Before launching an instance, make sure you own a lease. About how to create a lease, please see Reservations. Once your lease is started, you are almost ready to start an instance. But first, you need to make sure that you will be able to connect to it by setting up Key Pairs.
Launching Instances with the GUI¶
Launch an Instance¶
To launch an instance with the GUI, follow the steps:
In the navigation side bar, click Project > Compute > Instances to get to the Instances page.
Click the Launch Instance button in the upper right corner. This will open the Launch Instance wizard with several configuration steps. Steps with
In the Details step, enter a name for your instance that is unique within your project and select a currently active reservation for the instance.
In the Source step, select an image for your instance and click the “up” arrow. The image should move to the Allocated list, and can be removed by clicking the “Down” arrow if you wish to select a different image.
In the Flavor step, select the baremetal flavor by clicking the “up” arrow next to it. This is the only flavor available.
If you are familiar with Openstack, other implementations allow for the selection of flavors based on machine disk size and RAM. On Chameleon, the only flavor available is “baremetal” because hardware selection is performed in reservations.
In the Networks step, select a network by clicking the “up” arrow next to it. About Chameleon default network and how to create your own network, please see Networking.
In the Key Pair step, select one of your SSH key pairs. If you only have one key pair associated with your account, then it is selected by default.
It is a good practice to make sure that the instance is launching with the key pair of your choice, or you will not be able to access your instance.
You may import or create key pairs directly through this step.
Optionally, you may configure Scheduler Hints. This is useful if you would like to launch an instance on a specific node in your multi-node reservation by UUID.
- In the Custom text box, type
queryand click the + button. This will add a query hint to the list on the right.
- In the query hint, enter your scheduler hint. For example, if you require a specific node, type
<node_uuid>is the node you are requesting.
- In the Custom text box, type
If you want to customize your instance after it has launched, you can add a customization script in the Configuration step.
- You can type in the script in Customization Script.
- Or you can upload your script via Load script from a file.
Finish configuring and start launching the instance by clicking on the Launch Instance button. The instance will show up in the instance list, at first in Build status. It takes a few minutes to deploy the instance on bare metal hardware and reboot the machine.
After a few minutes, the instance should become Active. The power state will show as Running. You can now Associate a Floating IP.
To view instance details, click on the instance.
Associate a Floating IP¶
To make your instance publicly accessible over the Internet, you must associate a Floating IP Address to it.
On the Instances page, click the Associate Floating IP button next to your bare metal instance. The following Manage Floating IP Assocations dialog will show.
If you have previously allocated an IP address to your project that is not currently used, you may select it in the IP Address dropdown. Otherwise, click the + button to open the Allocate Floating IP dialog. In this dialog, you may simply click the Allocate IP button.
The correct value for “Port to be associated” should already be selected. You may click the Associate button.
You will be sent back to the instance list, where you can see the floating IP attached to the instance (you may need to refresh your browser to see the floating IP).
Launching Instances with the CLI¶
Reading The Command Line Interface is highly recommanded before continuing on the following sections.
Creating an Instance with the Nova Client¶
To launch an instance inside a reservation, run:
openstack server create \ --image CC-CentOS7 \ --flavor baremetal \ --key-name <key_name> \ --nic net-id=<sharednet1_id> \ --hint reservation=<reservation_id> \ my-instance
The ID of the
sharednet1 network can be obtained using the command:
openstack network list
Alternatively, you may look it up in the GUI in the Network > Networks page. You can obtain your reservation ID via the web interface or by running:
blazar lease-show <lease_name>
The reservation ID and the lease ID are different
Running a Shell Script on Boot¶
You might want to automatically execute some code after launching an instance, whether it is installing packages, changing configuration files, or running an application. OpenStack provides a mechanism called User Data to pass information to instances. This information can be any data in any format, but if it is a shell script it will be automatically executed after boot by cloudinit. You can provide this shell script either via the web interface in the Configuration tab when launching an instance, or by providing a file to the nova command line using the
Chameleon supported images are configured with appliance agents, including collectd and Heat agents.
To turn off appliance agents on boot, in order to remove the potential impact on experimental measurements, pass the following script as
#!/bin/bash systemctl stop collectd.service systemctl disable collectd.service systemctl stop os-collect-config.service systemctl disable os-collect-config.service
Customizing the Kernel¶
Before the February 2016 upgrade, support for kernel customizing on bare metal was limited due to the fact that instances were always booting their kernel directly using PXE and a common kernel command line. This required uploading kernel and ramdisk files to the Glance image repository as well as updating or creating a new OS image using these artifacts.
