openstack-manuals/doc/image-guide/source/openstack-images.rst
Akihiro Motoki afa18f4f37 Clean up remaining references to underscore files of user-guide
Change-Id: I0e0699d4aa479c181a79ad71472be87b4c1136e7
Implements: blueprint consistency-file-rename
2016-08-12 13:48:38 +09:00

17 KiB

Image requirements

Linux

For a Linux-based image to have full functionality in an OpenStack Compute cloud, there are a few requirements. For some of these, you can fulfill the requirements by installing the cloud-init package. Read this section before you create your own image to be sure that the image supports the OpenStack features that you plan to use.

  • Disk partitions and resize root partition on boot (cloud-init)
  • No hard-coded MAC address information
  • SSH server running
  • Disable firewall
  • Access instance using ssh public key (cloud-init)
  • Process user data and other metadata (cloud-init)
  • Paravirtualized Xen support in Linux kernel (Xen hypervisor only with Linux kernel version < 3.0)

Disk partitions and resize root partition on boot (cloud-init)

When you create a Linux image, you must decide how to partition the disks. The choice of partition method can affect the resizing functionality, as described in the following sections.

The size of the disk in a virtual machine image is determined when you initially create the image. However, OpenStack lets you launch instances with different size drives by specifying different flavors. For example, if your image was created with a 5 GB disk, and you launch an instance with a flavor of m1.small. The resulting virtual machine instance has, by default, a primary disk size of 10 GB. When the disk for an instance is resized up, zeros are just added to the end.

Your image must be able to resize its partitions on boot to match the size requested by the user. Otherwise, after the instance boots, you must manually resize the partitions to access the additional storage to which you have access when the disk size associated with the flavor exceeds the disk size with which your image was created.

Xen: one ext3/ext4 partition (no LVM)

If you use the OpenStack XenAPI driver, the Compute service automatically adjusts the partition and file system for your instance on boot. Automatic resize occurs if the following conditions are all true:

  • auto_disk_config=True is set as a property on the image in the image registry.
  • The disk on the image has only one partition.
  • The file system on the one partition is ext3 or ext4.

Therefore, if you use Xen, we recommend that when you create your images, you create a single ext3 or ext4 partition (not managed by LVM). Otherwise, read on.

Non-Xen with cloud-init/cloud-tools: one ext3/ext4 partition (no LVM)

You must configure these items for your image:

  • The partition table for the image describes the original size of the image.
  • The file system for the image fills the original size of the image.

Then, during the boot process, you must:

  • Modify the partition table to make it aware of the additional space:
    • If you do not use LVM, you must modify the table to extend the existing root partition to encompass this additional space.
    • If you use LVM, you can add a new LVM entry to the partition table, create a new LVM physical volume, add it to the volume group, and extend the logical partition with the root volume.
  • Resize the root volume file system.

The simplest way to support this is to install in your image the:

  • cloud-utils package, which contains the growpart tool for extending partitions.
  • cloud-initramfs-growroot package for Ubuntu, Debian and Fedora, which supports resizing root partition on the first boot.
  • cloud-initramfs-growroot package for CentOS and RHEL.
  • cloud-init package.

With these packages installed, the image performs the root partition resize on boot. For example, in the /etc/rc.local file. These packages are in the Ubuntu and Debian package repository, as well as the EPEL repository (for Fedora/RHEL/CentOS/Scientific Linux guests).

If you cannot install cloud-initramfs-tools, Robert Plestenjak has a GitHub project called linux-rootfs-resize that contains scripts that update a ramdisk by using growpart so that the image resizes properly on boot.

If you can install the cloud-utils and cloud-init packages, we recommend that when you create your images, you create a single ext3 or ext4 partition (not managed by LVM).

Non-Xen without cloud-init/cloud-tools: LVM

If you cannot install cloud-init and cloud-tools inside of your guest, and you want to support resize, you must write a script that your image runs on boot to modify the partition table. In this case, we recommend using LVM to manage your partitions. Due to a limitation in the Linux kernel (as of this writing), you cannot modify a partition table of a raw disk that has partitions currently mounted, but you can do this for LVM.

Your script must do something like the following:

  1. Detect if any additional space is available on the disk. For example, parse the output of parted /dev/sda --script "print free".
  2. Create a new LVM partition with the additional space. For example, parted /dev/sda --script "mkpart lvm ...".
  3. Create a new physical volume. For example, pvcreate /dev/sda6.
  4. Extend the volume group with this physical partition. For example, vgextend vg00 /dev/sda6.
  5. Extend the logical volume contained the root partition by the amount of space. For example, lvextend /dev/mapper/node-root /dev/sda6.
  6. Resize the root file system. For example, resize2fs /dev/mapper/node-root.

