Self-Hosted Docker & Container Build Runners
Short answer: Cold container pulls burn minutes on ephemeral runners. Strict runs jobs on dedicated hosts with warm local Docker BuildKit caches and uses Tier 5 maintenance to prune objects idle for >24 hours only when disk headroom drops.
Keep Docker layer caches local on NVMe while Tier 5 automatically prunes layers idle over 24 hours.
Drop-in workflow YAML & zero-sudo enrollment
Enroll your host with strict add-runner and route jobs directly via GitHub's native runs-on labels:
jobs:
container:
runs-on: [self-hosted, Linux, X64]
steps:
- uses: actions/checkout@v4
- run: docker build -t app:ci .
curl -fsSL https://strict.ai/install.sh | sh
strict add-runner owner/repo --labels self-hosted,Linux,X64,docker
How Strict keeps Docker & Builds runners healthy
| Subsystem | How it applies to Docker & Builds |
|---|---|
| Disk Cleanup | Tier 5 specifically prunes dangling and idle Docker images/containers older than 24 hours after Tiers 1–4 run—only triggered when free disk drops below lowGiB. |
| Preflight Doctor | disk-headroom prevents builds from hitting ENOSPC mid-image export; runner-hooks verifies before-job.sh drain barriers. |
| Tracked Toolchains | docker, go, node, bun, rust, python, runner |
| Cache & Balancing | Because Strict tracks CPU load-skew and RAM pressure alongside disk usage, heavy parallel docker build jobs spread across cooler hosts. |
When Strict fits vs. when hosted runners win
Use Strict for Docker & Builds when
- Persistent local BuildKit & base-image layers without paying $0.50/GB sticky disk fees
- Smart >24h Docker prune in Tier 5 preserves today's active base layers
- Private container registry access over Tailscale without exposing ports publicly
Use cloud-hosted runners when
- Untrusted arbitrary Dockerfile builds from public contributors
- Massive ephemeral fan-out matrix builds exceeding local host count
Frequently asked questions
Does Strict prune active Docker layers during a build?
No. Tier 5 targets Docker objects idle for more than 24 hours and only executes under maintenance.lock when free space drops below lowGiB, stopping as soon as targetGiB is reached.
How does Strict prevent runaway concurrent Docker builds from OOMing a box?
computeFleetDirective enters compute pressure above 90% RAM or sustained CPU overload and drains excess runner slots smoothly via before-job.sh.
Does Strict require Docker-in-Docker?
No. Runners invoke your host's native Docker or container engine directly on bare metal or VM storage.
Try Strict on one spare Mac in under two minutes
Signed Apple Silicon app or single-line CLI · Zero accounts required