Before you attempt to install Eucalyptus Cloud, make sure of:
- all machines part of the deployment have been cleanly installed (currently with CentOS or RHEL 7) with as default options as possible (firewalld can be left running, SELinux in permissive mode);
- select the machines roles (CLC, CC, NC, ceph) in case you have more than one machine in the deployment. You should be able to ssh without any password from the machine designated as CLC all other machines;
- make sure the network is setup accordingly to your planning. We suggest at least 2 networks the public facing one and the private facing one: the public will be the one used from clients to connects to the cloud, and the private will be the one mostly used internally for Eucalyptus and Ceph communication;
- all the machine should be able to reach access to Internet for packages, ntp etc... The CLC should have a public IP while all the others can have a masquerade access (or public IP);
- a set of public IPs should reserved for the Eucalyptus Cloud to use with the instances, elastic load balancers, etc ...
- a DNS (sub)domain should be reserved to the cloud, and it should be delegated to the CLC as the authoritative nameserver for it;
- if you use this ansible playbook for ceph deployment, we suggest to have full disks available for OSD: the disks should be empty, and one OSD will be started per disk.
Create an inventory for your environment:
cp inventory_example.yml inventory.yml
vi inventoy.yml
and modify accordingly. If you want a single node deployment (Cloud In A Box option) use instead:
cp inventory_example_local.yml inventory.yml
vi inventory.yml
To operate the cloud will need a subdomain, and a set of public IPs to be used for the instances and services. The options to set them up are:
cloud_system_dns_dnsdomain: <DNS subdomain this deployment is responsible for>
the DNS subdomain needs to be delegated to the cloud (CLC) machine. For example to delegate ats.mydomain.foo with dnsmasq add to dnsmasq configuration (for bind or other DNS server, follow their respective documentation):
server=/ats.mydomain.foo/<CLC IP address>
to enable the delegation. For the public IPs, you will need to define the network they are in and list them:
vpcmido_public_ip_cidr: <public IPs for example 10.1.1.0/24>
vpcmido_public_ip_range: <IPs list for example 10.1.1.5-10.1.1.50>
There are few variables to use to configure the cloud:
cloud_firewalld_configure: yes
default is no, but without this one you either disable firewalld or you configure it by hand;
cloud_use_dnsmasq: no
default is yes and it is recommended to keep it. Without a local dnsmasq running on the CLC configured to delegate the subdomain to the CLC, your deployment will depends on external DNS server to operate some functionalities (for example ELBs and more);
ceph_osd_data_path:
- /dev/sdg
- /dev/sdh
if you choose to let this script setup ceph for you, this is the list (1 or more) of hard disks to delegate to each ceph machine. The disks needs to be empty, without tags;
cloud_region_name: my_lovely_cloud
you can customize here your deployment region's name (from the unimaginative default of cloud-1);
cloud_zone_1_name: developers
cloud_zone_2_name: sales
cloud_zone_3_name: production
you can customize here the name if each AZ you have configured. Defaults is to add a, b etc... to the deployment name;
eucalyptus_console_cname: ec2.clouddns_subdomain
this is the CNAME to be used with your deployment Route53 to set an entry for the console. The default assumes you installed the eucaconsole on the CLC machine, and thus we add an entry of ec2.your_dns_cloud_subdomain. If you prefer to set a full A record instead add also eucalyptus_console_ipv4: IP_address;
enable_M1_instances: False
enable_M2_instances: False
enable_M3_instances: False
enable_M4_instances: False
enable_M5D_instances: False
enable_T1_instances: False
enable_T3_instances: False
turns each one of the above to True to have more instance types available in your deployment.
To install with VPCMIDO network mode (recommended):
ansible-playbook -i inventory.yml playbook_vpcmido.yml
this is the recommended network mode to have access to all VPC options.
To install with EDGE network mode:
ansible-playbook -i inventory.yml playbook[_edge].yml
To remove a eucalyptus installation and main dependencies:
ansible-playbook -i inventory.yml playbook_clean.yml
NOTE: if ceph was installed with this ansible playbook, it will not be touched, thus you will have to clear all the disks yourself if you desire to do a full clean install. For example if you have /dev/sdb and /dev/sdc dedicated to ceph and you want to clear them on each ceph node do:
for x in `lvdisplay |awk '/\/dev\/ceph/ { print $3 }'`; do lvremove -f $x; done
for x in `pvdisplay |awk '/ceph/ {print $3}'`; do vgremove $x; done
pvremove /dev/sdb /dev/sdc
Tags can be used to control which aspects of the playbook are used:
image:packagesand generic configurationpackages: installs yum repositories and rpmsmidonet-packages: downloads only Midonet packages locally
Example tag use:
ansible-playbook --tags image -i inventory.yml playbook.yml
ansible-playbook --skip-tags image -i inventory.yml playbook.yml
ansible-playbook --skip-tags selinux,firewalld -i inventory.yml playbook.yml
which would run the playbook in two parts, the first installing packages and non-deployment specific configuration, and the second completing the deployment.
The ceph cluster can be configured with its own cluster and public network which can and should be separate from the concept of public or cluster network for Eucalyptus. The main purpose here is to setup a separate storage network to be used by the deploymnet (the ceph_public_network), and/or a private ceph network to be used by OSD for synchronization (ceph_private_network).
Ceph hosts can be explicitely told which IP is where with ceph_public_ipv4 and ceph_private_ipv4.
Without any specification the Eucalyputs public and cluster networks will be used.