Skip to content

Latest commit

 

History

History
116 lines (84 loc) · 4.56 KB

USAGE.md

File metadata and controls

116 lines (84 loc) · 4.56 KB
Raw

Usage Scenario

Pumphouse application is intended to be used by cloud operators or service engineers to perform forklift upgrade of OpenStack-based cloud. This type of upgrade assumes that the new OpenStack installation works on the same set of hardware as original cloud. Workloads and resources are moved one by one from the original cloud to upgrade target cloud. Every physical hypervisor node of the existing cloud eventually gets added to upgraded cluster and re-installed as it`s Compute node.

Pumphouse assumes that upgrade target cloud is managed by Fuel deployment engine. For original cloud, both Fuel API and IPMI management modes are supported.

Install pumphouse

Pumphouse is a python application and can be installed as a pip package:

$ pip install pumphouse

Acquire initial nodes

Optionally, Pumphouse might allow to prepare nodes for deployment of initial Upgrade Target cloud. This preparation includes following steps:

  • install pumphouse package onto a node that has access to following networks: public/management API network of Source cloud; IPMI management network of Source cloud.
  • use Pumphouse to move virtual instances from certain nodes in existing cloud (via evacuate or live migrate or rebuild)
  • move released bare metal nodes to isolated L2 network (VLAN) which will serve as Admin network for Fuel deployment framework

Install Upgrade Target cloud

Currently, Pumphouse does not handle installation of 'seed' for Upgrade Target cloud. It is operator's responsibility to properly configure environment and initial set of Compute nodes via Fuel UI or API. Pumphouse will use the configuration of initail Compute nodes as a template to add upgraded hypervisors to target environment.

Configure source and target clouds in Pumphouse

Use configuration file config.yaml to configure source and target clouds, see example in samples/config.yaml

Discover cloud

Given the auth URL and credentials, we could learn basic information about the source and the destination clouds. Basic information includes the list of services available in those clouds and URLs to API endpoints of those services.

Moving workloads

We need to migrate workloads and virtual resources associated with them from (arbitrary) custom OpenStack cloud to the cloud based on the latest release of Mirantis OpenStack platform.

Migration must be as seamless as possible, with minimal downtime for workload being migrated.

Operators must be able to select workload resources they want to move based on different grouping criteria. The exact criteria depend on the type of resource being moved about. See details on migration strategies in RESOURCES document.

Migrate resources

Pumphouse provides Web service with HTTP-based API. See API specification in API document.

CLI script included with Pumphouse allows for granular migration of resources, including individual images, all servers of a project or all identity resources (tenants, users, roles, etc).

Maintenance mode on hosts

Pumphouse leverages Fuel deployment framework for upgrade of hypervisor hosts. Preparations for upgrade include removing all virtual resources from that host and disabling of provisioning new resources to it. This is also called maintenance mode of host.

Maintenance mode requires 2 steps:

  • moving targeted workloads to the Upgrade Target cluster with Pumphouse application;
  • moving remaining workloads to other hosts in source cloud with live/block migration or evacuate/rebuild.

Upgrade host

To perform the upgrade, you have to connect the host to Fuel-managed PXE network, reboot it into PXE bootstrap, configure it in the Fuel API and start deployment.

Following steps are performed by Pumphouse during upgrade procedure:

  • configure host to boot from PXE boot server using IPMI
  • reset host via IPMI Power Control function
  • wait for host to boot into bootstrap image via PXE and self-register as node in Fuel deployment framework
  • assign 'compute' role to the node via Fuel API
  • configure disk drives and network interfaces consistently with existing Compute nodes preconfigured by operator
  • start deployment of the host as Compute node OpenStack cluster via Fuel API

After deployment has started, you could check it's progress in Fuel UI or via Fuel CLI client.

Decomission source cloud

If the ultimate goal of the migration process is to completely replace original OpenStack infrastructure with new release, then as a final step of the move controllers of the source cloud should be decomissioned and added to Mirantis OpenStack cluster.