No edit summary |
|||
| Line 14: | Line 14: | ||
$> sudo yum -y openstack-utils qpid-cpp-server avahi | $> sudo yum -y openstack-utils qpid-cpp-server avahi | ||
== | == Adjust qpidd authentication == | ||
Ensure auth=no is set in /etc/qpidd.conf | Ensure auth=no is set in /etc/qpidd.conf | ||
Revision as of 19:28, 2 November 2012
Basic Setup
These steps will setup OpenStack services to be accessed by the OpenStack dashboard web UI on a *single node*, and also show steps to launch our first instance (virtual machine).
Many of the examples here require 'sudo' to be properly configured, please see Configuring Sudo if you need help.
Enable EPEL repositories
$> sudo rpm -Uvh http://download.fedoraproject.org/pub/epel/6/i386/epel-release-6-7.noarch.rpm
First let's pull in some preliminary dependencies
$> sudo yum -y openstack-utils qpid-cpp-server avahi
Adjust qpidd authentication
Ensure auth=no is set in /etc/qpidd.conf
Disable mDNS (optional)
As per Bug #746111, unless you have installed and are using avahi-libs, you should disable mdns_adv in /etc/libvirt/libvirtd.conf
$> sudo sed -i 's/#mdns_adv = 0/mdns_adv = 0/' /etc/libvirt/libvirtd.conf
(Note: This has been fixed upstream, but may still affect older versions)
Setup DHCP control depending on distro
On RHEL 6.2 or below:
$> sudo openstack-config --set /etc/nova/nova.conf DEFAULT force_dhcp_release False
On RHEL 6.3 or above:
$> sudo yum install dnsmasq-utils # from the Red Hat optional channel
Start support services
Nova requires dbus to be running:
$> sudo service messagebus start && sudo chkconfig messagebus on
Install and Setup OpenStack on a single node
This script will install nova, glance, quantum, cinder, swift, keystone, dashboard. Note it configures and enables all but the quantum service at this time
sudo openstack-demo-install
Now copy the keystonerc file created by the install script above to your home directory, so that it can be easily sourced to provide the credentials used by the various OpenStack command line utilities.
sudo cp /root/keystonerc ~
Installing within a VM
If you are testing OpenStack in a virtual machine, openstack-demo-install will have already detected this and configured nova to use qemu without KVM and hardware virtualization. Note nested virtualization will be the much slower TCG variety, and you should provide lots of memory to the top level guest, as the openstack created guests default to 2GM RAM with no overcommit.
We still need some manual config to work around a libvirt issue fixed in RHEL 6.4.
$> sudo ln -s /usr/libexec/qemu-kvm /usr/bin/qemu-system-x86_64 $> sudo service libvirtd restart $> sudo service openstack-nova-compute restart
Enable external access to OpenStack Dashboard
To open up the firewall ports for HTTP:
$> sudo lokkit -p http:tcp $> sudo lokkit -p https:tcp
The dashboard should be accessible at http://$HOST/dashboard . Account and password should be what you configured for the keystone setup, which is displayed when openstack-demo-install completes.
Basic Folsom Operations
The operational instructions below have not been updated for Folsom at present. For Folsom specific details please see OpenStack Folsom test cases
Basic OpenStack Operations
The items below have only been tested on the Essex version of OpenStack, and may need adjustment for OpenStack Folsom.
Setup volume storage
The openstack-nova-volume service requires an LVM Volume Group called nova-volumes to exist. We simply create this using a loopback sparse disk image.
