# The 14 GB cold-start problem: Rebuilding self-hosted GitHub Actions runners on Azure Kubernetes Service

DevFeed: [The 14 GB cold-start problem: Rebuilding self-hosted GitHub Actions runners on Azure Kubernetes Service](<https://devfeed.tech/articles/the-14-gb-cold-start-problem-rebuilding-self-hosted-github-actions-runners-on-azure-kubernetes-service-105698.md>)

Original publisher: [Read original article](<https://www.thoughtworks.com/insights/articles/rebuilding-self-hosted-github-actions-runners-on-aks>)

Author: Renicius Pagotto Fostaini

Published: 2026-10-28T00:00:00Z

Content type: article

Language: en

Sources: [Thoughtworks Insights](<https://devfeed.tech/sources/thoughtworks-insights.md>)

Topics: [Azure Kubernetes Service](<https://devfeed.tech/topics/azure-kubernetes-service.md>), [self-hosted runner](<https://devfeed.tech/topics/self-hosted-runner.md>), [GitHub Actions](<https://devfeed.tech/topics/github-actions.md>), [CI/CD](<https://devfeed.tech/topics/cicd.md>), [internal developer platform automation](<https://devfeed.tech/topics/internal-developer-platform-automation.md>), [platform engineering automation](<https://devfeed.tech/topics/platform-engineering-automation.md>), [Azure](<https://devfeed.tech/topics/azure.md>)

Tags: [actions](<https://devfeed.tech/tags/actions.md>), [aks](<https://devfeed.tech/tags/aks.md>), [arc](<https://devfeed.tech/tags/arc.md>), [autoscaling](<https://devfeed.tech/tags/autoscaling.md>), [azure](<https://devfeed.tech/tags/azure.md>), [azure-container-apps](<https://devfeed.tech/tags/azure-container-apps.md>), [azure-kubernetes-service](<https://devfeed.tech/tags/azure-kubernetes-service.md>), [azure-virtual-machines](<https://devfeed.tech/tags/azure-virtual-machines.md>), [build](<https://devfeed.tech/tags/build.md>), [build-times](<https://devfeed.tech/tags/build-times.md>), [build-tools](<https://devfeed.tech/tags/build-tools.md>), [cache](<https://devfeed.tech/tags/cache.md>), [caching](<https://devfeed.tech/tags/caching.md>), [capacity](<https://devfeed.tech/tags/capacity.md>), [cold-start-problem](<https://devfeed.tech/tags/cold-start-problem.md>), [github](<https://devfeed.tech/tags/github.md>), [github-actions](<https://devfeed.tech/tags/github-actions.md>), [kubernetes](<https://devfeed.tech/tags/kubernetes.md>), [net-framework](<https://devfeed.tech/tags/net-framework.md>), [self-hosted](<https://devfeed.tech/tags/self-hosted.md>)

## AI overview

For this team, combining ephemeral Windows runners on Azure Kubernetes Service (AKS), GitHub Actions Runner Controller (ARC), cached container images and prebuilt toolchains reduced reported pipeline duration from about 40 minutes to 10-15 minutes. Serverless containers first enabled more parallel work and brought runs to about 30 minutes, but cold starts spent 5-10 minutes pulling a 14 GB image; on AKS, a cached pull reportedly took 10-20 seconds. The team separated lightweight runner coordination from scalable worker nodes, used off-hours scale-down, ephemeral OS disks and image garbage collection, and packaged reusable workflows for application teams. These results are specific to the implementation: scaling workers to zero can lose the cache and restore cold-start delays, while keeping warm capacity trades lower latency for idle cost. The excerpt also reports an approximately 66% compute-cost reduction (about $7,800 annualized), contingent on comparable measurement scope. AKS adds operational responsibilities; the article recommends measuring queue and end-to-end time, cold versus warm starts, costs, and behavior at configured concurrency limits.

## Source excerpt

The 14 GB cold-start problem: Rebuilding self-hosted GitHub Actions runners on Azure Kubernetes Service