Skip to content

jriguera/bosh-ironic-cpi-release

BranchesTags

Repository files navigation

bosh-ironic-cpi

A Bosh CPI (Cloud Provider Interface) to manage baremetal servers via Ironic (Standalone mode, without the rest of OpenStack components). The idea is make the most of OpenStack Ironic to provide an alternative way to deploy physical servers with Bosh. If you do not know about how amazing is Bosh, have a look here: http://bosh.io/

Tested using https://github.com/jriguera/ansible-ironic-standalone but it should work with any Ironic installation.

Some screenshots about the project:

$ bosh status
Config
             /home/jriguera/.bosh_config

Director
  Name       ironic-bosh
  URL        https://10.0.0.10:25555
  Version    1.3262.9.0 (00000000)
  User       admin
  UUID       1c9788b5-46c9-4a11-bc15-9963c438dfb5
  CPI        ironic_cpi
  dns        disabled
  compiled_package_cache disabled
  snapshots  disabled

Deployment
  not set

Bosh perspective:

$ bosh instances
https://10.0.0.10:25555
Acting as user 'admin' on deployment 'carbon-c-relay' on 'ironic-bosh'

Director task 25720

Task 25720 done

+------------------------------------------------+---------+------+---------+---------------+
| Instance                                       | State   | AZ   | VM Type | IPs           |
+------------------------------------------------+---------+------+---------+---------------+
| test/0 (e633088f-11a9-4ee6-b26f-2388f29d2d1d)* | running | dogo | pool    | 10.0.0.250    |
+------------------------------------------------+---------+------+---------+---------------+

(*) Bootstrap node

Instances total: 1

Ironic point of view of the instances:

$ ironic node-list
+--------------------------------------+--------+--------------------------------------+-------------+--------------------+-------------+
| UUID                                 | Name   | Instance UUID                        | Power State | Provisioning State | Maintenance |
+--------------------------------------+--------+--------------------------------------+-------------+--------------------+-------------+
| 666e20fe-4681-411e-85c9-21d615171714 | None   | None                                 | None        | enroll             | False       |
| 87f8401e-9f13-4afc-b015-bf75738b44b6 | None   | None                                 | None        | enroll             | False       |
| 05c389e8-70d4-4473-941a-d5cde6b05558 | None   | None                                 | None        | enroll             | False       |
| a1b555ed-9555-4a74-a091-c5cd00b58ed1 | None   | None                                 | power off   | available          | False       |
| db7f1933-d375-458d-b77e-75dee63a85ef | lab1   | None                                 | power off   | manageable         | False       |
| 4e5d7a7b-75e6-4d3e-bec8-38a047fdab18 | test-0 | e633088f-11a9-4ee6-b26f-2388f29d2d1d | power on    | active             | False       |
+--------------------------------------+--------+--------------------------------------+-------------+--------------------+-------------+

Motivations

  • Hack "Day" (actually months) project to learn about Bosh internals. One of the first ideas was about creating a smart CPI capable of proxying some requests to make use of physical hardware. This could be a different approach to have a CPI able to deploy VMs and physical servers using the same Bosh Director. One could have VMware/OpenStack Nova for VMs and Ironic for special jobs (like Cloud Foundry runners). The CPI would delegate the calls on the other one for VM management. Anyway, there are plans to support multi-CPI in Bosh, so I think this project will be focused only on Ironic.

  • Make most of the Ironic Standalone project, in order to define a standard way to manage physical servers using Ironic as backend. Apart of having provisioning tools (Ansible, or other programs) using the Ironic HTTP API to manage physical servers, this will bring the oportunity of managing the physical servers in a centralized way with Ironic. Ironic supports multiple Conductors (or even by using the concept of chassis) it is easy to use them as Availability Zones within Bosh.

This project is made with Python ... I know most of the CPIs are written in Ruby or Golang, so why Python? ... why not? It is a great language which does not need to be compiled, OpenStack is made with it and its libraries provide the functionality I needed (e.g. to create ConfigDrive(s)). Also, having a CPI in another language made with a different approach helps to define a way to document and understand the functionalities. The architecture of the source code src/bosh_ironic_cpi was created in a flexible way to be re-used, to write a new CPI, one only needs to create or change the CPIAction sub-classes defined in src/bosh_ironic_cpi/ironic_cpi/actions/ folder, each one on each file. Each class will be registered automatically and exceptions or returns will be transformed in JSON format and write via stdout and stderr. This was a hacking project, made to understand the Bosh internal workflow, I have not got enough time to focus on writing the (needed) tests :-/

General considerations

Before running this sofware, take into account:

  • Servers must be predefined on Ironic (as manageable state) or defined using the ironic_params properties. When the CPI runs, depending on the configuration, it wil take one of the available servers which fits with the properties (searching by MAC or by hardware specs: memory, cpu, disk ...) or it will define the baremetal server in Ironic. When the CPI will delete the node, it will remember if the server was predefined to avoid removing it from Ironic.

