Provisioning Bare Metal with OpenStack
47 likes20,328 views
This document discusses provisioning bare metal servers with OpenStack's bare metal driver. It provides background on the driver and its status. The driver allows OpenStack to directly control physical servers without a hypervisor. This improves performance by eliminating hypervisor overhead and enables sharing of physical resources. The document outlines how the driver works, including hardware discovery, image deployment via PXE, and network configuration. It also discusses plans to improve automation of hardware discovery and firmware management.
![Enabling the Driver
[DEFAULT]
compute_driver = nova.virt.baremetal.driver.BareMetalDriver
scheduler_host_manager =
nova.scheduler.baremetal_host_manager.BaremetalHostManager
firewall_driver = nova.virt.firewall.NoopFirewallDriver
ram_allocation_ratio = 1.0
reserved_host_memory_mb = 0](https://image.slidesharecdn.com/odshavana-provisioningbaremetalwithopenstack-130417175151-phpapp02/85/Provisioning-Bare-Metal-with-OpenStack-24-320.jpg)
![Tuning the Driver
[baremetal]
net_config_template =
/opt/stack/nova/nova/virt/baremetal/net-static.ubuntu.template
tftp_root = /tftpboot
power_manager = nova.virt.baremetal.ipmi.IPMI
driver = nova.virt.baremetal.pxe.PXE
instance_type_extra_specs = cpu_arch:{i386|x86_64|arm}
sql_connection = mysql://{user}:{pass}@{host}/nova_bm
http://docs.openstack.org/trunk/openstack-compute/
admin/content/list-of-compute-config-options.html](https://image.slidesharecdn.com/odshavana-provisioningbaremetalwithopenstack-130417175151-phpapp02/85/Provisioning-Bare-Metal-with-OpenStack-25-320.jpg)
Provisioning Bare Metal with OpenStack
- 1. Provisioning Bare Metal With OpenStack Devananda van der Veen @devananda HP Cloud Services
- 2. A Technical Dive into the BareMetal Driver ● Background & Status ● Why? ● How Deployment Works ● Dev/Test ● Further ideas...
- 3. Who am I?
- 4. History of the BareMetal Driver
- 5. Why?
- 6. Why? ● HPC & BMaaS Scott Beale / Laughing Squid
- 7. Why? ● HPC & BMaaS ● Stop re-inventing the wheel
- 8. Why? ● HPC & BMaaS ● Stop re-inventing the wheel ● New possibilities...
- 10. What is BareMetal? A Hypervisor Driver for Nova
- 11. What is BareMetal? A Hypervisor Driver for Nova like the drivers for: libvirt, xen, vmware, hyperv, powervm, ...
- 12. What is BareMetal? A Hypervisor Driver for Nova … but different.
- 13. “normal” Nova Compute Operating System nova-compute Hypervisor (xen, kvm, lxc, ...) VM VM VM VM Neighbors steal resources.... Memory wasted running OS inside each VM... Hypervisor adds overhead...
- 14. BareMetal Nova Compute Control services on a separate host! No bad neighbors! Native IO! No hypervisor!Operating System Your applications nova-compute PXE IPMI Operating System
- 15. One Host to Rule them All (well, not really)
- 16. Some Things had to Change... ● scheduler_host_manager ● (host, node) ● nova_bm
- 17. Many varieties of hardware ● Pluggable sub-drivers – Power, Image deployment, Volume, Network
- 18. Many varieties of hardware ● Pluggable sub-drivers – Power, Image deployment, Volume, Network – HW discovery? – BIOS & Firmware management? – Vendor-specific control modules?