$> sudo dd if=/dev/zero of=/var/lib/nova/nova-volumes.img bs=1M seek=20k count=0 $> sudo vgcreate nova-volumes $(sudo losetup --show -f /var/lib/nova/nova-volumes.img)
Nova Network Setup
To create the network do:
$> sudo nova-manage network create demonet 10.0.0.0/24 1 256 --bridge=demonetbr0
NB the network range here, should *not* be the one used on your existing physical network. It should be a range dedicated for the network that OpenStack will configure. So if 10.0.0.0/24 clashes with your local network, pick another range
Register an Image
To run an instance, you are going to need an image. There are prebuilt Fedora 16 JEOS (Just Enough OS) images that can be downloaded. Note this will download a 200MB image (without a progress bar)
$> glance add name=f16-jeos is_public=true disk_format=qcow2 container_format=ovf \
copy_from=http://berrange.fedorapeople.org/images/2012-02-29/f16-x86_64-openstack-sda.qcow2
Launch an Instance
Create a keypair:
$> nova keypair-add mykey > oskey.priv $> chmod 600 oskey.priv
Configure key injection mode, to allow guestfs to inject into multiple guest types:
$> sudo openstack-config --set /etc/nova/nova.conf DEFAULT libvirt_inject_partition -1 $> sudo service openstack-nova-compute restart
Launch an instance:
$> nova boot myserver --flavor 2 --key_name mykey \
--image $(glance index | grep f16-jeos | awk '{print $1}')
And then observe the instance running, observe the KVM VM running and SSH into the instance:
$> sudo virsh list $> nova list $> ssh -i oskey.priv root@10.0.0.2 # use ec2-user instead of root if launching the image f16-x86_64-openstack-sda $> nova console-log myserver $> nova delete myserver
Additional Functionality
Using Eucalyptus tools
Set up a rc file for EC2 access (this expects a prior keystone configuration)
$> . ./keystonerc
$> USER_ID=$(keystone user-list | awk '/admin / {print $2}')
$> ACCESS_KEY=$(keystone ec2-credentials-list --user-id $USER_ID | awk '/admin / {print $4}')
$> SECRET_KEY=$(keystone ec2-credentials-list --user-id $USER_ID | awk '/admin / {print $6}')
$> cat > novarc <<EOF
export EC2_URL=http://localhost:8773/services/Cloud
export EC2_ACCESS_KEY=$ACCESS_KEY
export EC2_SECRET_KEY=$SECRET_KEY
EOF
$> chmod 600 novarc
$> . ./novarc
You should now be able to launch an image:
$> euca-run-instances f16-jeos -k nova_key $> euca-describe-instances $> euca-get-console-output i-00000001 $> euca-terminate-instances i-00000001
Images
Rather than the prebuilt Fedora 16 JEOS image referenced above, there are other image options.
- Building a Fedora 16 JEOS image using Oz
- Downloading ttylinux based minimal images used by OpenStack developers for testing
Building Fedora 16 JEOS Images With Oz
You can very easily build an image using Oz. First, make sure it's installed:
$> sudo yum install /usr/bin/oz-install
Create a template definition file called f16-jeos.tdl containing:
<template>
<name>fedora16_x86_64</name>
<description>My Fedora 16 x86_64 template</description>
<os>
<name>Fedora</name>
<version>16</version>
<arch>x86_64</arch>
<install type='url'>
<url>http://download.fedoraproject.org/pub/fedora/linux/releases/16/Fedora/x86_64/os/</url>
</install>
</os>
<commands>
<command name='setup-rc-local'>
sed -i 's/rhgb quiet/console=ttyS0/' /boot/grub/grub.conf
cat >> /etc/rc.local << EOF
if [ ! -d /root/.ssh ]; then
mkdir -p /root/.ssh
chmod 700 /root/.ssh
fi
# Fetch public key using HTTP
ATTEMPTS=10
while [ ! -f /root/.ssh/authorized_keys ]; do
curl -f http://169.254.169.254/latest/meta-data/public-keys/0/openssh-key > /tmp/aws-key 2>/dev/null
if [ \$? -eq 0 ]; then
cat /tmp/aws-key >> /root/.ssh/authorized_keys
chmod 0600 /root/.ssh/authorized_keys
restorecon /root/.ssh/authorized_keys
rm -f /tmp/aws-key
echo "Successfully retrieved AWS public key from instance metadata"
else
FAILED=\$((\$FAILED + 1))
if [ \$FAILED -ge \$ATTEMPTS ]; then
echo "Failed to retrieve AWS public key after \$FAILED attempts, quitting"
break
fi
echo "Could not retrieve AWS public key (attempt #\$FAILED/\$ATTEMPTS), retrying in 5 seconds..."
sleep 5
fi
done
EOF
</command>
</commands>
</template>
Then simply do:
$> sudo oz-install -d4 -u f16-jeos.tdl
Once built, you simply have to register the image with Nova:
$> glance add name=f16-jeos is_public=true container_format=bare disk_format=raw < /var/lib/libvirt/images/fedora16_x86_64.dsk $> glance index
The last command should return a list of the images registered with the Glance image registry.