  • Baremetal servers need at least two disks. Because of how Bosh Agent works and how Ironic sets-up the ConfigDrive (as a small partition at the end of the first device /dev/sda), Bosh Agent is not able to manage partitions in such way on the first device, as consecuence, a second device (/dev/sdb) is needed for ephemeral data. A 3rd device would be used for persistent data.

  • The CPI can deploy normal OpenStack stemcells. They work on most of the hardware, but take into account there are some NICs or storage devices which would need aditional drivers. In that case, you will need to build your own stemcell including those drivers.

  • A physical server usually has several NICs available, but maybe some of them are not connected or not needed. When Bosh Agent runs, it ensures all the interfaces defined on the server must have an IP, if an NIC has no IP address, Bosh Agent fails. To overcome this limitation I have decided to assign an IP of the loopback local range 127.0.0.100/24 to the non used interfaces. That is why all the MAC addresses must be defined (or discovered) in Ironic, to create such static network definitions.

  • Bonding is needed! In the VM world it does not make sense, but in the physical worl, most of the times is desirable (unless there are a lot of servers and by taking aggregations one can assume losing some of them). In order to provide bonding, stemcells must be created with the proper settings and Bosh Agent functionality should be changed. A similar issue appears if one needs to define VLANs (besides the option of having OpenStack Neutron controlling the physical switch ports) for some NIC(s).

  • Limited support for persistent disks. Persistent disks are local to each server. A persistent disk device (usually /dev/sdc) cannot be attached or moved to a different server than the one where it was originally created, because is physically attached to the server. If a job uses a persistent disk and a pool of servers is defined, the CPI will select always the server where the disk lives: the MAC address of the server is encoded in the disk id, that is how to setup the nexus between persistent disk and server.

  • LVM and/or RAID are not supported. LVM would be nice, but I think it will require a lot of changes in Bosh Agent which makes no sense to enable support in a ephemeral world. RAID setup could be managed by Ironic if the servers are predefined.

Jobs defined in the release

The Bosh Release consist of two jobs. The core, called ironic_cpi, is the implementation of the Ironic CPI and webdav_metadata_registry is an optional job implementing the Bosh Registry API in Lua for Nginx. Making the most of the WebDAV protocol (which is already needed to manage the Config-Drive Metatata in Ironic Standalone) I have decided to create a simple implementation of the Registry API to delegate in WebDAV as backend.

Bosh Registry stores the settings needed by the Bosh Agent to carry on with the configuration the VM. These settings are a simple JSON file, no database is needed, it is possible to use WebDAB as storage already needed for the stemcells. The Lua implementation parses each request and does internal redirections to WebDAV metadata location, doing some checks and creating the JSON file. There are two ideas behing this implementation:

  • The Registry API server can live in the same WebDAV service as Ironic Config-Drive Metadata and Stemcell Images storage, centralizing all Bosh Agent specifications for every node in the same repository. The implementation is already included in https://github.com/jriguera/ansible-ironic-standalone, just run the setup-ironic-bosh-registry.yml playbook after changing/defining the Registry credentials of the Bosh Director. Instructions.

  • Run the WebDAV inside Bosh Director VM providing Stemcell/Image and Config-Drive Metadata storage together with the Registry API. If you are running Ironic as part of a big OpenStack deployment, this is the way to use this CPI for now, because it only supports WebDAV storage repositories for Stemcells/Images and Ironic Config-Drive data (sorry no Glance, Swift or S3).

Example Bosh-Init set-up and configuration

Use Bosh Init as described in https://bosh.io/docs/init.html . Take into account, Bosh-Init will use two CPIs, one just to deploy the Bosh Director VM and the ironic_cpi to be included on it. Follow the instructions for the platform you want to run the Bosh Director VM and change the settings below.

