bac30b76fagave gpu.yml the 00_core/ prefix that the tasks_from lint asks for, and the include-existence lint then rejected it: the first resolves the path against tasks/, the second against the including file's directory. Neither the bare name nor the prefixed one holds under both bases, which is exactly what tests/lint/ansible/tasks/test_tasks_from_resolves.py requires of a nested include below a tasks_from entrypoint, because the base depends on how the caller entered the role rather than on the file itself. All three includes go through "{{ role_path }}/tasks/00_core/..." now. That is the form230c231e7bsettled on for web-app-mediawiki after the path_join spelling turned out to be invisible to _include_target_paths, which reads a literal basename or this shape and nothing else. The two distro includes in nvidia.yml carry the same defect and only became visible once gpu.yml stopped failing ahead of them. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Docker Server
Description
This role installs and maintains the Docker service, including Docker Compose, on Linux systems.
It is part of the Infinito.Nexus Project, maintained and developed by Kevin Veen-Birkenbach.
Overview
The role ensures that Docker and Docker Compose are present, integrates essential backup, repair, and health check roles, and supports cleanup or full reset modes for a fresh Docker environment.
When enabled via MODE_CLEANUP or MODE_RESET, it will automatically prune unused Docker resources.
MODE_RESET additionally restarts the Docker service after cleanup.
Features
-
Automated Installation
Installs Docker and Docker Compose via the system package manager. -
Integrated Dependencies
Includes backup, repair, and health check sub-roles -
Cleanup & Reset Modes
MODE_CLEANUP: Removes unused Docker containers, networks, images, and volumes.MODE_RESET: Performs cleanup and restarts the Docker service.
-
Handler Integration
Restart handler ensures the Docker daemon is reloaded when necessary.
Resource limits
resource.yml.j2 renders the compose cpus, mem_reservation, mem_limit and pids_limit of every service, and deploy.yml.j2 renders the swarm equivalents. Each value comes from services.<service>.<key> of the role's meta/services.yml, overridable per host in the inventory, and falls back to the fair share lookup('resource', …) computes from the host.
cpus also takes a percentage, which is read against RESOURCE_HOST_CPUS: the host's raw core count, without the two-CPU host reserve and without the division by running containers that the fair-share default applies. 100% is the whole machine, 50% is half of it:
applications:
web-app-docs:
services:
docs:
cpus: 50%
A percentage is the only portable way to size against the machine. Docker refuses a cpus above the host's core count (range of CPUs is from 0.01 to N.00), so cpus: "4" is a hard failure on a two-core host while 50% is valid everywhere. 100% resolves to exactly that ceiling and can never be rejected.
Shares are rounded to 0.01, Docker's granularity, and never fall below it. A percentage above 100 is refused, and so is a bare cpus: 0 — Docker reads a zero as uncapped rather than as none, so it has to be written as 100% to mean the machine.
A role that needs the number itself, to size a worker pool or a thread count, reads it with lookup('cpus', application_id, '<service>'). Reaching for services.<service>.cpus through lookup('config', …) returns the raw text and would size that pool from the string 95%.
Two limits:
- It caps, it does not reserve. Handing one container every CPU removes the backpressure that keeps it from crowding out a parallel deploy, and nothing stops several services from each claiming
100%. - The resource budget the
ressourcesCLI and the variant lint compute parsescpusas a plain float, so a service written as a percentage contributes nothing to that total.
License
This role is released under the Infinito.Nexus Community License (Non-Commercial) (CNCL).
See license details.