Lessons Learned from a Large Scale OpenStack Deployment with Tripleo
summary
TripleO deploys an OpenStack Ussuri overcloud of 700+ compute nodes using one bare-metal undercloud, Ironic/PXE provisioning, three monolithic controllers and a three-node Ceph cluster; networking uses OVN, services run in Podman containers, installs use config-download with Ansible and PXE, and workload tests run with Rally while metrics come from Grafana/collectd. Result: successful scale to >700 nodes with no major regressions, Heat memory growth is linear and tunable, RabbitMQ and OVN hold, and multiple Ansible/OVN tunings and 15+ upstream bug fixes land. one undercloud, ansible config-download with limits
AI-written summary · about
chapters (14)
- 0:00Intro
- 1:05Performance and Scale
- 3:35What is Tripleo
- 4:42Tripleo Scale Testing
- 5:30Deployment Overview
- 7:30Software Specification
- 9:03Deployment Workflow
- 11:53Highlights
- 14:06Hit Engine Resource Consumption
- 15:01Heat Engine Resource Consumption
- 15:54Control Plane Performance
- 16:53Challenges
- 21:16Outcome
- 22:57Best Practices
related talks
54mTripleO: OpenStack on OpenStack
45mOpenstack-on-OpenStack Overview
18mChallenges of scaling a upstream Openstack Cloud
30mOpenStack Works, so now what?
18mLessons learned from multarchitecture, multinetwork OpenStack installation
38mHow OpenStack became boring (and successful)
17mBuilding the Rackspace Open Cloud with XenServer and OpenStack
27mExperience Report: Cloud Foundry Open Source Operations