Downloading Existing Images
If you don't need a functioning Fedora 16 and want the smallest possible images, just download this set of images commonly used by OpenStack developers for testing and register them with Nova:
$> mkdir images $> cd images $> curl http://images.ansolabs.com/tty.tgz | tar xvfzo - $> glance add name=aki-tty disk_format=aki container_format=aki is_public=true < aki-tty/image $> glance add name=ami-tty disk_format=ami container_format=ami is_public=true < ami-tty/image $> glance add name=ari-tty disk_format=ari container_format=ari is_public=true < ari-tty/image
Then to start the image:
$> euca-run-instances ami-tty --kernel aki-tty --ramdisk ari-tty -k mykey
Volumes
If you use the Chrome browser, kill it before embarking on this section, as it has been known to cause the lvcreate command to fail with 'incorrect semaphore state' errors.
Note when setting up volumes in production, make sure you don't put your volume nodes on the same network as your guests when using the default volume driver, as all the iscsi targets are discoverable and accessible without any security.
Start the SCSI target daemon
$> sudo systemctl start tgtd.service $> sudo systemctl enable tgtd.service
Create a new 1GB volume
$> VOLUME=$(euca-create-volume -s 1 -z nova | awk '{print $2}')
View the status of the new volume, and wait for it to become 'available'
$> watch "euca-describe-volumes | grep $VOLUME | grep available"
Re-run the previously terminated instance if necessary:
$> INSTANCE=$(euca-run-instances f16-jeos -k mykey | grep INSTANCE | awk '{print $2}')
or:
$> INSTANCE=$(euca-run-instances ami-tty --kernel aki-tty --ramdisk ari-tty -k mykey | grep INSTANCE | awk '{print $2}')
Make the storage available to the instance (note -d is the device on the compute node)
$> euca-attach-volume -i $INSTANCE -d /dev/vdc $VOLUME
ssh to the instance and verify that the vdc device is listed in /proc/partitions
$> cat /proc/partitions
Now make the device available if /dev/vdc is not already present
$> mknod /dev/vdc b 252 32
Create and mount a file system directly on the device
$> mkfs.ext3 /dev/vdc $> mkdir /mnt/nova-volume $> mount /dev/vdc /mnt/nova-volume
Display some file system details
$> df -h /dev/vdc
Create a temporary file:
$> echo foo > /mnt/nova-volume/bar
Terminate and re-run the instance, then re-attach the volume and re-mount within the instance as above. Your temporary file will have persisted:
$> cat /mnt/nova-volume/bar
Unmount the volume again:
$> umount /mnt/nova-volume
Exit from the ssh session, then detach and delete the volume:
$> euca-detach-volume $VOLUME $> euca-delete-volume $VOLUME
Floating IPs
You may carve out a block of public IPs and assign them to instances.
First thing you need to do is make sure that nova is configured with the correct public network interface. The default is eth0, but you can change it by e.g.
$> sudo openstack-config-set /etc/nova/nova.conf DEFAULT public_interface em1 $> sudo systemctl restart openstack-nova-network.service
Then you can do e.g.
$> sudo nova-manage floating create 172.31.0.224/28 $> euca-allocate-address $> euca-associate-address -i i-00000012 172.31.0.224 $> ssh -i nova_key.priv root@172.31.0.224 $> euca-disassociate-address 172.31.0.224 $> euca-release-address 172.31.0.224
VNC access
To setup VNC access to guests through the dashboard:
nova-novncproxy reads some parameters in /etc/nova/nova.conf file. First you need to configure your cloud controller to enable VNC
novncproxy_host = 0.0.0.0 novncproxy_port = 6080
and in the nova compute nodes you need something like this
novncproxy_base_url=http://NOVNCPROXY_FQDN:6080/vnc_auto.html vnc_enabled=true vncserver_listen=COMPUTE_FQDN vncserver_proxyclient_address=COMPUTE_FQDN
You should also make sure that openstack-nova-consoleauth has been started on the controller node:
$ controller> sudo /etc/init.d/openstack-nova-consoleauth restart
After restarting nova services on both nodes the newly created machines will run the qemu-kvm with a parameter -vnc compute_fqdn:display_number. Then after starting the novncproxy and connecting to the dashboard it will discover the host and point to the novncproxy with the appropriate values and connect to the VM.
