We’re excited to welcome Lucerne University of Applied Sciences and Arts (HSLU) as a Bronze Sponsor of DevOpsDays Zürich 2026!
HSLU stands for practice-driven education and applied research, exactly the mindset our community thrives on: learning by doing, sharing real stories, and building what works in the field.
Thanks, HSLU, we’re proud to have you with us!
DevOpsDays #DevOpsDaysZurich #DevOps #PlatformEngineering #SRE
Just wrapped reviewing #DevOpsDays Zurich 2026 proposals 🤯
387 talks
47 workshops
28 ignites
That’s a tidal wave of ideas from our community 🌊🙌 Huge thanks to everyone who submitted!
Program drops soon — stay tuned 👀
If you're building this right now, I'd sanity-check 3 questions:
Can jobs queue (not crash) when capacity is short? Adept hit this gap with orchestration vs Slurm.
Do you measure GPU busy vs allocated, not just "utilisation"?
Is preemption a documented promise users can plan around?
GPU scheduling is a governance system pretending to be a data structure.
The solution class isn't "pick the perfect scheduler."
It's make the allocation model legible:
→ Explicit baselines (quotas) so planning is possible
→ Borrowing of idle capacity so utilisation doesn't tank
→ Priority tiers with preemption contracts
→ Shared unit economics so finance and engineering argue from the same facts
Priority queues work when people believe the system is fair.
That belief is governance. The scheduler just enforces it.
#MLOps #PlatformEngineering #FinOps
Each phase optimises for something different. The Wild West optimises for local speed. Static quotas optimise for local safety. Flexible borrowing optimises for global throughput and, maybe more importantly, legitimacy. (Everyone understands the rules.)
I'll be writing more about queuing and priority over the next couple of weeks. There's a lot to unpack here.
If you're in one of these roles: you're doing something undeservedly difficult. The technical complexity alone is immense. Doing it while navigating organisational politics, competing priorities, and limited recognition? That takes something special.
Hats off to you. Your work matters... even when nobody says it!
New in Talos: Multi-doc config for networking. No more bricking a remote server just because there’s a single typo in a 200-line configuration file.
Talos Linux and Omni were purpose-built to solve the difficulty of bare metal. Our latest updates replace the high-stakes risk of physical deployments with cloud-like automated guardrails, enabling staged provisioning, seamless remote management, and stability rooted in automated governance rather than reactive troubleshooting. The result is infrastructure that is self-healing, immutable, and resistant to human […]
🛠️ Take your LLM skills to production! Join Rust for Interfacing with Language Models (LFWS309) and build real-world LLM apps in Rust with hands-on labs and expert guidance.
Focus on what matters in production:
🔹 Type-safe integrations
🔹 Streaming responses
🔹 Structured outputs
🔹 RAG pipelines
🔹 Agent workflows
Next session starts soon — enroll now: https://training.linuxfoundation.org/training/rust-for-interfacing-with-language-models-lfws309/
CNCF survey data shows cloud native technologies established as foundational infrastructure, with Kubernetes widely used in production.
Operational maturity increasingly shapes how organizations run AI workloads.
Watch the video highlights for key findings from the data and learn more at https://www.linuxfoundation.org/research/cncf-2025-annual-survey
#CloudNative #OpenSource #Kubernetes #CNCF #PlatformEngineering #DevOps #Infrastructure #AI
At some point I should write up on https://PhoenixTrap.com a segmented walkthrough of my #SelfHosted setup, including the things that the #Mastodon documentation leaves out about data retention and backups.
Someday.
#SelfHosting #MastoAdmin #FediAdmin #DevOps #PlatformEngineering
DevOpsDays Zurich
Romano Roth
strickvl
Sidero Labs
The Linux Foundation
Mark Gardner