Includes:
- Rewrite of test-deploy workflow to use isolated inner dockerd with privileged mode.
- Switch logging drivers to 'json-file' when IS_CONTAINER=true for compatibility with non-systemd CI runners.
- Adjust Dockerfile to install docker CLI and simplify package setup.
- Improve inventory creation and deploy steps for CI stability.
- Fully compatible with Ansible 2.20 variable handling.
Conversation reference:
https://chatgpt.com/share/6930e285-9604-800f-aad8-7a81c928548c
This commit updates multiple roles to ensure compatibility with Ansible 2.20.
Several include paths and task-loading mechanisms required adjustments,
as Ansible 2.20 applies stricter evaluation rules for complex Jinja expressions
and no longer resolves certain relative include paths the way Ansible 2.18 did.
Key changes:
- Replaced legacy once_finalize.yml and once_flag.yml with the new structure
under tasks/utils/once/finalize.yml and tasks/utils/once/flag.yml.
- Updated all include_tasks statements to use 'path_join' with playbook_dir,
ensuring deterministic and absolute file resolution across roles.
- Fixed all network helper includes by converting direct relative paths such as
'roles/docker-compose/tasks/utils/network.yml' to proper Jinja-evaluated paths.
- Normalized MATOMO_* variable names for consistency with the updated variable
scope behavior in Ansible 2.20.
- Removed deprecated patterns that were implicitly supported in Ansible 2.18
but break under the more strict variable and path resolution model in 2.20.
These changes are part of the full migration step required to ensure the
infinito-nexus roles remain stable, deterministic, and forward-compatible with
Ansible 2.20.
Details of the discussion and reasoning can be found in this conversation:
https://chatgpt.com/share/69300a8d-24d4-800f-bec0-e895a695618a
- Replace legacy utils/run_once.yml with the new helpers utils/once_flag.yml and utils/once_finalize.yml
- Introduce utils/compose_up.yml to ensure docker-compose stacks are up and to flush handlers safely without coupling to run-once flags
- Migrate all affected roles (desk-*, dev-*, sys-ctl-*, sys-svc-*, web-app-*, web-svc-*, util-*) to the new run-once helpers
- Rework sys-svc-msmtp to auto-load Mailu once per deploy, check reachability, and reuse the running stack instead of requiring multiple playbook passes
- Adjust web-app-mailu to integrate cert deployment, handler flushing, and run-once handling so Mailu is fully initialized in a single deploy
- Improve Matomo, CDN, logout and CSP/health-check related roles to cooperate with the new compose_up / once_* pattern
- Simplify alarm/backup/timer/service orchestration (sys-ctl-alm-*, sys-bkp-provider, sys-timer-cln-bkps, etc.) by moving run-once logic into dedicated 01_core.yml files
- Update integration tests so utils/once_flag.yml and utils/once_finalize.yml are recognised as valid run-once providers, keeping the global run_once_* guarantees consistent
- Align frontend injection and service dependencies so Mastodon- and Mailu-related services can be brought up coherently within a single deployment cycle rather than several iterations
- Add CPU, memory and PID limits to all services in config/main.yml to prevent OOM
- Replace old LDAP admin bootstrap with new 02_admin.yml using OPENPROJECT_ADMINISTRATOR_* vars
- Standardize variable names (uppercase convention)
- Fix HTTPS/HSTS port check (443 instead of 433)
- Allow docker_restart_policy override in base.yml.j2
- Cleanup redundant LDAP admin runner in 01_ldap.yml
See: https://chatgpt.com/share/68d40c6e-ab9c-800f-a4a0-d9338d8c1b32
- Introduce new resource_filter plugin (mandatory hard_default, auto entity_name fallback)
- Replace get_app_conf calls with resource_filter in resource.yml.j2
- Add WEBSERVER_CPUS_EFFECTIVE, WEBSERVER_WORKER_PROCESSES, WEBSERVER_WORKER_CONNECTIONS to 05_webserver.yml
- Update Nginx templates (sys-svc-webserver, web-app-magento, web-app-nextcloud) to use new vars
- Extend svc-prx-openresty config with cpus/mem limits
- Add unit tests for resource_filter
Details: https://chatgpt.com/share/68d3a493-9a5c-800f-8cd2-bd2e7a3e3fda
Compute per-container CPU/RAM shares based on active services (web-/svc-*, enabled=true or undefined). Cast host facts to numbers, add safe min=1, and output compose-ready values. Include robust unit test.
Also: include resource.yml.j2 in base template and minor formatting tidy-up.
https://chatgpt.com/share/68d2d96c-9bf4-800f-bbec-d4f2c0051c06
Replaced duplicated inline build definitions in multiple docker-compose.yml.j2
templates with a shared include (roles/docker-container/templates/build.yml.j2).
This ensures consistent use of pull_policy: never and Dockerfile context across
services (Postgres, Bookwyrm, Bridgy Fed, Chess, Confluence, Jira, Moodle,
OpenProject, Pretix, Roulette Wheel, WordPress, XWiki, Simpleicons).
Conversation: https://chatgpt.com/share/68cd8f35-b764-800f-9b00-2c837103d2fb
Ensure that the database host alias is only attached to the database
containers themselves, not to dependent application containers. This
avoids DNS collisions where multiple containers expose the same alias
(e.g. 'postgres') on the same network, which led to connection refused
errors in XWiki.
See conversation: https://chatgpt.com/share/68cae4e5-94e4-800f-b291-d2acdb36af21
Added explicit aliases in the networks configuration for database containers
(Postgres/MariaDB). This guarantees that the configured 'database_host' is always
resolvable across external networks, fixing intermittent 'UnknownHostException'
issues when restarting dependent services (e.g., Confluence).
Ref: https://chatgpt.com/share/68cabfac-8618-800f-bcf4-609fdff432ed
- Add Host header to curl healthcheck when container_hostname is defined
- Use PRETIX_PRETIX_ALLOWED_HOSTS to fix Django 400 Bad Request during healthcheck
- Centralize PRETIX_HOSTNAME from container_hostname var
- Add Redis broker/result backend config for Celery
See: https://chatgpt.com/share/68b59c42-c0fc-800f-9bfb-f1137c59b3de