Note ensure than the iptables entries for VNC ports (5900+DISPLAYNUMBER) are allowed.
Migrate and Resize
This is implemented currently by transferring the images between compute nodes over ssh. Therefore currently you need to make these adjustments on each compute node to allow that.
- Allow logins for the nova user
# usermod -s /bin/bash nova # su - nova $ mkdir .ssh && cd .ssh
- Disable host identity checking by adding this to ssh config
$ cat > config <<EOF Host * StrictHostKeyChecking no UserKnownHostsFile=/dev/null EOF
- Generate and distribute ssh key
$ ssh-keygen -f id_rsa -b 1024 -P $ scp /var/lib/nova/.ssh/id_rsa.pub root@otherHost:/var/lib/nova/.ssh/authorized_keys # chown nova:nova /var/lib/nova/.ssh/authorized_keys # semanage permissive -a sshd_t
To improve the SELinux config in future, the nova_var_lib_t context on /var/lib/nova will, need to be configured to allow search access by sshd_t. Also the ssh_home_t context will need to be associated with /var/lib/nova/.ssh
Live Migration of VM instances
First note the official OpenStack docs on the feature and a doc patch distinguishing libvirt live migration.
- Seting NFS server
- Make an nfs share with no_root_squash (nova uses root-wrap to chown the instance's disk to qemu:qemu)
- Make nova user and qemu user:
nova:x:162:162::/home/nova:/bin/bash qemu:x:107:107::/home/qemu:/bin/bash
- chown -R nova:nova /the/nfs/share
- Mount nfs share on each host at /var/lib/nova/instances
- Configure libvirt
- See the libvirt wiki as to how to create certificates.
- Edit /etc/libvirt/libvirt.conf
listen_tcp = 1 tcp_port = "16509" auth_tcp = "none"
- Edit /etc/sysconfig/libvirtd
LIBVIRTD_ARGS="--listen"
- Restart libvirtd & OpenStack compute services
Deployment
Adding a Compute Node
Okay, everything so far has been done on a single node. The next step is to add another node for running VMs.
Let's assume the machine you've set up above is called 'controller' and the new machine is called 'node'.
First, open the qpid, MySQL, Nova API and iSCSI ports on controller:
$ controller> sudo lokkit -p 3306:tcp $ controller> sudo lokkit -p 5672:tcp $ controller> sudo lokkit -p 9292:tcp $ controller> sudo lokkit -p 3260:tcp $ controller> sudo service libvirtd reload
Then make sure that ntp is enabled on both machines:
$> sudo yum install -y ntp $> sudo service ntpd start $> sudo chkconfig ntpd on
Install libvirt and nova on node:
$ node> sudo yum install --enablerepo=epel-testing openstack-nova python-keystone openstack-utils $ node> sudo service libvirtd start $ node> sudo chkconfig libvirtd on $ node> sudo setenforce 0
Configure nova so that node can find the services on controller:
$ node> sudo openstack-config --set /etc/nova/nova.conf DEFAULT qpid_hostname controller $ node> sudo openstack-config --set /etc/nova/nova.conf DEFAULT sql_connection mysql://nova:nova@controller/nova $ node> sudo openstack-config --set /etc/nova/nova.conf DEFAULT glance_api_servers controller:9292 $ node> sudo openstack-config --set /etc/nova/nova.conf DEFAULT iscsi_ip_prefix 172.31.0.107 $ node> sudo openstack-config --set /etc/nova/nova.conf DEFAULT auth_strategy keystone
(The {{{iscsi_ip_prefix}}} value is the IP address of the controller node)
Configure the Network interfaces The bridge name should match what use used in the nova-manage command on the controller
$ node> sudo openstack-config --set /etc/nova/nova.conf DEFAULT flat_network_bridge demonetbr0
The device which should be moved onto the bridge (nova will set up this bridge, once it done you can view it with the brctl command
$ node> sudo openstack-config --set /etc/nova/nova.conf DEFAULT flat_interface eth0 $ controller> sudo openstack-config --set /etc/nova/nova.conf DEFAULT flat_interface eth0
$ node> brctl show
Enable the compute service:
$ node> sudo service openstack-nova-compute start
Now everything should be running as before, except the VMs are launched either on controller or node. You will only be able to ping/ssh to vm's from the controller node.
