Re: [Rdo-list] Deploying an HA dev environment with tripleo-quickstart

On 02/19/2016 09:57 AM, Lars Kellogg-Stedman wrote: > I'm working on a slightly more in-depth article on this topic, but in > order for some people (pmyers I'm looking at you) to get started I /me looks around sheepishly... :) > wanted to write up some very quick instructions. Forgive me any typos > in this email because I'd like to send it out before actually running > through everything locally: while the process is automated, an HA > deploy can still take quite a while to complete. Also, the fetch of > the undercloud image *also* takes a chunk of time; there are > instructions in the tripleo-quickstart README for caching a local copy > of the image to speed up subsequent installs. https://github.com/redhat-openstack/tripleo-quickstart/blob/master/README... Definitely do this. > You will need a physical host with at least 32GB of RAM. More is > better, less *may* be possible but you will probably regret it. Wheee.... MemTotal: 65764284 kB MemFree: 58529856 kB > You will also need Ansible 2.0.x, which is what you will get if you > 'pip install ansible', or install Ansible from updates-testing > (Fedora) or epel-testing (RHEL/CentOS/...). Ok, since I'm starting with a pretty vanilla CentOS7 Server with libvirt, qemu, etc installed... # yum install ansible --enablerepo epel-testing ansible.noarch 0:2.0.0.2-1.el7 > Do *not* run Ansible HEAD from the git repository! This will lead to > sadness and disappointment. > > 1. Prepare your target host. > > You need a user on your target host to which you can (a) log in via > ssh without a password and then (b) sudo to root without a password. > We'll refer to this user as "targetuser", below. That is, the > following should work: > > ssh -tt targetuser@targethost sudo uptime using 'admin' account, verified passwordless ssh as admin works and passwordless sudo from admin to root works > 2. Clone the tripleo-quickstart repository: > > git clone https://github.com/redhat-openstack/tripleo-quickstart > cd tripleo-quickstart > > (Everything below is run from inside the tripleo-quickstart > directory) > > 2. Create an ansible inventory file. > > Create an inventory file that lists your target host in the 'virthost' > and that provides ansible with the necessary connection information: > > cat > inventory <<EOF > [virthost] > my.target.host ansible_user=targetuser duh, for those of us that are noobs... replace my.target.host with localhost So for me its: cat > inventory <<EOF [virthost] localhost ansible_user=admin > 3. Create an ansible playbook. > > cat > playbooks/ha.yml <<EOF > - hosts: virthost > roles: > - role: libvirt/teardown > - role: libvirt/setup > > - hosts: localhost > roles: > - rebuild-inventory > > - hosts: undercloud > roles: > - overcloud > EOF > > 4. Create a variables file that describes your architecture: > > cat > nodes.yml <<EOF > extra_args: >- > --control-scale 3 > -e /usr/share/openstack-tripleo-heat-templates/environments/puppet-pacemaker.yaml > --ntp-server pool.ntp.org > baremetal_vm_xml: | > <cpu mode='host-passthrough'/> > libvirt_args: --libvirt-type kvm > > # Set these to values appropriate for your target system. You > # by default get three controllers, one compute node, and > # one ceph node (so this example has a memory footprint of > # 32GB, which is probably too much for a host with only > # 32GB of RAM). > control_memory: 8192 > compute_memory: 4096 > ceph_memory: 4096 url: file:///usr/share/quickstart_images/mitaka/undercloud.qcow2 (if you pre-downloaded the undercloud image as per the instructions in the rst file link above) > EOF > > The above configuration will enabled nested KVM on the target host. > > It is possible to change the number of nodes of each type that are > created, but that's for another time. > > 5. Run it! > > ansible-playbook playbooks/ha.yml -i inventory -e @nodes.yml Ran into a small issue... since I was running as admin and needing sudo for root escalation, this patch from larsks was necessary to apply: http://chunk.io/f/574614d4738c460db656714931591694 Kicked off at 11am EST sharp. Currently running overcloud deploy. Lars, let us know when this is merged :) > This will: > > - First attempt to clean up the virtual environment from any > previous run of tripleo-quickstart > - Deploy a new virtual undercloud and virtual overcloud > - Install the undercloud > - Deploy the overcloud > - Validate the overcloud > > If you don't trust your copying-and-pasting, the example files > referenced in this email are also available from: > > https://gist.github.com/larsks/e02ca28982d1daacfa5d > > E.g.: > > git clone https://gist.github.com/e02ca28982d1daacfa5d.git To login to undercloud ssh -F ~/.quickstart/ssh.config.ansible undercloud hewbrocca also mentions that in the future, heat.conf in the undercloud will set by default max_resources_per_stack to -1 which should make things go much faster trown notes that undercloud node will be slow to deploy with only 1 vCPU. Since my box has 16 real (32 with HT), this seems like a waste of computing power :) Adding: undercloud_vcpu: 4 control_vcpu: 2 to nodes.yml may make sense At least on my machine given that I have 64gb/32 cores Validate step finished around 70 minutes after initial run of the ansible playbook, but it failed. larsks suggested that it might on a HA deployment. This will need looking into. I was able to ssh into the undercloud and from there ssh into overcloud nodes after getting their control plane IP addresses via undercloud nova list I was also able to source overcloudrc and nova list, and that worked. Probably I'll want to set up some ssh tunnels so that I can access the overcloud horizon (just for the heck of it) without needing to be ON the undercloud node And for the uninitiated... there is no Undercloud Horizon. Only CLI So far, I think this is all fairly accessible. It's really not that much more time consuming than Packstack and I think some optimizations can be put into place (max_resources_per_stack and undercloud vcpus) to make things speedier. That being said... I am running on a machine which most developers wouldn't have, so the next steps will be to make it reasonable on a 32gb machine and to provide a non-HA setup for those with 16gb machines. Perry