7/dist-packages/nova/virt/libvirt

3) # cp
4) Add the hack below to
   if self.os_mach_type is not None:
       type_node.set("machine", self.os_mach_type)
   ### begin hack
       type_node.set("machine", "pc-i440fx-bionic-hpb")
   ### end hack
5) # service nova-compute restart
2018-12-05 13:28:46 -0500 commented question Can't create instances larger 1TB Memory

I have the same problem. Did you ever find a fix or work-around? thx

2018-11-11 15:48:36 -0500 received badge  Nice Answer (source)
2018-11-06 13:34:51 -0500 commented answer PXE/TFTP boot VM Instance within OpenStack

I know you said you don't have any code to share, but can you share the "DHCP options we redirect the VMs to our own PXE/TFTP server"? Thanks.

2018-09-07 14:50:16 -0500 asked a question Get rid of Ubuntu dashboard theme on Queens

Every version, Ubuntu has been making it harder to remove their dashboard theme as the default. In a Queens vanilla tutorial build on ubuntu, it seems they are ignoring the DEFAULT_THEME and AVAILABLE_THEMES settings in

Does anyone have a work around to make the standard OpenStack "default" theme, the actual, err, default theme?

2018-07-24 06:27:39 -0500 received badge  Popular Question (source)
2018-06-23 11:18:22 -0500 answered a question queens linuxbridge tap interface mtu not set to jumbo

This was solved by deleting and recreating the provider network. The provider network had been created before the neutron jumbo configuration was applied and hence was set to a mtu of 1500.

2018-06-23 10:19:52 -0500 asked a question queens linuxbridge tap interface mtu not set to jumbo

On a standard Queens Ubuntu 18.04 tutorial build, the linuxbridge provider tap interface mtus are not being set to jumbo.

Settings for jumbo mtu are configured as usual:

neutron.conf:global_physnet_mtu = 9000
ml2_conf.ini:path_mtu = 9000
ml2_conf.ini:physical_network_mtus = provider:9000

Manually forcing the tap interface mtu to 9000 with ifconfig after the fact works.

2017-10-11 06:22:06 -0500 commented question My VM can not be resize, but the flavor is subsistent exist.

Are you able to migrate an instance? Since resize uses migration, make sure migrate works.

2017-08-26 07:45:22 -0500 commented answer ERROR: could not access cell mapping database - has api db been created?

It's normal to get the cell error you mentioned above when installing the nova compute package. Look elsewhere for the cause of the 503 error.

2017-08-25 10:05:51 -0500 answered a question ERROR: could not access cell mapping database - has api db been created?

Just keep going with the install. The compute node placement stuff gets setup further down the guide.

2017-07-27 21:23:13 -0500 answered a question Why does a compute node require at least 2 network interfaces (minimal architecture ) ??

One for the management network and one for the provider network. If I only have one interface, I just use vlan subinterfaces.

2017-07-27 09:02:08 -0500 commented question Linux bridge agent on compute node

Did you install openstack-selinux?

2017-07-26 14:13:25 -0500 commented question Linux bridge agent on compute node

local_ip should be the ip address of the compute node. If, as in the tutorial guides, you are using the management network for tunnels then change to the COMPUTE node mgt. interface ip and restart the linuxbridge service on the compute node.

2017-07-24 13:42:09 -0500 answered a question why am i not getting the option "Create New Volume checked" and "volume" while lanuching instance in openstack ocata?

The horizon .js code currently only checks for cinder v1 endpoints to decide whether to display the options. The work around I used is to just go ahead and add the v1 endpoints. On controller:

. admin-openrc
openstack endpoint create --region RegionOne   volume public http://controller:8776/v1/%\(project_id\)s
openstack endpoint create --region RegionOne   volume internal http://controller:8776/v1/%\(project_id\)s
openstack endpoint create --region RegionOne   volume admin http://controller:8776/v1/%\(project_id\)s
2017-07-22 11:21:24 -0500 commented answer IPv6 Support in Ocata

The controller (network) node really isn't involved in ipv6 in this case. I'm on ubuntu, so "apt list --installed" gives:

2017-07-22 11:12:10 -0500 commented answer IPv6 Support in Ocata

The compute ifconfig excerpt looks like:

eth0 Link encap:Ethernet  HWaddr f4:e9:d4:ee:64:40  
          inet6 addr: fe80::f4e9:d4ee:feae:6440/64 Scope:Link
2017-07-22 11:10:33 -0500 commented answer IPv6 Support in Ocata

See added details above. Controller and compute have self assigned ipv6 link addresses. I don't want them to get global addresses, because then they are exposed. So I disable ipv6 autoconfiguration:

2017-07-21 12:06:23 -0500 commented question DHCP not providing IPv6 address to VM

This article describes tenant ipv6 deployment with good example commands and explanations for the different options. It is from 2015 but still helpful.