First of all, you have to add the ironic_cpi release to the releases section like this:

releases:
- name: bosh
  url: https://bosh.io/d/github.com/cloudfoundry/bosh?v=257.9
  sha1: 3d6168823f5a8aa6b7427429bc727103e15e27af
- name: bosh-vsphere-cpi
  url: https://bosh.io/d/github.com/cloudfoundry-incubator/bosh-vsphere-cpi-release?v=27
  sha1: 7b9cd2b47138b49cdf9c7329ec6d85324d572743
- name: bosh-ironic-cpi
  url: https://github.com/jriguera/bosh-ironic-cpi-release/releases/download/v1/bosh-ironic-cpi-1.tgz
  sha1: 5c361af42a2afcc9cdb9ef1d8a8420eabe7917c1

Now, add the new jobs to the section templates from this release. Only one *_cpi job is needed in this section:

jobs:
- name: bosh
  instances: 1

  templates:
  - {name: nats, release: bosh}
  - {name: postgres, release: bosh}
  - {name: blobstore, release: bosh}
  - {name: director, release: bosh}
  - {name: health_monitor, release: bosh}
  - {name: ironic_cpi, release: bosh-ironic-cpi}
  - {name: webdav_metadata_registry, release: bosh-ironic-cpi}

Remember, job webdav_metadata_registry is not needed if you are using Ironic Standalone and it was deployed with setup-ironic-boshregistry.yml following the instructions, otherwise just include both jobs.

In the director section, make sure you have changed the cpi_job to point to ironic_cpi job/template:

    director:
      address: 127.0.0.1
      name: ironic-bosh
      db: *db
      cpi_job: ironic_cpi
      user_management:
        provider: local
        local:
          users:
          - {name: admin, password: admin}
          - {name: hm, password: hm-password}

At the same indentation level, define the main CPI configuration properties:

    ironic_cpi:
      ironic_url: "http://IRONIC-API:6385/"
      ironic_auth_token: "fake"
      repository_stemcell_url: "http://IRONIC-API/images"
      repository_metadata_url: "http://IRONIC-API/metadata"
      metadata_publickeys:
      - "ssh-rsa AAAAB3NzaC1yc ... dsfasdfa"
      metadata_nameservers: ["8.8.8.8", "4.4.4.4"]

Change IRONIC-API with the IP address or dns name of Ironic API. If you are not using Ironic Standalone (so also webdav_metadata_registry job), put here the IP of the Registry specified in registry.host item. metadata_publickeys is a list of SSH public keys which will be incorporated to vcap user in the servers deployed.

Done!, now just run Bosh-Init: bosh-init deploy <manifest.yaml>, it will take half an hour, so for a coffee! There are examples of manifests for different platforms in the checks folder.

Stemcells

The CPI can operate with the OpenStack images, created with those parameters:

  • Disk format: qcow2
  • container Format: bare
bosh upload stemcell https://bosh.io/d/stemcells/bosh-openstack-kvm-ubuntu-trusty-go_agent

Those stemcells are optimized to run in KVM hypervisors, but because the Linux kernel includes a lot of device drivers, they should run on most of the common hardware. Take into account there is hardware which requires additional drivers or firmware to get it working, in such, case you will have to build your own stemcells.

Cloud-Config setup

Focusing in Bosh v2 style manifests, before starting using the new Bosh Director, you have to upload a cloud-config definition to describe and use the infrastructure. Given the following example:

azs:
- name: dogo
  cloud_properties:
    availability_zone: dogo

vm_types:

# Define and use these servers in Ironic
- name: pe-prod-dogo-lab-01
  cloud_properties:
    macs: ['e4:1f:13:e6:33:3c']
    ironic_params:
      driver: "agent_ipmitool"
      driver_info:
        ipmi_address: "10.100.0.3"
        ipmi_username: "admin"
        ipmi_password: "pass"
        deploy_kernel: "file:///var/lib/ironic/http/deploy/coreos_production_pxe.vmlinuz"
        deploy_ramdisk: "file:///var/lib/ironic/http/deploy/coreos_production_pxe_image-oem.cpio.gz"
- name: pe-prod-dogo-lab-02
  cloud_properties:
    macs: [ 'e4:1f:13:e6:d6:d4', 'e4:1f:13:e6:d6:d6', '00:0a:cd:26:f1:79', '00:0a:cd:26:f1:7a', 'e6:1f:13:e8:c4:eb']
    ironic_params:
      driver: "agent_ipmitool"
      driver_info:
        ipmi_address: "10.100.0.4"
        ipmi_username: "admin"
        ipmi_password: "pass"
        deploy_kernel: "file:///var/lib/ironic/http/deploy/ctinyipa-master.vmlinuz"
        deploy_ramdisk: "file:///var/lib/ironic/http/deploy/tinyipa-master.gz"
- name: pe-prod-dogo-lab-03
  cloud_properties:
    macs: ['e4:1f:13:e6:3d:38', 'e4:1f:13:e6:3d:3a', '00:0a:cd:26:f2:1c', '00:0a:cd:26:f2:1b']
    ironic_params:
      driver: "agent_ipmitool"
      driver_info:
        ipmi_address: "10.100.0.5"
        ipmi_username: "admin"
        ipmi_password: "pass"
        deploy_kernel: "file:///var/lib/ironic/http/deploy/ctinyipa-master.vmlinuz"
        deploy_ramdisk: "file:///var/lib/ironic/http/deploy/tinyipa-master.gz"

