Evaluating two years in production with Juju and Charmed OpenStack
summary
Two years of Juju-managed Charmed OpenStack run in production with 12 Nova nodes and 24 Ceph OST nodes, deployed across three racks on 100Gb switches and BGP/EVPN routing, but that host‑routing choice breaks Juju automation and requires custom scripts. Juju handles Ceph and OpenStack upgrades as rolling operations reliably, controller raft leadership and hook failures are single points of failure, and custom machine charms for switches (PXE + FRR + netplan) plus MAAS tags enable GPU PCI passthrough and observability; NFS proved flaky so Samba CTDB and Trilio backups fill gaps.
AI-written summary · about
chapters (17)
- 0:00Intro
- 1:15JUJU WHOAMI-ABOUT THE AUTHOR
- 2:31OPENSTACK @IBEO (2019-)
- 4:04RACK LAYOUT
- 4:48OPENSTACK NETWORKING USING MAAS+JUJU
- 5:54L3 APPROACH AUTOMATED OUTSIDE JUJU
- 8:31GPU PCI PASSTHROUGH USING JUJU (2/2)
- 11:17CEPH TRAP #1
- 13:09BASIC TUNING: CPU GOVERNOR
- 14:40UNDERSTANDING CHARMS (1/2): WHAT AWAITS THE OPERATOR?
- 16:12THE JUJU CONTROLLER (2/5)
- 20:23UPGRADING CHARMS
- 21:07OPENSTACK & CEPH UPGRADES
- 22:17SERIES UPGRADES
- 23:49DEPLOYING MELLANOX SWITCHES USING JUJU
- 25:36OPENSTACK NAS SHORTCOMINGS
- 26:51BACKUP IS YOUR PROBLEM
related talks
32mYour own personal supercomputer within 15 minutes or less
11mInterview
53mPrivate Cloud with OpenSource – "Now for Real!"
25mLOKI (Linux, OpenStack, Kubernetes, Infrastructure)
23mDeploying OpenSearch Solutions on Hybrid Multi-Cloud Environments
38mHow OpenStack became boring (and successful)
18mOpenstack Continious Performance tests (CPT)
21mBack to the original mission: an open cloud operating system