You do not need a /boot partition unless your image is an older Linux distribution that requires that /boot is not managed by LVM.

No hard-coded MAC address information

You must remove the network persistence rules in the image because they cause the network interface in the instance to come up as an interface other than eth0. This is because your image has a record of the MAC address of the network interface card when it was first installed, and this MAC address is different each time the instance boots. You should alter the following files:

  • Replace /etc/udev/rules.d/70-persistent-net.rules with an empty file (contains network persistence rules, including MAC address).
  • Replace /lib/udev/rules.d/75-persistent-net-generator.rules with an empty file (this generates the file above).
  • Remove the HWADDR line from /etc/sysconfig/network-scripts/ifcfg-eth0 on Fedora-based images.

Note

If you delete the network persistent rules files, you may get a udev kernel warning at boot time, which is why we recommend replacing them with empty files instead.

Ensure ssh server runs

You must install an ssh server into the image and ensure that it starts up on boot, or you cannot connect to your instance by using ssh when it boots inside of OpenStack. This package is typically called openssh-server.

Disable firewall

In general, we recommend that you disable any firewalls inside of your image and use OpenStack security groups to restrict access to instances. The reason is that having a firewall installed on your instance can make it more difficult to troubleshoot networking issues if you cannot connect to your instance.

Access instance by using ssh public key (cloud-init)

The typical way that users access virtual machines running on OpenStack is to ssh using public key authentication. For this to work, your virtual machine image must be configured to download the ssh public key from the OpenStack metadata service or config drive, at boot time.

If both the XenAPI agent and cloud-init are present in an image, cloud-init handles ssh-key injection. The system assumes cloud-init is present when the image has the cloud_init_installed property.

Use cloud-init to fetch the public key

The cloud-init package automatically fetches the public key from the metadata server and places the key in an account. The account varies by distribution. On Ubuntu-based virtual machines, the account is called ubuntu. On Fedora-based virtual machines, the account is called ec2-user.

You can change the name of the account used by cloud-init by editing the /etc/cloud/cloud.cfg file and adding a line with a different user. For example, to configure cloud-init to put the key in an account named admin, edit the configuration file so it has the line:

user: admin

Write a custom script to fetch the public key

If you are unable or unwilling to install cloud-init inside the guest, you can write a custom script to fetch the public key and add it to a user account.

To fetch the ssh public key and add it to the root account, edit the /etc/rc.local file and add the following lines before the line touch /var/lock/subsys/local. This code fragment is taken from the rackerjoe oz-image-build CentOS 6 template <https://github.com/ rackerjoe/oz-image-build/blob/master/templates/centos60_x86_64.tdl>.

if [ ! -d /root/.ssh ]; then
  mkdir -p /root/.ssh
  chmod 700 /root/.ssh
fi

# Fetch public key using HTTP
ATTEMPTS=30
FAILED=0
while [ ! -f /root/.ssh/authorized_keys ]; do
  curl -f http://169.254.169.254/latest/meta-data/public-keys/0/openssh-key > /tmp/metadata-key 2>/dev/null
  if [ $? -eq 0 ]; then
    cat /tmp/metadata-key >> /root/.ssh/authorized_keys
    chmod 0600 /root/.ssh/authorized_keys
    restorecon /root/.ssh/authorized_keys
    rm -f /tmp/metadata-key
    echo "Successfully retrieved public key from instance metadata"
    echo "*****************"
    echo "AUTHORIZED KEYS"
    echo "*****************"
    cat /root/.ssh/authorized_keys
    echo "*****************"
  else
    FAILED=`expr $FAILED + 1`
    if [ $FAILED -ge $ATTEMPTS ]; then
      echo "Failed to retrieve public key from instance metadata after $FAILED attempts, quitting"
      break
    fi
    echo "Could not retrieve public key from instance metadata (attempt #$FAILED/$ATTEMPTS), retrying in 5 seconds..."
    sleep 5
  fi
done

Note

Some VNC clients replace : (colon) with ; (semicolon) and _ (underscore) with - (hyphen). If editing a file over a VNC session, make sure it is http: not http; and authorized_keys not authorized-keys.

Process user data and other metadata (cloud-init)

In addition to the ssh public key, an image might need additional information from OpenStack, such as Provide user data to instances <http://docs.openstack.org/ user-guide/cli-provide-user-data-to-instances.html>, that the user submitted when requesting the image. For example, you might want to set the host name of the instance when it is booted. Or, you might wish to configure your image so that it executes user data content as a script on boot.