# Specific predefined server in Ironic
- name: e4-1f-13-e6-3f-42
  cloud_properties:
    macs: ['e4:1f:13:e6:3f:42']

# Pool definition
- name: pool
  cloud_properties:
    ironic_properties:
      local_gb: 500
      cpus: 16

# Persistent disk /dev/sdc (on each server)
disk_types:
- name: default
  disk_size: 1
  cloud_properties:
    device: /dev/sdc

# Two networks, Ironic DHCP pool to deploy and compile packages
# and static IPs for the jobs
networks:
- name: default
  type: manual
  subnets:
  - range: 10.0.0.0/24
    az: dogo
    name: pxe
    gateway: 10.0.0.1
    reserved:
      - 10.0.0.1-10.0.0.100
      - 10.0.0.150-10.0.0.255
    dns: [8.8.8.8]
    cloud_properties:
      macs: ['e4:1f:13:e6:3d:38']
    static:
      - 10.0.0.250
      - 10.0.0.251
- name: compilation
  type: dynamic

# The second server is reserved for compilation purposes, because it boots
# faster than the others ...
compilation:
  workers: 1
  reuse_compilation_vms: true
  vm_type: pe-prod-dogo-lab-02
  network: compilation

The sections are documented in https://bosh.io/docs/cloud-config.html , but there are cloud_properties parameters for some of them which have to be explained to understand the Bosh Ironic CPI functionalities.

Take into account that some properties depend on your hardware and Ironic settings. Check the Ironic documentation to know more.

vm_types section can be used in 3 different ways

  • To specifically define a server in Ironic, such server can be referenced in the deployment manifest. You can define as many servers as you need, and, in the same way Bosh CPI defines each server, it will be deleted from Ironic when it will be no longer needed in the deployment manifest. All the configuration remains in Bosh Cloud-Config.

    - name: pe-prod-dogo-lab-03
  cloud_properties:
    macs: ['e4:1f:13:e6:3d:38', 'e4:1f:13:e6:3d:3a', '00:0a:cd:26:f2:1c', '00:0a:cd:26:f2:1b']
    ironic_params:
      driver: "agent_ipmitool"
      driver_info:
        ipmi_address: "10.100.0.5"
        ipmi_username: "admin"
        ipmi_password: "pass"
        deploy_kernel: "file:///var/lib/ironic/http/deploy/ctinyipa-master.vmlinuz"
        deploy_ramdisk: "file:///var/lib/ironic/http/deploy/tinyipa-master.gz"

    Two parameters must be defined in cloud properties: macs a list of all the MAC addresses of the physical server (even if they will not be used, in such case, an IP of the local loopback range will be assigned, see General considerations section above) and ironic_params is a dictionary of parameters for Ironic, basically the IPMI settings of the physical server and the image used to deploy the stemcell. ironic_params values depends on the physical server and Ironic configuration, it can include all the parameters supported by the ironic.node.create API method such as: chassis_uuid, driver, driver_info, extra, uuid, properties, name, network_interface, resource_class. The example configuration here should be enough for most of the cases, just change the MACs and the IPMI parameters acording to your server, for more information, check https://github.com/openstack/python-ironicclient/blob/master/ironicclient/v1/node.py and https://github.com/jriguera/ansible-ironic-standalone

  • To reference a predefined server by MAC. In this case, Bosh CPI will perform a search over all physical servers defined in Ironic which are in state available (not in use). Specific parameters of the selected server are already defined in Ironic (IPMI and driver_info settings) and Bosh CPI does not touch and know them. When the server is not longer needed by Bosh, it will be put in available state again, ready to be used. This mode is useful to target a specific server for a deployment.

