Update multi-tenancy documentation

This updates the multi-tenancy documentation. The changes include:
- use the service name (e.g. the Bare Metal service) instead of the
  project name (e.g. ironic)
- fix some grammatical issues
- refer to the ironic conductor as 'ironic-conductor' instead of 'conductor'
- remove note about setting enabled_network_interfaces config option in
  ironic-api's config file; that is no longer needed since the checks
  have been moved to the conductor side (when node creation was moved).

Change-Id: Ic4b1512672c16c1bff5b43595abd90190e23b0fe
This commit is contained in:
Ruby Loo 2016-09-27 21:43:04 -04:00
parent 2a4d3cc5e8
commit 7556dc6a76

View File

@ -1,8 +1,8 @@
.. _multitenancy:
==================================
Multitenancy in Bare Metal service
==================================
=======================================
Multi-tenancy in the Bare Metal service
=======================================
Overview
========
@ -14,8 +14,8 @@ nodes in a separate provisioning network. The result of this is that multiple
tenants can use nodes in an isolated fashion. However, this configuration does
not support trunk ports belonging to multiple networks.
Network interface is one of the driver interfaces, that manages network
switching for nodes. Currently there are 3 network interfaces available in
Network interface is one of the driver interfaces that manages network
switching for nodes. There are 3 network interfaces available in
the Bare Metal service:
- ``noop`` interface is used for standalone deployments, and does not perform
@ -25,58 +25,55 @@ the Bare Metal service:
a single layer 2 network, separated from the cleaning network;
- ``neutron`` interface provides tenant-defined networking by integrating with
neutron, while also separating tenant networks from the provisioning and
cleaning provider networks.
the Networking service, while also separating tenant networks from the
provisioning and cleaning provider networks.
Configuring Ironic
==================
Configuring the Bare Metal service
==================================
Below is an example flow of how to setup ironic so that node provisioning will
happen in a multitenant environment (which means using ``neutron`` network
interface as stated above):
Below is an example flow of how to set up the Bare Metal service so that node
provisioning will happen in a multi-tenant environment (which means using
``neutron`` network interface as stated above):
#. Network interfaces can be enabled on ironic-conductor by adding them to
``enabled_network_interfaces`` configuration option under the default
section::
#. Network interfaces can be enabled on ironic-conductor by adding them to the
``enabled_network_interfaces`` configuration option under the ``default``
section of the configuration file::
[DEFAULT]
...
enabled_network_interfaces=noop,flat,neutron
Please note that ``enabled_network_interfaces`` has to be set in the
ironic-api configuration file too, as its value is used on the API side to
check if the requested network interface is available.
Keep in mind that, ideally, all ironic-conductors should have the same list
of enabled network interfaces, but it may not be the case during
ironic-conductor upgrades. This may cause problems if one of the
ironic-conductors dies and some node that is taken over is mapped to an
ironic-conductor that does not support the node's network interface.
Any actions that involve calling the node's driver will fail until that
network interface is installed and enabled on that ironic-conductor.
Keep in mind that, ideally, all conductors should have the same list of
enabled network interfaces, but it may not be the case during conductor
upgrades. This may cause problems if one of the conductors dies and some
node that is taken over is mapped to a conductor that does not support the
node's network interface. Any actions that involve calling the node's driver
will fail until that network interface is installed and enabled on that
conductor.
#. It is recommended to set the default network interface via
``default_network_interface`` configuration option under the default
section::
#. It is recommended to set the default network interface via the
``default_network_interface`` configuration option under the ``default``
section of the configuration file::
[DEFAULT]
...
default_network_interface=neutron
It will be used for all nodes that will be created without explicitly
specifying the network interface.
This default value will be used for all nodes that don't have a network
interface explicitly specified in the creation request.
If it is not set, the default network interface is determined by looking at
the ``[dhcp]dhcp_provider`` configuration option value. If it is
``neutron`` - ``flat`` network interface becomes the default, otherwise
``noop`` is the default.
If this configuration option is not set, the default network interface is
determined by looking at the ``[dhcp]dhcp_provider`` configuration option
value. If it is ``neutron``, then ``flat`` network interface becomes the
default, otherwise ``noop`` is the default.
#. Define a provider network in neutron, which we shall refer to as the
"provisioning" network, and add it in under the neutron section in
ironic-conductor configuration file. Using ``neutron`` network interface
#. Define a provider network in the Networking service, which we shall refer to
as the "provisioning" network, and add it in the ``neutron`` section of
the ironic-conductor configuration file. Using ``neutron`` network interface
requires that ``provisioning_network`` and ``cleaning_network``
configuration options are set to a valid neutron network UUIDs or names,
otherwise cleaning or provisioning will fail to start::
configuration options are set to valid Networking service's network
identifiers (UUID or name); otherwise cleaning or provisioning will fail to
start::
[neutron]
...
@ -87,53 +84,54 @@ interface as stated above):
information about cleaning.
.. warning::
Please make sure ironic is exclusive to the provisioning and cleaning
network. Spawning instances by non-admin users in these networks and
getting access to ironic control plane is a security risk. For this
reason, the provisioning and cleaning network should be configured as
non-shared network in the admin tenant.
Please make sure that the Bare Metal service has exclusive access to the
provisioning and cleaning networks. Spawning instances by non-admin users
in these networks and getting access to the Bare Metal service's control
plane is a security risk. For this reason, the provisioning and cleaning
