systemd @ Meta: From upstream to production
summary
Meta runs systemd everywhere: bare metal, containers and VMs, currently on CentOS Stream 9 moving to 10 and shipping systemd 257 (skipping 256). Development is upstream‑first with backports, building RPMs via CBS, hosting CI in the CentOS GitLab namespace and running integration tests on Fedora Testing Farm. Deploys use staged rollouts (1%, 5%, 10% …) with RPM git tags and exploded source branches for reproducible debugging. Built a service‑based container flow by adding delegate namespaces and user/BPF mounts, added varlink interfaces, cut ~3,000 unused lines and sped builds ~33%, and keeps journald minimal, forwarding to syslog.
AI-written summary · about
chapters (12)
- 0:00systemd at Meta
- 1:03systemd adoption and community
- 3:10systemd components and journaling
- 5:20Upstream-first policy
- 6:22Upstream to production path
- 7:47RPM packaging and CI/CD
- 15:13Fedora and upstream contributions
- 17:23Container runtime with systemd
- 21:20Performance and varlink adoption
- 25:36Future plans and deployment
- 26:41Debugging systemd bugs
- 33:03Q&A
related talks
35mBuilding the Community Enterprise Operating System through CentOS Stream
33msystemd-sysext in Prod: What We Learned Extending /usr Without a Package Manager
31mSolving All the Problems with systemd
48mDemystifying Systemd for Embedded Systems
25mCentOS Stream: Progress so far
35mNew security features in systemd
50mPortable Services Are Ready To Use
22mWhat's new in systemd in 2016?