However, it is now easy to customize the operating system kernel or modify the kernel command line. You now have the option of modifying the boot loader configuration (
/boot/grub2/grub.cfg on CentOS 7 images) to point it to a new kernel on the local disk, or specifying kernel parameters and then rebooting using this modified configuration.
To do this, you must be using a whole disk image rather than a partition image. Whole disk images contain their own kernel and ramdisk files and do not have
ramdisk_id properties in the image repository, unlike partition images.
Running Virtual Machines on Bare Metal Hardware¶
For cloud computing and virtualization experiments, you might want to run virtual machines on bare hardware that you fully control rather than use the shared OpenStack KVM cloud. There are many different ways to configure networking for virtual machines. The configuration described below will enable you to connect your virtual machines to the Internet using a KVM Public Bridge which you must first configure manually on your host on the default network interface.
pip install python-neutronclient
Then, for each virtual machine you want to run, request a Neutron port with:
openstack port-create sharednet1
This should display, among other information:
- A fixed IP in the same private network as the physical nodes
- A MAC address
Finally, start your virtual machine while assigning it the MAC address provided by OpenStack. If your image is configured to use DHCP, the virtual machine should receive the allocated IP.
Neutron ports allocated this way are not automatically deleted, so please delete them after your experiment is over using:
openstack port delete <id>
You may find the ID of your ports using:
openstack port list
Launching Instances on Specific Nodes¶
If you have a reservation for multiple physical nodes, explicitly identified with their UUIDs, you might want to force an instance to be launched on a specific node rather than letting the scheduler select one. This can be done with the CLI using a scheduler hint:
openstack server create \ --image CC-CentOS7 \ --flavor baremetal \ --key-name <key_name> \ --nic net-id=<sharednet1_id> \ --hint reservation=<reservation_id> \ --hint query='["=","$hypervisor_hostname","<node_uuid>"]' \ <instance_name>
From within an instance you have already launched, you can discover which node it is running on by executing
This will return a JSON dictionary describing site, cluster, and node.
In its default configuration, the bare metal deployment system used by Chameleon (OpenStack Ironic) is restricted to using a single shared network per site. The network configuration features available in the dashboard are not supported (Networks and Routers). On CHI@UC, network layer 2 isolation is optionally available for compute nodes. You may find more details on the documentation for Networking.
Interacting with Instances¶
Once your bare metal instance has launched, you may interact with it by using SSH if you have associated a Floating IP to it or by using the Serial Console from the GUI.
Connecting via SSH¶
If you have associated a Floating IP with the instance and you have the private key in place, you should be able to connect to the instance via SSH using the
To access the instance using SSH, type the command in your terminal:
If you get errors:
@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ @ WARNING: REMOTE HOST IDENTIFICATION HAS CHANGED! @ @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ IT IS POSSIBLE THAT SOMEONE IS DOING SOMETHING NASTY! ...
It is likely that you have saved a previous entry for the instance’s Floating IP in your
~/.ssh/known_hosts file on your computer. Simply removing the entry from the file should solve the issue.
You can remove the entry from the
~/.ssh/known_hosts file by using the command:
ssh-keygen -R <floating_ip>
You may receive the response below. Type
yes and hit enter:
The authenticity of host '188.8.131.52 (184.108.40.206)' can't be established. RSA key fingerprint is 5b:ca:f0:63:6f:22:c6:96:9f:c0:4a:d8:5e:dd:fd:eb. Are you sure you want to continue connecting (yes/no)?
When logged in, your prompt may appear like this:
If you notice SSH errors such as connection refused, password requests, or failures to accept your key, it is likely that the physical node is still going through the boot process. In that case, please wait before retrying. Also make sure that you use the
cc account. If after 10 minutes you still cannot connect to the machine, please open a ticket with our help desk.
You can now check whether the resource matches its known description in the resource registry. For this, simply run:
sudo cc-checks -v
cc-checks program prints the result of each check in green if it is successful and red if it failed. You can now run your experiment directly on the machine via SSH. You can run commands with root privileges by prefixing them with
sudo. To completely switch user and become root, use the
sudo su - root command.
Connecting via the Serial Console¶
Chameleon now allows you to connect to the serial console of your bare metal nodes via the GUI. Once your instance is deployed, click on the Console button in the instance contextual menu.
This should open a screen showing an interactive serial console (it could take some time to show up, give it 30 seconds or so).
Our latest images are configured to auto-login into the
cc account. Other images may show you a login prompt. You can set a password on the
cc account by accessing it via SSH, using the command
sudo passwd cc, and then using this password to connect to the console.