    - name: e4-1f-13-e6-3f-42
  cloud_properties:
    macs: ['e4:1f:13:e6:3f:42']

    Only one parameter is needed: macs, a list of macs to search the server (or servers) pre-defined in Ironic. The first available server which has one of the MACs will be used.

  • To define a pool of pre-defined servers by hardware properties. Similar to the previous use case, but instead of searching by MAC, Bosh CPI will search for a server which fits on the hardware properties required. The server has to be in manageable state to be selected. By using Ironic-Inspector to automatically discover and define the new servers in Ironic, without human intervention, they will become ready to use by Bosh CPI. Example:

    - name: pool
  cloud_properties:
    ironic_properties:
      local_gb: 500
      cpus: 16

    This will define a selection pool with two filters: local_gb wich will filter servers with at least 500 GB of disk storage and cpus which will select those ones with at least 16 cpus. The ironic_properties available are: memory_mb, cpu_arch, local_gb, cpus which are the same properties used by OpenStack Nova Scheduler as well the ones automatically defined by Ironic Inspector. When server is no longer needed, Bosh CPI will move it back to the pool.

disk_types section

Persistent disks are local to the server where it was originally created. There is no way to attach a physical disk to a different server (without using network tecnologies). If a job uses a persistent disk and a pool of servers is defined, the CPI will select always the server where the disk lives: the MAC address of the server is encoded in the disk id, that is how to setup the nexus between persistent disk and server.

Because of the combination of how Bosh Agent works and the Config-Drive partition created by Ironic driver at the end of the device where the stemcell is deployed, Bosh Agent cannot handle ephemeral partitions on the same disk, causing that a server needs at least 2 disks: /dev/sda for the operating system, /dev/sdb for ephemeral data like swap and logs. With this mapping, the next available disk is /dev/sdc (for persistent data):

- name: default
  disk_size: 1
  cloud_properties:
    device: /dev/sdc

The parameter disk_size does not make sense here, but it is required. The device is already physically attached to the server, and the size is determined by its capabilities. The parameter device is not required, if it is not present, the default value is /dev/sdc.

Also, there is a limitation naming the devices. Bosh Agent assumes always devices like 'sdX', and when it creates the partitions on the disk, it assumes the partitions will be named as sdX1, sdX2 ... if you specify a device like /dev/disk/by-id/scsi.... it will not fit with the partition schema and bosh Agent will fail. You have to specify devices with format '/dev/sdX' here. Moreover, be aware those devices can point to a different physical devices when the stemcell is being deployed vs when the stemcell runs. The kernel used to deploy the stemcells can discover the devices in a different order than the kernel of the stemcell, so it could happen that the ephemeral device would be /dev/sda, check the BIOS settings to change the order.

networks section

Different networks definition are supported, it is only limited by the configuration of network devices (switches, routers, ...) connected to the NICs. Usually networks used for compilacion can be dynamic, which means the IPs will be assigned from the DHCP pool of Ironic (check its configuration).

Example:

networks:
- name: default
  type: manual
  subnets:
  - range: 10.0.0.0/24
    az: dogo
    name: pxe
    gateway: 10.0.0.1
    reserved:
      - 10.0.0.1-10.0.0.100
      - 10.0.0.150-10.0.0.255
    dns: [8.8.8.8]
    cloud_properties:
      macs: ['e4:1f:13:e6:3d:38']
    static:
      - 10.0.0.250
      - 10.0.0.251
- name: compilation
  type: dynamic

There is only one parameter available for cloud properties here: macs, a list of MACs addresses from different servers which should be used in the present network. The list allows to determine which NICs (from the servers) must be assigned to each network, in case of a server with multiple NICs and connected to multiple networks.

A simple solution to avoid this weird setting could be allow an extra parameter in the networks block of the manifest to specify the MAC address, or allow the MAC address as a parameter in the networks.default array (apart of dns and gateway).

