Walk through of running OpenStack on Fedora 17 using DevStack
When first getting involved in the OpenStack project as a developer, most people will probably recommend use of DevStack. When I first started hacking, I skipped this because it wasn’t reliable on Fedora at that time, but these days it works just fine and there are even basic instructions for DevStack on Fedora. Last week I decided to finally give DevStack a go, since my hand-crafted dev environment was getting kind of nasty. The front page on the DevStack website says it is only supported on Fedora 16, but don’t let that put you off; aside from one bug which does not appear distro specific, it all seemed to work correctly. What follows is an overview of what I did / learnt
Setting up the virtual machine
I don’t like really like letting scripts like DevStack mess around with my primary development environment, particularly when there is little-to-no-documentation about what changes they will be making and they ask for unrestricted sudo (sigh) privileges ! Thus running DevStack inside a virtual machine was the obvious way to go. Yes, this means actual VMs run by Nova will be forced to use plain QEMU emulation (or nested KVM if you are brave), but for dev purposes this is fine, since the VMs don’t need todo anything except boot. My host is Fedora 17, and for simplicity I decided that my guest dev environment will also be Fedora 17. With that decided installing the guest was a simple matter of running virt-install on the host as root
# virt-install --name f17x86_64 --ram 2000 --file /var/lib/libvirt/images/f17x86_64.img --file-size 20 --accelerate --location http://mirror2.hs-esslingen.de/fedora/linux//releases/17/Fedora/x86_64/os/ --os-variant fedora17
I picked the defaults for all installer options, except for reducing the swap file size down to a more sensible 500 MB (rather than the 4 G it suggested). NB if copying this, you probably want to change the URL used to point to your own best mirror location.
Once installation completed, run through the firstboot wizard, creating yourself an unprivileged user account, then login as root. First add the user to the wheel group, to enable it to run sudo commands:
# gpasswd -a YOURUSERNAME wheel
The last step before getting onto DevStack is to install GIT
# yum -y install git
Setting up DevStack
The recommended way to use DevStack, is to simply check it out of GIT and run the latest code available. I like to keep all my source code checkouts in one place, so I’m using $HOME/src/openstack for this project
$ mkdir -p $HOME/src/openstack $ cd $HOME/src/openstack $ git clone git://github.com/openstack-dev/devstack.git
Arguably you can now just kick off the stack.sh script at this point, but there are some modifications that are a good idea to do. This involves creating a “localrc” file in the top level directory of the DevStack checkout
$ cd devstack $ cat > localrc <<EOF # Stop DevStack polluting /opt/stack DESTDIR=$HOME/src/openstack # Switch to use QPid instead of RabbitMQ disable_service rabbit enable_service qpid # Replace with your primary interface name HOST_IP_IFACE=eth0 PUBLIC_INTERFACE=eth0 VLAN_INTERFACE=eth0 FLAT_INTERFACE=eth0 # Replace with whatever password you wish to use MYSQL_PASSWORD=badpassword SERVICE_TOKEN=badpassword SERVICE_PASSWORD=badpassword ADMIN_PASSWORD=badpassword # Pre-populate glance with a minimal image and a Fedora 17 image IMAGE_URLS="http://launchpad.net/cirros/trunk/0.3.0/+download/cirros-0.3.0-x86_64-uec.tar.gz,http://berrange.fedorapeople.org/images/2012-11-15/f17-x86_64-openstack-sda.qcow2" EOF
With the localrc created, now just kick off the stack.sh script
$ ./stack.sh
At time of writing there is a bug in DevStack which will cause it to fail to complete correctly – it is checking for existence paths before it has created them. Fortunately, just running it for a second time is a simple workaround
$ ./unstack.sh $ ./stack.sh
From a completely fresh Fedora 17 desktop install, stack.sh will take a while to complete, as it installs a large number of pre-requisite RPMs and downloads the appliance images. Once it has finished it should tell you what URL the Horizon web interface is running on. Point your browser to it and login as “admin” with the password provided in your localrc file earlier.
Because we told DevStack to use $HOME/src/openstack as the base directory, a small permissions tweak is needed to allow QEMU to access disk images that will be created during testing.
$ chmod o+rx $HOME
Note, that SELinux can be left ENFORCING, as it will just “do the right thing” with the VM disk image labelling.
UPDATE: if you want to use the Horizon web interface, then you do in fact need to set SELinux to permissive mode, since Apache won’t be allowed to access your GIT checkout where the Horizon files live.
$ sudo su - # setenforce 0 # vi /etc/sysconfig/selinux ...change to permissive...
