This patch set adds support for rootwrap in order to execute iptables. Co-Authored-By: Dmitry Tantsur <dtantsur@redhat.com> Change-Id: I7c424c17222f119730b8c5ac0daafd9906282e4d Closes-bug: #1495844
Hardware introspection for OpenStack Bare Metal
This is an auxiliary service for discovering hardware properties for a node managed by Ironic. Hardware introspection or hardware properties discovery is a process of getting hardware parameters required for scheduling from a bare metal node, given it's power management credentials (e.g. IPMI address, user name and password).
A special ramdisk is required to collect the information on a node. The default one can be built using diskimage-builder and ironic-discoverd-ramdisk element (see Configuration below).
- Free software: Apache license
- Source: http://git.openstack.org/cgit/openstack/ironic-inspector
- Bugs: http://bugs.launchpad.net/ironic-inspector
- Blueprints: https://blueprints.launchpad.net/ironic-inspector
- Downloads: https://pypi.python.org/pypi/ironic-inspector
- Python client library and CLI tool: python-ironic-inspector-client.
Refer to CONTRIBUTING.rst for instructions on how to contribute.
Note
ironic-inspector was called ironic-discoverd before version 2.0.0.
Version Support Matrix
ironic-inspector currently requires bare metal API
version 1.6 to be provided by Ironic. This version is
available starting with Ironic Kilo release.
Here is a mapping between Ironic versions and supported ironic-inspector versions. The Standalone column shows which ironic-inspector versions can be used in standalone mode with each Ironic version. The Inspection Interface column shows which ironic-inspector versions can be used with the Ironic inspection interface in each version of Ironic.
|
----------+--------------------+ Standalone|Inspection Interface |
+==============+==========+====================+ 1.0 +--------------+----------+--------------------+ 1.0 - 2.2 +--------------+----------+--------------------+ 1.1 - 2.X +--------------+----------+--------------------+
Note
2.X means we don't have specific plans on deprecating
support for this Ironic version. This does not imply that we'll support
it forever though.
Workflow
Usual hardware introspection flow is as follows:
Operator enrolls nodes into Ironic e.g. via ironic CLI command. Power management credentials should be provided to Ironic at this step.
Nodes are put in the correct state for introspection as described in Node States.
Operator sends nodes on introspection using ironic-inspector API or CLI (see Usage).
On receiving node UUID ironic-inspector:
- validates node power credentials, current power and provisioning states,
- allows firewall access to PXE boot service for the nodes,
- issues reboot command for the nodes, so that they boot the ramdisk.
The ramdisk collects the required information and posts it back to ironic-inspector.
On receiving data from the ramdisk, ironic-inspector:
- validates received data,
- finds the node in Ironic database using it's BMC address (MAC address in case of SSH driver),
- fills missing node properties with received data and creates missing ports.
Note
ironic-inspector is responsible to create Ironic ports for some or all NIC's found on the node. ironic-inspector is also capable of deleting ports that should not be present. There are two important configuration options that affect this behavior:
add_portsandkeep_ports(please refer toexample.conffor detailed explanation).Default values as of ironic-inspector 1.1.0 are
add_ports=pxe,keep_ports=all, which means that only one port will be added, which is associated with NIC the ramdisk PXE booted from. No ports will be deleted. This setting ensures that deploying on introspected nodes will succeed despite Ironic bug 1405131.Ironic inspection feature by default requires different settings:
add_ports=all,keep_ports=present, which means that ports will be created for all detected NIC's, and all other ports will be deleted. Refer to the Ironic inspection documentation for details.Separate API (see Usage) can be used to query introspection results for a given node.
Nodes are put in the correct state for deploying as described in Node States.
Starting DHCP server and configuring PXE boot environment is not part of this package and should be done separately.
Installation
Install from PyPI (you may want to use virtualenv to isolate your environment):
pip install ironic-inspector
Also there is a DevStack plugin for ironic-inspector - see CONTRIBUTING.rst for the current status.
Finally, some distributions (e.g. Fedora) provide ironic-inspector packaged, some of them - under its old name ironic-discoverd.
Configuration
Copy example.conf to some permanent place (e.g.