Defining manually a pool of physical servers

Normally this task should be done automatically by Ironic-Inspector, but those are the Ironic commands to do it manually. Change the variables and properties according to each one:

### lab1

NAME=lab1
MAC="e4:1f:13:e6:44:6a e4:1f:13:e6:44:6b"
IPMI=10.100.0.203
IPMI_USER=admin
IPMI_PASS=pass

# Define the node
ironic node-create -n "$NAME" \
    -d agent_ipmitool \
        -i ipmi_address=$IPMI \
        -i ipmi_username=$IPMI_USER \
        -i ipmi_terminal_port=20010 \
        -i ipmi_password=$IPMI_PASS \
    -i deploy_kernel="file:///var/lib/ironic/http/deploy/coreos_production_pxe.vmlinuz" \
    -i deploy_ramdisk="file:///var/lib/ironic/http/deploy/coreos_production_pxe_image-oem.cpio.gz"

# Get the UUID
UUID=$(ironic node-list | awk "/$NAME/ { print \$2 }")

# Add hardware propeties
ironic node-update "$UUID" add \
    properties/memory_mb="100000" \
    properties/cpus="16" \
    properties/local_gb="256"

# Add ports (macs)
for m in $MAC; do
    ironic port-create -n "$UUID" -a "$m"
done

# Set to manageable state in order to make it discoverable by the CPI
# States flow: http://docs.openstack.org/developer/ironic/_images/states.svg
ironic node-set-provision-state "$UUID" manage


### lab2

NAME=lab2
MAC="e4:1f:13:e6:44:d4 e4:1f:13:e6:44:d6 00:0a:cd:26:44:79 00:0a:cd:26:44:7a e6:1f:13:e8:44:eb"
IPMI=10.100.0.204
IPMI_USER=admin
IPMI_PASS=pass

# Define the node
ironic node-create -n "$NAME" \
    -d agent_ipmitool \
        -i ipmi_address=$IPMI \
        -i ipmi_username=$IPMI_USER \
        -i ipmi_terminal_port=20011 \
        -i ipmi_password=$IPMI_PASS \
    -i deploy_kernel="file:///var/lib/ironic/http/deploy/coreos_production_pxe.vmlinuz" \
    -i deploy_ramdisk="file:///var/lib/ironic/http/deploy/coreos_production_pxe_image-oem.cpio.gz"

# Get the UUID
UUID=$(ironic node-list | awk "/$NAME/ { print \$2 }")

# Add hardware propeties
ironic node-update "$UUID" add \
    properties/memory_mb="200000" \
    properties/cpus="32" \
    properties/local_gb="512"

# Add ports (macs)
for m in $MAC; do
    ironic port-create -n "$UUID" -a "$m"
done

# Set to manageable state in order to make it discoverable by the CPI
# States flow: http://docs.openstack.org/developer/ironic/_images/states.svg
ironic node-set-provision-state "$UUID" manage


### List nodes

ironic node-list
# +--------------------------------------+--------+--------------------------------------+-------------+--------------------+-------------+
# | UUID                                 | Name   | Instance UUID                        | Power State | Provisioning State | Maintenance |
# +--------------------------------------+--------+--------------------------------------+-------------+--------------------+-------------+
# | 78eafda7-e89a-4eec-a65e-29fac864088c | lab1   | None                                 | power off   | manageable         | False       |
# | 5958d8a1-4df7-406e-ae06-30f4bd44f1cf | lab2   | None                                 | power off   | manageable         | False       |
# +--------------------------------------+--------+--------------------------------------+-------------+--------------------+-------------+

Dev environment

Run ./bosh_prepare download to download the sources of the packages. In the examples folder there are example manifests to check the jobs with Bosh Lite.

Known issues

Since Bosh Agent assumes that new disks are always empty and clean (specially the ephemeral one), some old versions will fail to deploy because they try to create some folders which are there. There are two workarounds for this issue:

  • Delete (or overwrite with dd) the ephemeral disk manually.
  • Setup Ironic clean steps to automatically delete disks when the node is deleted. This is the implemented solution. The clean steps incorporated in the current stable Ironic only supports wiping disks and this operation takes a lot of time!. The new version adds an additional clean step which only destroys the partitions tables, which is faster than wiping the disk (and enough for this purpose). The clean steps can be defined in Ironic or in the CPI via configuration parameters. More info: http://docs.openstack.org/developer/ironic/deploy/cleaning.html

TODO

  • Write code tests
  • Find out how to support bonding on the Stemcells
  • Create specific stemcells for Ironic
  • Support for non defined NICs*
  • Create a specific stemcell for Ironic
  • VLANs support?*
  • Include RAID setup in predefined servers

* Changes in the Bosh Agent are required

Author & License

José Riguera López

Apache 2.0

About

Managing baremetal servers with Bosh and Ironic

Resources

Readme

License

Apache-2.0 license
Activity

Stars

3 stars

Watchers

1 watching

Forks

2 forks
Report repository

Releases 1

v1 Latest
Sep 14, 2016

Packages

No packages published

Languages