UPDATE:If you want to use Horizon, you must also manually install Node.js from a 3rd party repositoryh, because it is not yet included in Fedora package repositories:
# yum localinstall --nogpgcheck http://nodejs.tchol.org/repocfg/fedora/nodejs-stable-release.noarch.rpm # yum -y install nodejs nodejs-compat-symlinks # systemctl restart httpd.service
Testing DevStack
Before going any further, it is a good idea to make sure that things are operating somewhat normally. DevStack has created an file containing the environment variables required to communicate with OpenStack, so load that first
$ . openrc
Now check what images are available in glance. If you used the IMAGE_URLS example above, glance will have been pre-populated
$ glance image-list +--------------------------------------+---------------------------------+-------------+------------------+-----------+--------+ | ID | Name | Disk Format | Container Format | Size | Status | +--------------------------------------+---------------------------------+-------------+------------------+-----------+--------+ | 32b06aae-2dc7-40e9-b42b-551f08e0b3f9 | cirros-0.3.0-x86_64-uec-kernel | aki | aki | 4731440 | active | | 61942b99-f31c-4155-bd6c-d51971d141d3 | f17-x86_64-openstack-sda | qcow2 | bare | 251985920 | active | | 9fea8b4c-164b-4f54-8e74-b53966e858a6 | cirros-0.3.0-x86_64-uec-ramdisk | ari | ari | 2254249 | active | | ec3e9b72-0970-44f2-b442-58d0042448f7 | cirros-0.3.0-x86_64-uec | ami | ami | 25165824 | active | +--------------------------------------+---------------------------------+-------------+------------------+-----------+--------+
Incidentally the glance sort ordering is less than helpful here – it appears to be sorting based on the UUID strings rather than the image names :-(
Before booting a instance, Nova likes to be given an SSH public key, which it will inject into the guest filesystem to allow admin login
$ nova keypair-add --pub-key $HOME/.ssh/id_rsa.pub mykey
Finally an image can be booted
$ nova boot --key-name mykey --image f17-x86_64-openstack-sda --flavor m1.tiny f17demo1
+------------------------+--------------------------------------+
| Property | Value |
+------------------------+--------------------------------------+
| OS-DCF:diskConfig | MANUAL |
| OS-EXT-STS:power_state | 0 |
| OS-EXT-STS:task_state | scheduling |
| OS-EXT-STS:vm_state | building |
| accessIPv4 | |
| accessIPv6 | |
| adminPass | NsddfbJtR6yy |
| config_drive | |
| created | 2012-11-19T15:00:51Z |
| flavor | m1.tiny |
| hostId | |
| id | 6ee509f9-b612-492b-b55b-a36146e6833e |
| image | f17-x86_64-openstack-sda |
| key_name | mykey |
| metadata | {} |
| name | f17demo1 |
| progress | 0 |
| security_groups | [{u'name': u'default'}] |
| status | BUILD |
| tenant_id | dd3d27564c6043ef87a31404aeb01ac5 |
| updated | 2012-11-19T15:00:55Z |
| user_id | 72ae640f50434d07abe7bb6a8e3aba4e |
+------------------------+--------------------------------------+
Since we’re running QEMU inside a KVM guest, booting the image will take a little while – several minutes or more. Just keep running the ‘nova list’ command to keep an eye on it, until it shows up as ACTIVE
$ nova list +--------------------------------------+----------+--------+------------------+ | ID | Name | Status | Networks | +--------------------------------------+----------+--------+------------------+ | 6ee509f9-b612-492b-b55b-a36146e6833e | f17demo1 | ACTIVE | private=10.0.0.2 | +--------------------------------------+----------+--------+------------------+
Just to prove that it really is working, login to the instance with SSH
$ ssh ec2-user@10.0.0.2 The authenticity of host '10.0.0.2 (10.0.0.2)' can't be established. RSA key fingerprint is 9a:73:e5:1a:39:e2:f7:a5:10:a7:dd:bc:db:6e:87:f5. Are you sure you want to continue connecting (yes/no)? yes Warning: Permanently added '10.0.0.2' (RSA) to the list of known hosts. [ec2-user@f17demo1 ~]$ sudo su - [root@f17demo1 ~]#
Working with DevStack
The DevStack setup runs all the python services under a screen session. To stop/start individual services, attach to the screen session with the ‘rejoin-stack.sh’ script. Each service is running under a separate screen “window”. Switch to the window containing the service to be restarted, and just Ctrl-C it and then use bash history to run the same command again.
$ ./rejoin-stack.sh
Sometimes the entire process set will need to be restarted. In this case, just kill the screen session entirely, which causes all the OpenStack services to go away. Then the same ‘rejoin-stack.sh’ script can be used to start them all again.
One annoyance is that unless you have the screen session open, the debug messages from Nova don’t appear to end up anywhere useful. I’ve taken to editing the file “stack-screen” to make each service log to a local file in its checkout. eg I changed
stuff "cd /home/berrange/src/openstack/nova && sg libvirtd /home/berrange/src/openstack/nova/bin/nova-compute"
to
stuff "cd /home/berrange/src/openstack/nova && sg libvirtd /home/berrange/src/openstack/nova/bin/nova-compute 2>&1 | tee nova-compute.log"