You can access this information through the metadata service or referring to Store metadata on the configuration drive. As the OpenStack metadata service is compatible with version 2009-04-04 of the Amazon EC2 metadata service, consult the Amazon EC2 documentation on Using Instance Metadata <http://docs.amazonwebservices.com/ AWSEC2/2009-04-04/UserGuide/AESDG-chapter-instancedata.html> for details on how to retrieve the user data.

The easiest way to support this type of functionality is to install the cloud-init package into your image, which is configured by default to treat user data as an executable script, and sets the host name.

Ensure image writes boot log to console

You must configure the image so that the kernel writes the boot log to the ttyS0 device. In particular, the console=ttyS0 argument must be passed to the kernel on boot.

If your image uses grub2 as the boot loader, there should be a line in the grub configuration file. For example, /boot/grub/grub.cfg, which looks something like this:

linux /boot/vmlinuz-3.2.0-49-virtual root=UUID=6d2231e4-0975-4f35-a94f-56738c1a8150 ro console=ttyS0

If console=ttyS0 does not appear, you must modify your grub configuration. In general, you should not update the grub.cfg directly, since it is automatically generated. Instead, you should edit the /etc/default/grub file and modify the value of the GRUB_CMDLINE_LINUX_DEFAULT variable:

GRUB_CMDLINE_LINUX_DEFAULT="console=ttyS0"

Next, update the grub configuration. On Debian-based operating systems such as Ubuntu, run this command:

# update-grub

On Fedora-based systems, such as RHEL and CentOS, and on openSUSE, run this command:

# grub2-mkconfig -o /boot/grub2/grub.cfg

Paravirtualized Xen support in the kernel (Xen hypervisor only)

Prior to Linux kernel version 3.0, the mainline branch of the Linux kernel did not have support for paravirtualized Xen virtual machine instances (what Xen calls DomU guests). If you are running the Xen hypervisor with paravirtualization, and you want to create an image for an older Linux distribution that has a pre 3.0 kernel, you must ensure that the image boots a kernel that has been compiled with Xen support.

Manage the image cache

Use options in the nova.conf file to control whether, and for how long, unused base images are stored in the /var/lib/nova/instances/_base/. If you have configured live migration of instances, all your compute nodes share one common /var/lib/nova/instances/ directory.

For information about the libvirt images in OpenStack, see The life of an OpenStack libvirt image from Pádraig Brady.

Image cache management configuration options
Configuration option=Default value (Type) Description
preallocate_images=none (StrOpt) VM image preallocation mode:
none

No storage provisioning occurs up front.

space

Storage is fully allocated at instance start. The $instance_dir/ images are fallocated to immediately determine if enough space is available, and to possibly improve VM I/O performance due to ongoing allocation avoidance, and better locality of block allocations.

remove_unused_base_images=True (BoolOpt) Should unused base images be removed? When set to True, the interval at which base images are removed are set with the following two settings. If set to False base images are never removed by Compute.
remove_unused_original_minimum_age_seconds=86400 (IntOpt) Unused unresized base images younger than this are not removed. Default is 86400 seconds, or 24 hours.
remove_unused_resized_minimum_age_seconds=3600 (IntOpt) Unused resized base images younger than this are not removed. Default is 3600 seconds, or one hour.

To see how the settings affect the deletion of a running instance, check the directory where the images are stored:

# ls -lash /var/lib/nova/instances/_base/

In the /var/log/compute/compute.log file, look for the identifier:

2012-02-18 04:24:17 41389 WARNING nova.virt.libvirt.imagecache [-] Unknown base file: /var/lib/nova/instances/_base/06a057b9c7b0b27e3b496f53d1e88810a0d1d5d3_20
2012-02-18 04:24:17 41389 INFO nova.virt.libvirt.imagecache [-] Removable base files: /var/lib/nova/instances/_base/06a057b9c7b0b27e3b496f53d1e88810a0d1d5d3 /var/lib/nova/instances/_base/06a057b9c7b0b27e3b496f53d1e88810a0d1d5d3_20
2012-02-18 04:24:17 41389 INFO nova.virt.libvirt.imagecache [-] Removing base file: /var/lib/nova/instances/_base/06a057b9c7b0b27e3b496f53d1e88810a0d1d5d3

Because 86400 seconds (24 hours) is the default time for remove_unused_original_minimum_age_seconds, you can either wait for that time interval to see the base image removed, or set the value to a shorter time period in the nova.conf file. Restart all nova services after changing a setting in the nova.conf file.