Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

How a placeholder container restores GPU workloads with CRIU

AI summary

A technical walkthrough organized from repository implementations explains how a Kubernetes recovery Pod uses an inert placeholder container to maintain Linux namespaces while an agent restores captured processes through CRIU. It distinguishes the lightweight placeholder image from the placeholder process and covers CUDA state restoration, multi-container mappings, scheduling preflight checks, cleanup, and the restore-complete signal.

In-depth insight

The design lets restored business processes resume without rerunning the original entrypoint, avoiding a second initialization that could load another model copy and compete for GPU resources.

[snapshot]7-placeholder与恢复流程

对照仓库实现(agent/Dockerfile 的 placeholder target、api/podcontract、agent/internal/controller、executor、nsmount、cmd/nsrestore)整理。总览见 [snapshot0-snapshot介绍]。

结论先说: placeholder 的唯一职责就是给 Restore 当落点。它不是 dump 工具,也不跑业务初始化。Capture 拍的是源 Pod;Restore 需要一个已经调度好、命名空间还活着、但进程尽量什么都不做的容器,agent 再把快照灌进去。


1. 两个容易混的「placeholder」

日常说 placeholder,其实叠了两层。编 agent 镜像时看到的那个,和恢复 Pod 里跑着的那个,不是同一件事。

#Kubernetes#CRIU#NVIDIA#CUDA#nsrestore#vLLM#PodSnapshot

Original · 8 min readRead the original
Group chat

No comments yet. Start the conversation.