- 19. Reuse & Recycle ● External dependencies – dnsmasq – ipmitool – open-iscsi – syslinux
- 21. Current Capabilities ● Pluggable Power sub-drivers – IPMI – Virtual – TileraPDU (in trunk)
- 22. Current Capabilities ● Pluggable Deployment sub-drivers – PXE – Tilera (in trunk)
- 23. Current Capabilities ● OpenStack Network integration (partial) ● Nova API extension – add/delete nodes and interfaces – show status
- 24. Enabling the Driver [DEFAULT] compute_driver = nova.virt.baremetal.driver.BareMetalDriver scheduler_host_manager = nova.scheduler.baremetal_host_manager.BaremetalHostManager firewall_driver = nova.virt.firewall.NoopFirewallDriver ram_allocation_ratio = 1.0 reserved_host_memory_mb = 0
- 25. Tuning the Driver [baremetal] net_config_template = /opt/stack/nova/nova/virt/baremetal/net-static.ubuntu.template tftp_root = /tftpboot power_manager = nova.virt.baremetal.ipmi.IPMI driver = nova.virt.baremetal.pxe.PXE instance_type_extra_specs = cpu_arch:{i386|x86_64|arm} sql_connection = mysql://{user}:{pass}@{host}/nova_bm http://docs.openstack.org/trunk/openstack-compute/ admin/content/list-of-compute-config-options.html
- 26. Deployment
- 27. Deployment ● Prerequisite:You need a running OpenStack cloud! – devstack – diskimage-builder + tripleo-image-elements “boot-stack” – or something else
- 28. Deployment ● Prerequisite: operational requirements – flavors mapped to hardware – images loaded in Glance ● deploy kernel & ramdisk ● machine images
- 29. Deployment ● Current(*) Prerequisites – HW already configured (BIOS, RAID, etc) – HW inventory ● MAC addr ● IPMI credentials ● CPU/RAM characteristics – HW enrolled via Nova BareMetal API (*) https://etherpad.openstack.org/HavanaBaremetalNextSteps https://wiki.openstack.org/wiki/BaremetalOperationsSpec
- 30. Deployment ● Enrollment nova baremetal-node-create --pm_address=... --pm_user=... --pm_password=... HYPERVISOR_HOSTNAME CPU RAM DISK nova baremetal-add-interface NODE_ID AA:BB:CC:DD:EE:FF
- 31. Deployment ● Deploy Process – Initiated by “nova boot” – nova-scheduler selects host & node – nova-compute prepares bootloader, notifies OS Network... – power-on & deploy image – reboot
- 32. Deployment (PXE)
- 33. Step-by-step Baremetal Compute Host Baremetal Node IPMI Power-On DHCP Request Serve Deploy K & R Expose disks via iSCSI partition & copy image Reboot command DHCP Request Serve User K & R Control given to user ”driver.spawn(...)” Cloud-Init Fetch deploy & user images Build TFTP config (*) PXE deploy process subject to change without notice Blue: nova-compute Red: nova-baremetal-deploy-helper
- 34. Setting up your Dev env Full walk-through: http://bit.ly/tripleo-dev
- 35. Setting up your Dev env ● A minimal OpenStack cloud + BareMetal driver – devstack – download: http://bit.ly/dib-bootstack (*) – build it yourself! git clone https://github.com/stackforge/diskimage-builder git clone https://github.com/stackforge/tripleo-image-elements git clone https://github.com/tripleo/incubator incubator/scripts/boot-elements -n boot-stack (*) YMMV. Build this locally for best results.
- 36. Setting up your Dev env ● Some VMs to mock hardware – create them by hand – automate it! git clone https://github.com/tripleo/bm_poseur bm_poseur/bm_poseur –vms 1 –arch i686 create-vm bm_poseur/bm_poseur get-macs
- 37. Setting up your Dev env ● Deploy ramdisk created for you ● But you need a cloud image: – download: http://bit.ly/dib-ubuntu – diskimage-builder/bin/disk-image-create ubuntu -a i386 -o ubuntu
- 38. Put it all together ● Start boot-stack VM ● Load ubuntu image into Glance incubator/scripts/load-image ubuntu.qcow2 ● Enroll VM with BareMetal ● Go!
- 39. What about real hardware? Only the networking is different
- 40. What about real hardware? Only the networking is different … except for when other things are different.
- 41. Plans
- 44. Plans ● Automate discovery ● Firmware management ● HA baremetal deploy nodes
- 45. Further Reading ● Driver Wiki – https://wiki.openstack.org/wiki/Baremetal ● Disk Image Builder – github/stackforge/diskimage-builder/blob/master/README.md ● TripleO team's developer resources – github/tripleo/incubator/blob/master/README.md – github/tripleo/incubator/blob/master/notes.md