CNV - Scale to Zero and Live Migration of Stateful Apps on Kubernetes

Join Felicitas, Head of Research and Development at Loophole Labs, for a technical deep dive into the world of Checkpoint Restore and Live Migration. Recorded at Cloud Native Vancouver 2025, this talk explores how to capture the complete state of an application, including CPU registers, memory, and file descriptors, to enable instant scaling and seamless maintenance.

Felicitas breaks down the mechanics behind major implementations like CRIU, gVisor, and Firecracker/MicroVMs, explaining how to amortize startup times for heavy workloads (like Java applications and LLMs) and achieve true "scale-to-zero" capabilities. The session concludes with a live demo of "Architect" running on Kubernetes, showcasing live process migration with state preservation.

For sponsorship, and talk submission visit: https://luma.com/lkt74bpf

📂 Resources & Slides: https://github.com/loopholelabs/architect-...

⏱️ Timestamps:
00:00 - Introduction & Technical Setup
00:51 - Who is Loophole Labs?
01:16 - What is Checkpoint Restore? (Capturing Application State)
02:13 - Use Case 1: Amortizing Startup Time (Java, LLMs)
03:46 - Use Case 2: Scaling to Zero & Serverless
05:47 - Use Case 3: Live Migrations & Zero-Downtime Maintenance
07:41 - Technical Implementation: CRIU (Checkpoint Restore in Userspace)
12:41 - Technical Implementation: gVisor (Sandboxing & User Space Kernel)
19:03 - Technical Implementation: Firecracker & MicroVMs
21:08 - PVM: Page Table vs. Virtual Machine (Alibaba Patches)
23:13 - Handling Network Migrations
24:03 - Live Demo: Architect on Kubernetes
31:25 - Q&A: Distributed Applications & Coordination
32:16 - Q&A: eBPF/XDP vs. Architect Networking
34:56 - Q&A: GPU Support & Virtualization

#Kubernetes #CloudNative #LiveMigration #CheckpointRestore #Serverless #gVisor #CRIU #DevOps #CNCFVancouver #eBPF