/etc/ironic-inspector/inspector.conf). Fill in at least
these configuration values:
os_username,os_password,os_tenant_name- Keystone credentials to use when accessing other services and check client authentication tokens;os_auth_url,identity_uri- Keystone endpoints for validating authentication tokens and checking user roles;database- where you want ironic-inspector SQLite database to be placed;dnsmasq_interface- interface on whichdnsmasq(or another DHCP service) listens for PXE boot requests (defaults tobr-ctlplanewhich is a sane default for TripleO-based installations but is unlikely to work for other cases).
See comments inside example.conf for the other possible configuration options.
Note
Configuration file contains a password and thus should be owned by
root and should have access rights like
0600.
ironic-inspector requires root rights for managing
iptables. It gets them by running ironic-inspector-rootwrap
utility with sudo. To allow it, copy
rootwrap.conf to the configuration directory (e.g. as
/etc/ironic-inspector/rootwrap.conf and create file
/etc/sudoers.d/ironic-inspector-rootwrap with the following
content:
stack ALL=(root) NOPASSWD: /usr/bin/ironic-inspector-rootwrap /etc/ironic-inspector/rootwrap.conf *
Note
rootwrap.conf must be writeable only by root.
Replace stack with whatever user you'll be using to run
ironic-inspector.
Configuring PXE
As for PXE boot environment, you'll need:
TFTP server running and accessible (see below for using dnsmasq). Ensure
pxelinux.0is present in the TFTP root.You need PXE boot server (e.g. dnsmasq) running on the same machine as ironic-inspector. Don't do any firewall configuration: ironic-inspector will handle it for you. In ironic-inspector configuration file set
dnsmasq_interfaceto the interface your PXE boot server listens on. Here is an example dnsmasq.conf:port=0 interface={INTERFACE} bind-interfaces dhcp-range={DHCP IP RANGE, e.g. 192.168.0.50,192.168.0.150} enable-tftp tftp-root={TFTP ROOT, e.g. /tftpboot} dhcp-boot=pxelinux.0You have to install and configure one of 2 available ramdisks: simple bash-based (see Using simple ramdisk) or more complex based on ironic-python-agent (See Using IPA).
Here is inspector.conf you may end up with:
[DEFAULT]
debug = false
[ironic]
identity_uri = http://127.0.0.1:35357
os_auth_url = http://127.0.0.1:5000/v2.0
os_username = admin
os_password = password
os_tenant_name = admin
[firewall]
dnsmasq_interface = br-ctlplane
Note
Set debug = true if you want to see complete logs.
Using simple ramdisk
Build and put into your TFTP the kernel and ramdisk created using the diskimage-builder ironic-discoverd-ramdisk element:
ramdisk-image-create -o discovery fedora ironic-discoverd-ramdiskYou need diskimage-builder 0.1.38 or newer to do it (using the latest one is always advised).
Configure your
$TFTPROOT/pxelinux.cfg/defaultwith something like:default introspect label introspect kernel discovery.kernel append initrd=discovery.initramfs discoverd_callback_url=http://{IP}:5050/v1/continue ipappend 3Replace
{IP}with IP of the machine (do not use loopback interface, it will be accessed by ramdisk on a booting machine).Note
There are some prebuilt images which use obsolete
ironic_callback_urlinstead ofdiscoverd_callback_url. Modifypxelinux.cfg/defaultaccordingly if you have one of these.
Using IPA
ironic-python-agent is a new ramdisk developed for Ironic. During the Liberty cycle support for ironic-inspector was added. This is experimental for now, but we plan on making IPA the default ramdisk in Mitaka cycle.
Note
You need at least 1.5 GiB of RAM on the machines to use this ramdisk.
To build an ironic-python-agent ramdisk, do the following:
Get the latest diskimage-builder:
sudo pip install -U "diskimage-builder>=1.1.2"Build the ramdisk:
disk-image-create ironic-agent fedora -o ironic-agentNote
Replace "fedora" with your distribution of choice.
Copy resulting files
ironic-agent.vmlinuzandironic-agent.initramfsto the TFTP root directory.
Next, set up $TFTPROOT/pxelinux.cfg/default as
follows:
default introspect
label introspect
kernel ironic-agent.vmlinuz
append initrd=ironic-agent.initramfs ipa-inspection-callback-url=http://{IP}:5050/v1/continue systemd.journald.forward_to_console=yes
ipappend 3
Replace {IP} with IP of the machine (do not use loopback
interface, it will be accessed by ramdisk on a booting machine).
Note
While systemd.journald.forward_to_console=yes is not
actually required, it will substantially simplify debugging if something
goes wrong.
Running
ironic-inspector --config-file /etc/ironic-inspector/inspector.conf
A good starting point for writing your own systemd unit should be one used in Fedora (note usage of old name).
Usage
Refer to HTTP-API.rst for information on the HTTP API. Refer to the client page for information on how to use CLI and Python library.
Using from Ironic API
Ironic Kilo introduced support for hardware introspection under name of "inspection". ironic-inspector introspection is supported for some generic drivers, please refer to Ironic inspection documentation for details.
Node States
The nodes should be moved to
MANAGEABLEprovision state before introspection (requires python-ironicclient of version 0.5.0 or newer):ironic node-set-provision-state <UUID> manageAfter successful introspection and before deploying nodes should be made available to Nova, by moving them to
AVAILABLEstate:ironic node-set-provision-state <UUID> provideNote
Due to how Nova interacts with Ironic driver, you should wait 1 minute before Nova becomes aware of available nodes after issuing this command. Use
nova hypervisor-statscommand output to check it.
Introspection Rules
Inspector supports a simple JSON-based DSL to define rules to run during introspection. Inspector provides an API to manage such rules, and will run them automatically after running all processing hooks.
A rule consists of conditions to check, and actions to run. If conditions evaluate to true on the introspection data, then actions are run on a node. All actions have "rollback actions" associated with them, which are run when conditions evaluate to false. This way we can safely rerun introspection.
Available conditions and actions are defined by plugins, and can be extended, see CONTRIBUTING.rst for details. See HTTP API for specific calls to define introspection rules.
Conditions
A condition is represented by an object with fields:
op the type of comparison operation, default available
operators include : eq, le, ge,
ne, lt, gt (basic comparison
operators), in-net (checks that IP address is in a given
network).
field a JSON path to the field
in the introspection data to use in comparison.
multiple how to treat situations where the
field query returns multiple results (e.g. the field
contains a list), available options are:
any(the default) require any to match,allrequire all to match,firstrequrie the first to match.
All other fields are passed to the condition plugin, e.g. numeric
comparison operations require a value field to compare
against.
Actions
An action is represented by an object with fields:
action type of action. Possible values are defined by
plugins.
All other fields are passed to the action plugin.
Default available actions include:
failfail introspection. Requires amessageparameter for the failure message.set-attributesets an attribute on an Ironic node. Requires apathfield, which is the path to the attribute as used by ironic (e.g./properties/something), and avalueto set.set-capabilitysets a capability on an Ironic node. Requiresnameandvaluefields, which are the name and the value for a new capability accordingly. Existing value for this same capability is replaced.extend-attributethe same asset-attribute, but treats existing value as a list and appends value to it. If optionaluniqueparameter is set toTrue, nothing will be added if given value is already in a list.
Setting IPMI Credentials
If you have physical access to your nodes, you can use ironic-inspector to set IPMI credentials for them without knowing the original ones. The workflow is as follows:
Ensure nodes will PXE boot on the right network by default.
Set
enable_setting_ipmi_credentials = truein the ironic-inspector configuration file, restart ironic-inspector.Enroll nodes in Ironic with setting their
ipmi_addressonly (or equivalent driver-specific property, as peripmi_address_fieldsconfiguration option).With Ironic Liberty use ironic API version
1.11, so that new node gets intoenrollprovision state:ironic --ironic-api-version 1.11 node-create -d <DRIVER> -i ipmi_address=<ADDRESS>Providing
ipmi_addressallows ironic-inspector to distinguish nodes.With Ironic Kilo or older, set maintenance mode on nodes. That's an important step, otherwise Ironic might interfere with introspection process. This is replaced by
enrollstate in Ironic Liberty.Start introspection with providing additional parameters:
new_ipmi_passwordIPMI password to set,new_ipmi_usernameIPMI user name to set, defaults to one in node driver_info.
Manually power on the nodes and wait.
After introspection is finished (watch nodes power state or use ironic-inspector status API) you can move node to
manageableand thenavailablestates - see Node States. With Ironic Kilo you have to move a node out of maintenance mode.