2017-07-20 19:58:14 -0500 commented answer cpu overcommit does not work

I think you also need at least the AvailabilityZoneFilter.

2017-07-20 09:03:26 -0500 received badge  Associate Editor (source)
2017-07-20 08:50:39 -0500 answered a question IPv6 Support in Ocata

The easy way to give a tenant ipv6 support is to do nothing! Just create the tenant on a provider network that is ipv6 enabled. Linuxbridge passes the ipv6 packets out of the box. When I say the provider network is ipv6 enabled, I mean the external network is ipv6 enabled (i.e. my campus subnet has a radvd daemon running, ipv6 tunnel, or whatever is your use case).

For example, on an standard Ocata ubuntu tutorial build with no special ipv6 configuration added, the cirros tenant picks up necessary ipv6 info from the external provider network (addresses changed to protect the innocent):

$ ifconfig
  eth0      Link encap:Ethernet  HWaddr FA:16:3E:C7:66:55  
            inet addr:  Bcast:  Mask:
            inet6 addr: 2011:470:c092:2bf:cafe:3eff:fec7:6655/64 Scope:Global
            inet6 addr: fe80::cafe:3eff:fec7:6655/64 Scope:Link

$ ping6 -c 3
PING (2607:f8b0:400f:800::200e): 56 data bytes
64 bytes from 2607:f8b0:400f:800::200e: seq=0 ttl=54 time=50.347 ms
64 bytes from 2607:f8b0:400f:800::200e: seq=1 ttl=54 time=50.174 ms
64 bytes from 2607:f8b0:400f:800::200e: seq=2 ttl=54 time=50.483 ms

You won't see the ipv6 address in the horizon dashboard listing, but as you can see above the tenant gets a global address and has ipv6 connectivity. To add ipv6 security group rules, use ::0 instead of for the remote prefix.

The key is ipv6 on your network segment OUTSIDE of OpenStack must work. So, if I have campus vlan 600 and I plug a laptop into that segment, I should get a global ipv6 address and have v6 connectivity. In my case I'm using slaac on my campus segment and an radvd daemon on a linux box. And, of course that segment also has ipv4. My /etc/radvd.conf looks like this (address changed):

interface em3.600
   AdvSendAdvert on;
   prefix 2011:670:cafe:2cf::/64
        AdvOnLink on;
        AdvAutonomous on;

I use this network segment as my provider network. I follow the vanilla ubuntu tutorial build: So, the interface that is connected to the campus vlan 600 segment is just what you would normally have on the controller (network) and computes for provider:

auto eth0
  iface eth0 inet manual
  up ip link set dev $IFACE up
  down ip link set dev $IFACE down

I don't change anything from the tutorial guide when I configure neutron or when I create the provider network. I just create it for ipv4 (NO ipv6 subnet, NO slaac options, NO dhcp6 options). ipv4 only, like in the guide:

When the tenant gets created on the provider network, it gets it's ipv4 address and info from neutron, but ipv6 router advertisements and neighbor discovery will ... (more)

2017-07-19 21:41:06 -0500 commented answer Devstack live-migration configuration problem

I don't know devstack well enough to answer, sorry. From what you say it sounds like devstack doesn't need a nova user account. In my standard build passwordless ssh for the root user was needed for live migration. Also, try with the "Block Migration" checked if you don't use shared instance storage

2017-07-19 20:39:49 -0500 answered a question cpu overcommit does not work

I think the filter names changed. Should have something in /etc/nova/nova.conf like:



btw, you can't just put the ComputeFilter. You need some of the others for it to work.

2017-07-19 17:43:38 -0500 received badge  Commentator
2017-07-19 17:43:38 -0500 commented question novnc did not respond when enter command

I noticed the same thing on my builds (so I switched to spice console). However,I don't recall if I only had the problem with Chrome or all browsers.

2017-07-19 17:27:03 -0500 commented answer Devstack live-migration configuration problem

You also need to setup passwordless root ssh between all compute nodes. See:

(Whereas for "resize" functionality you also need passwordless nova ssh between compute nodes).

2017-07-19 17:16:03 -0500 edited answer Devstack live-migration configuration problem

I haven't tried this on devstack, but on a plain ubuntu tutorial build you need to:

Edit /etc/libvirt/libvirtd.conf to add/change:
listen_tls = 0
listen_tcp = 1
auth_tcp = "none"

Then edit /etc/default/libvirtd to add:

/etc/init.d/libvirtd restart