networks should be configured as non-shared networks in the ``admin``
tenant.
.. note::
Spawning a bare metal instance onto the provisioning network is
impossible, the deployment will fail. The node should be deployed onto a
different network than the provisioning network. When you boot a bare
metal instance from nova, you should choose a different network in
neutron for your instance.
metal instance from the Compute service, you should choose a different
network in the Networking service for your instance.
.. note::
The "provisioning" and "cleaning" networks may be the same neutron
provider network, or may be distinct networks. To ensure communication
between ironic and the deploy ramdisk works, it's important to ensure
that security groups are disabled for these networks, *or* the default
The "provisioning" and "cleaning" networks may be the same network or
distinct networks. To ensure that communication between the Bare Metal
service and the deploy ramdisk works, it is important to ensure that
security groups are disabled for these networks, *or* that the default
security groups allow:
* DHCP
* TFTP
* egress port used for ironic (6385 by default)
* egress port used for the Bare Metal service (6385 by default)
* ingress port used for ironic-python-agent (9999 by default)
* if using the iSCSI deploy method (``pxe_*`` and ``iscsi_*`` drivers),
the ingress port used for iSCSI (3260 by default)
* if using the direct deploy method (``agent_*`` drivers), the egress
port used for swift (typically 80 or 443)
port used for the Object Storage service (typically 80 or 443)
* if using iPXE, the egress port used for the HTTP server running
on the ironic conductor nodes (typically 80).
on the ironic-conductor nodes (typically 80).
#. This step is optional and applicable only if you want to use security
groups during provisioning and/or cleaning of the nodes. If not specified,
default security groups are used.
First define security groups in neutron to be used for provisioning
and/or cleaning networks. Then add the list of these security group
UUIDs under the neutron section in ironic-conductor configuration file
as shown below::
First define security groups in the Networking service, to be used for
provisioning and/or cleaning networks. Then add the list of these security
group UUIDs under the ``neutron`` section of ironic-conductor's
configuration file as shown below::
[neutron]
...
cleaning_network=$CLEAN_UUID_OR_NAME
cleaning_network_security_groups=[$LIST_OF_CLEAN_SEC_GROUPS]
cleaning_network_security_groups=[$LIST_OF_CLEAN_SECURITY_GROUPS]
provisioning_network=$PROVISION_UUID_OR_NAME
provisioning_network_security_groups=[$LIST_OF_PROVISION_SEC_GROUPS]
provisioning_network_security_groups=[$LIST_OF_PROVISION_SECURITY_GROUPS]
Multiple security groups may be applied to a given network, hence,
they are specified as a list.
@ -143,10 +141,10 @@ interface as stated above):
.. warning::
If security groups are configured as described above, do not
set the "port_security_enabled" flag to False for the corresponding
neutron network or port. This will cause the deploy to fail.
Networking service's network or port. This will cause the deploy to fail.
For example: if provisioning_network_security_groups configuration
option is used, ensure that "port_security_enabled" flag for
For example: if ``provisioning_network_security_groups`` configuration
option is used, ensure that "port_security_enabled" flag for the
provisioning network is set to True. This flag is set to True by
default; make sure not to override it by manually setting it to False.
@ -155,7 +153,8 @@ interface as stated above):
<http://docs.openstack.org/networking-guide/config-ml2.html>`_
for details.
#. Restart the ironic conductor and API services after the modifications:
#. Restart the ironic-conductor and ironic-api services after the
modifications:
- Fedora/RHEL7/CentOS7::
@ -167,8 +166,8 @@ interface as stated above):
sudo service ironic-api restart
sudo service ironic-conductor restart
#. Make sure that the conductor is reachable over the provisioning network
by trying to download a file from a TFTP server on it, from some
#. Make sure that the ironic-conductor is reachable over the provisioning
network by trying to download a file from a TFTP server on it, from some
non-control-plane server in that network::
tftp $TFTP_IP -c get $FILENAME
@ -178,7 +177,7 @@ interface as stated above):
Configuring nodes
=================
#. Multitenancy support was added in the 1.20 API version. The following
#. Multi-tenancy support was added in the 1.20 API version. The following
examples assume you are using python-ironicclient version 1.5.0 or higher.
They show the usage of both ``ironic`` and ``openstack baremetal`` commands.
@ -192,10 +191,11 @@ Configuring nodes
export OS_BAREMETAL_API_VERSION=1.20
#. Node's ``network_interface`` field should be set to valid network interface
that is listed in the ``[DEFAULT]/enabled_network_interfaces`` configuration
option in the ironic-api config. Set it to ``neutron`` to use neutron ML2
driver:
#. The node's ``network_interface`` field should be set to a valid network
interface. Valid interfaces are listed in the
``[DEFAULT]/enabled_network_interfaces`` configuration option in the
ironic-conductor's configuration file. Set it to ``neutron`` to use the
Networking service's ML2 driver:
- ``ironic`` command::
@ -208,11 +208,12 @@ Configuring nodes
--driver agent-ipmitool
.. note::
If the ``[DEFAULT]/default_network_interface`` configuration option was
If the ``[DEFAULT]/default_network_interface`` configuration option is
set, the ``--network-interface`` option does not need to be specified
when defining the node.
when creating the node.
#. To update existing node's network interface, use the following commands:
#. To update an existing node's network interface to ``neutron``, use the
following commands:
- ``ironic`` command::
@ -224,7 +225,7 @@ Configuring nodes
--network-interface neutron
#. The Bare Metal service provides the ``local_link_connection`` information to
the Networking service ML2 driver. The ML2 driver uses that information to
the Networking service's ML2 driver. The ML2 driver uses that information to
plug the specified port to the tenant network.
.. list-table:: ``local_link_connection`` fields
@ -239,7 +240,7 @@ Configuring nodes
- Required. Port ID on the switch, for example, Gig0/1.
* - ``switch_info``
- Optional. Used to distinguish different switch models or other
vendor specific-identifier. Some ML2 plugins may require this
vendor-specific identifier. Some ML2 plugins may require this
field.
Create a port as follows: