Hardware introspection daemon for OpenStack Ironic
Go to file
OpenStack Proposal Bot b338739037 Updated from global requirements
Change-Id: I51b8d11a58aaad12fe38811d4e40f7cbf5a35e56
2015-09-15 02:23:19 +00:00
devstack Add introspection rules support 2015-09-14 14:56:09 +02:00
ironic-inspector/locale Prerelease translation source update 2015-07-08 21:49:33 +02:00
ironic_inspector Add introspection rules support 2015-09-14 14:56:09 +02:00
.gitignore Copy gitignore from ironic 2015-08-19 12:32:10 +02:00
.gitreview Update .gitreview file for project rename 2015-06-12 23:12:30 +00:00
CONTRIBUTING.rst Add introspection rules support 2015-09-14 14:56:09 +02:00
HTTP-API.rst Add introspection rules support 2015-09-14 14:56:09 +02:00
LICENSE Apache license 2014-10-03 17:15:55 +02:00
README.rst Add introspection rules support 2015-09-14 14:56:09 +02:00
babel.cfg Add scripts to manage translations 2015-03-04 00:07:37 +00:00
example.conf Make list of ipmi_address-alike driver fields configurable 2015-09-01 09:25:11 +02:00
ironic-inspector.8 Final discoverd -> inspector rename/move pass 2015-06-15 15:34:00 +02:00
plugin-requirements.txt Store and expose introspection data 2015-08-27 20:18:54 -04:00
requirements.txt Updated from global requirements 2015-09-15 02:23:19 +00:00
setup.cfg Add introspection rules support 2015-09-14 14:56:09 +02:00
setup.py Updated from global requirements 2015-07-15 01:32:51 +00:00
test-requirements.txt Convert functional test to a framework and decouple from client 2015-07-23 09:31:19 +02:00
tox.ini Add func3 tox environment for functional testing with Python 3. 2015-08-13 14:51:58 +02:00


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).

Refer to CONTRIBUTING.rst for instructions on how to contribute.


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.

Ironic Version

Inspector (Discoverd) Version

----------+--------------------+ Standalone|Inspection Interface

+==============+==========+====================+ 1.0 +--------------+----------+--------------------+ 1.0 - 2.2 +--------------+----------+--------------------+ 1.1 - 2.X +--------------+----------+--------------------+


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.


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.


    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_ports and keep_ports (please refer to example.conf for 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.


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.


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 which dnsmasq (or another DHCP service) listens for PXE boot requests (defaults to br-ctlplane which 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.


Configuration file contains a password and thus should be owned by root and should have access rights like 0600.

As for PXE boot environment, you'll need:

  • TFTP server running and accessible (see below for using dnsmasq). Ensure pxelinux.0 is present in the TFTP root.

  • Build and put into your TFTP directory kernel and ramdisk from the diskimage-builder ironic-discoverd-ramdisk element:

    ramdisk-image-create -o discovery fedora ironic-discoverd-ramdisk

    You need diskimage-builder 0.1.38 or newer to do it (using the latest one is always advised).

  • 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_interface to the interface your PXE boot server listens on. Here is an example dnsmasq.conf:

    dhcp-range={DHCP IP RANGE, e.g.,}
    tftp-root={TFTP ROOT, e.g. /tftpboot}
  • Configure your $TFTPROOT/pxelinux.cfg/default with something like:

    default discover
    label discover
    kernel discovery.kernel
    append initrd=discovery.initramfs discoverd_callback_url=http://{IP}:5050/v1/continue
    ipappend 3

    Replace {IP} with IP of the machine (do not use loopback interface, it will be accessed by ramdisk on a booting machine).


    There are some prebuilt images which use obsolete ironic_callback_url instead of discoverd_callback_url. Modify pxelinux.cfg/default accordingly if you have one of these.

Here is inspector.conf you may end up with:

debug = false
identity_uri =
os_auth_url =
os_username = admin
os_password = password
os_tenant_name = admin
dnsmasq_interface = br-ctlplane


Set debug = true if you want to see complete logs.


Run as root:

ironic-inspector --config-file /etc/ironic-inspector/inspector.conf


Running as root is not required if ironic-inspector does not manage the firewall (i.e. manage_firewall is set to false in the configuration file).

A good starting point for writing your own systemd unit should be one used in Fedora (note usage of old name).


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 MANAGEABLE provision state before introspection (requires python-ironicclient of version 0.5.0 or newer):

    ironic node-set-provision-state <UUID> manage
  • After successful introspection and before deploying nodes should be made available to Nova, by moving them to AVAILABLE state:

    ironic node-set-provision-state <UUID> provide


    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-stats command 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.


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.

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,
  • all require all to match,
  • first requrie 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.


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:

  • fail fail introspection. Requires a message parameter for the failure message.
  • set-attribute sets an attribute on an Ironic node. Requires a path field, which is the path to the attribute as used by ironic (e.g. /properties/something), and a value to set.

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 = true in the ironic-inspector configuration file, restart ironic-inspector.

  • Enroll nodes in Ironic with setting their ipmi_address only (or equivalent driver-specific property, as per ipmi_address_fields configuration option).

    With Ironic Liberty use ironic API version 1.11, so that new node gets into enroll provision state:

    ironic --ironic-api-version 1.11 node-create -d <DRIVER> -i ipmi_address=<ADDRESS>

    Providing ipmi_address allows 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 enroll state in Ironic Liberty.

  • Start introspection with providing additional parameters:

    • new_ipmi_password IPMI password to set,
    • new_ipmi_username IPMI 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 manageable and then available states - 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.


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:


reports error, if error field is set by the ramdisk, also optionally stores logs from logs field, see HTTP API for details.


validates and updates basic hardware scheduling properties: CPU number and architecture, memory and disk size.


validates network interfaces information.

Here are some plugins that can be additionally enabled:


example plugin logging it's input and output.

raid_device (deprecated name root_device_hint)

gathers block devices from ramdisk and exposes root device in multiple runs.


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.


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 AVAILABLE for 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):

  1. Fatal failure in processing chain before node was found in the local cache. See Troubleshooting data processing for the hints.
  2. Failure to load the ramdisk on the target node. See Troubleshooting PXE boot for the hints.
  3. 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).


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:

  1. network switches configuration does not prevent PXE boot requests from propagating,
  2. 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:

  1. TFTP server is running and accessible (use tftp utility to verify),
  2. no firewall rules prevent access to TFTP port,
  3. DHCP server is correctly set to point to the TFTP server,
  4. pxelinux.cfg/default within 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.