Note that due to various limitations on password value in different BMC, ironic-inspector will only accept passwords with length between 1 and 20 consisting only of letters and numbers.
Plugins
ironic-inspector heavily relies on plugins for data
processing. Even the standard functionality is largely based on plugins.
Set processing_hooks option in the configuration file to
change the set of plugins to be run on introspection data. Note that
order does matter in this option.
These are plugins that are enabled by default and should not be disabled, unless you understand what you're doing:
ramdisk_error-
reports error, if
errorfield is set by the ramdisk, also optionally stores logs fromlogsfield, see HTTP API for details. scheduler-
validates and updates basic hardware scheduling properties: CPU number and architecture, memory and disk size.
validate_interfaces-
validates network interfaces information.
Here are some plugins that can be additionally enabled:
example-
example plugin logging it's input and output.
raid_device(deprecated nameroot_device_hint)-
gathers block devices from ramdisk and exposes root device in multiple runs.
extra_hardware-
stores the value of the 'data' key returned by the ramdisk as a JSON encoded string in a Swift object.
Refer to CONTRIBUTING.rst for information on how to write your own plugin.
Troubleshooting
Errors when starting introspection
Invalid provision state "available"
In Kilo release with python-ironicclient 0.5.0 or newer Ironic defaults to reporting provision state
AVAILABLEfor newly enrolled nodes. ironic-inspector will refuse to conduct introspection in this state, as such nodes are supposed to be used by Nova for scheduling. See Node States for instructions on how to put nodes into the correct state.
Introspection times out
There may be 3 reasons why introspection can time out after some time
(defaulting to 60 minutes, altered by timeout configuration
option):
- Fatal failure in processing chain before node was found in the local cache. See Troubleshooting data processing for the hints.
- Failure to load the ramdisk on the target node. See Troubleshooting PXE boot for the hints.
- Failure during ramdisk run. See Troubleshooting ramdisk run for the hints.
Troubleshooting data processing
In this case ironic-inspector logs should give a good idea what went wrong. E.g. for RDO or Fedora the following command will output the full log:
sudo journalctl -u openstack-ironic-inspector
(use openstack-ironic-discoverd for version <
2.0.0).
Note
Service name and specific command might be different for other Linux distributions (and for old version of ironic-inspector).
If ramdisk_error plugin is enabled and
ramdisk_logs_dir configuration option is set,
ironic-inspector will store logs received from the
ramdisk to the ramdisk_logs_dir directory. This depends,
however, on the ramdisk implementation.
Troubleshooting PXE boot
PXE booting most often becomes a problem for bare metal environments with several physical networks. If the hardware vendor provides a remote console (e.g. iDRAC for DELL), use it to connect to the machine and see what is going on. You may need to restart introspection.
Another source of information is DHCP and TFTP server logs. Their location depends on how the servers were installed and run. For RDO or Fedora use:
$ sudo journalctl -u openstack-ironic-inspector-dnsmasq
(use openstack-ironic-discoverd-dnsmasq for version <
2.0.0).
The last resort is tcpdump utility. Use something like
:
$ sudo tcpdump -i any port 67 or port 68 or port 69
to watch both DHCP and TFTP traffic going through your machine.
Replace any with a specific network interface to check that
DHCP and TFTP requests really reach it.
If you see node not attempting PXE boot or attempting PXE boot on the wrong network, reboot the machine into BIOS settings and make sure that only one relevant NIC is allowed to PXE boot.
If you see node attempting PXE boot using the correct NIC but failing, make sure that:
- network switches configuration does not prevent PXE boot requests from propagating,
- there is no additional firewall rules preventing access to port 67 on the machine where ironic-inspector and its DHCP server are installed.
If you see node receiving DHCP address and then failing to get kernel and/or ramdisk or to boot them, make sure that:
- TFTP server is running and accessible (use
tftputility to verify), - no firewall rules prevent access to TFTP port,
- DHCP server is correctly set to point to the TFTP server,
pxelinux.cfg/defaultwithin TFTP root contains correct reference to the kernel and ramdisk.
Troubleshooting ramdisk run
Connect to the remote console as described in Troubleshooting PXE boot to see
what is going on with the ramdisk. The ramdisk drops into emergency
shell on failure, which you can use to look around. There should be file
called logs with the current ramdisk logs.