Manual Setup of MySQL
As of openstack-nova-2011.3-9.el6 and openstack-nova-2011.3-8.fc16, openstack-nova is now set up to use MySQL by default. If you're updating an older installation or prefer to set up MySQL manually instead of using the openstack-nova-db-setup script, this section shows how to do it.
First install and enable MySQL:
$> sudo yum install -y mysql-server $> sudo service mysqld start $> sudo chkconfig mysqld on
Set a password for the root account and delete the anonymous accounts:
$> mysql -u root
mysql> update mysql.user set password = password('iamroot') where user = 'root';
mysql> delete from mysql.user where user = '';
Create a database and user account specifically for nova:
mysql> create database nova; mysql> create user 'nova'@'localhost' identified by 'nova'; mysql> create user 'nova'@'%' identified by 'nova'; mysql> grant all on nova.* to 'nova'@'%';
(If anyone can explain why nova@localhost is required even though the anonymous accounts have been deleted, I'd be very grateful :-)
Then configure nova to use the DB and install the schema:
$> sudo openstack-config-set /etc/nova/nova.conf DEFAULT sql_connection mysql://nova:nova@localhost/nova $> sudo nova-manage db sync
As a final sanity check:
$> mysql -u nova -p nova Enter password: mysql> select * from migrate_version;
Miscellaneous
Smoke Tests
Nova comes with a selection of fairly basic smoke tests which you can run against your installation. It can be useful to use these to sanity check your configuration.
First off, you need the nova-adminclient python library which isn't yet packaged:
$> sudo yum install python-pip $> sudo pip-python install nova-adminclient
Then you need a user and project both named admin:
$> sudo nova-manage user admin admin $> sudo nova-manage project create admin admin $> sudo nova-manage project zipfile admin admin $> unzip nova.zip $> . ./novarc
Make sure you have the tty images imported as described above. You also need a block of floating IPs created, also as described above.
Then, run the tests from a fedpkg checkout:
$> fedpkg clone openstack-nova $> cd openstack-nova $> fedpkg switch-branch f16 $> fedpkg prep $> cd nova-2011.3/smoketests $> python ./run_tests.py
All the tests should pass.
If you run into import errors such as:
ImportError: No module named nose
or:
ImportError (No module named paramiko)
simply install the missing dependency as follows:
$> sudo yum install -y python-nose.noarch $> sudo yum install -y python-paramiko.noarch
Cleanup
While testing OpenStack, you might want to delete everything related to OpenStack and start testing with a clean slate again.
Here's how. First, make sure to terminate all running instances:
$> euca-terminate-instances ...
Double check that you have no lingering VMs, perhaps saved to disk:
$> virsh list --all && virsh undefine $> rm -f /var/lib/libvirt/qemu/save/instance-00000*
Then stop all the services:
$> for iii in /usr/lib/systemd/system/openstack-*.service; do sudo systemctl stop $(basename $iii); done
Delete all the packages:
$> sudo yum erase python-glance python-nova* python-keystone* openstack-swift* memcached
Delete the nova and keystone tables from the MySQL DB:
$> mysql -u root -p -e 'drop database nova;' $> mysql -u root -p -e 'drop database keystone;'
Delete the nova-volumes VG:
$> sudo vgchange -an nova-volumes $> sudo losetup -d /dev/loop0 $> sudo rm -f /var/lib/nova/nova-volumes.img
Take down the bridge and kill dnsmasq:
$> sudo ip link set demonetbr0 down $> sudo brctl delbr demonetbr0 $> sudo kill -9 $(cat /var/lib/nova/networks/nova-demonetbr0.pid)
Remove all directories left behind from the packages:
$> sudo rm -rf /etc/{glance,nova,swift,keystone,openstack-dashboard} /var/lib/{glance,nova,swift,keystone} /var/log/{glance,nova,swift,keystone} /var/run/{glance,nova,swift,keystone}
Remove swift storage device (if we don't want the data)
$> sudo umount /srv/node/partitions $> sudo losetup -d $DEVICE $> rm /tmp/swiftstorage
Finally, restart iptables to clear out all rules added by Nova. You also need to reload libvirt's iptables rules:
$> sudo service iptables restart $> sudo service libvirtd restart