Files
infrastructure/llm-server/llm/ansible
spektr 46dab5dca7
ci-static / secret-scan (push) Successful in 21s
ci-static / yamllint (push) Successful in 30s
ci-static / ansible-syntax (push) Successful in 28s
ci-static / markdown-links (push) Successful in 22s
ci-static / secret-scan (pull_request) Successful in 20s
ci-static / yamllint (pull_request) Successful in 30s
ci-static / ansible-syntax (pull_request) Successful in 27s
ci-static / markdown-links (pull_request) Successful in 20s
M6/GNAI-021: OpenWebUI-Theme gnai-catppuccin-mocha + Ghost-Favicon als read-only Mounts; ai_backup um /etc/gnmon und openwebui-data (SQLite-Snapshot) erweitert
- open_webui-Rolle: Theme (files/theme/gnai-catppuccin-mocha.css) und Favicon
  (files/favicon/*, byte-identisch zum Dashboard-Original) werden host-seitig
  deployt und im Quadlet read-only ueber die Image-Platzhalter in
  /app/backend/open_webui/static/ gemountet — update-persistent, weil der
  Mount host-seitig liegt und OpenWebUIs Start-Wipe von static/ eine ro-Datei
  nicht anfassen kann. Inode-Befund: copy->Rename tauscht den Inode, daher
  notify Restart + flush_handlers vor den Served-Checks.
- Served-Verifikation in der Rolle: /static/custom.css (Marker) und
  /static/favicon.png (byte-identisch via content_length).
- ai_backup: zusaetzliche Pfade etc/gnmon und
  var/lib/ghostnet-ai/production/openwebui-data; die live webui.db (+wal/shm)
  wird aus dem tar ausgeschlossen und stattdessen als konsistenter Snapshot
  via SQLite-Online-Backup-API (python3 stdlib) an derselben Archiv-Position
  angehaengt (tar uncompressed -> append -> gzip).
- Doku: GNAI-021 (neu) + ai-README-Index; Link auf GNMON-000 korrigiert.
- Live ausgerollt 14.11.2026 auf 10.0.0.90 (webui.yml + ai.yml konvergent
  changed=0): custom.css 200/5311 B mit Marker, favicon.png 200/2729 B,
  alle 6 ro-Mounts am laufenden Container gebunden.

Generated by Mistral Vibe.
Co-Authored-By: Mistral Vibe <vibe@mistral.ai>
2026-11-14 14:37:13 +01:00
..

LLM serving stack – Ansible control project

Executes the LLM serving stack for exactly one environment on ghostnet-ki (GNAI-008 section 7.4). Host baseline, Podman runtime, workload accounts and AMD GPU access are managed by ../../host/ansible; this project never changes host state.

Layout

  • inventories/<environment>/ – one inventory per KI environment (currently testing); deployment uses the ansible-deploy account with become: true.
  • playbooks/ai.yml – applies model_store then inference_engine.
  • roles/model_store/ – shared, pinned model weights under /var/lib/ghostnet-ai/shared/models (GNAI-010 sections 3 and 4).
  • roles/inference_engine/ – rootless Quadlets, network, image provisioning and the exclusive profile switch for llama.cpp (GNAI-010 sections 5 and 6).

Control node setup

Use the repository-wide control venv built from ../../../shared/ansible/requirements-control.txt (ansible-core==2.21.4), for example:

python3 -m venv .venv
.venv/bin/pip install -r shared/ansible/requirements-control.txt
cd llm-server/llm/ansible
../../../.venv/bin/ansible-playbook -i inventories/testing playbooks/ai.yml --syntax-check

Usage (Testing)

All commands use the operator deploy key ~/.ssh/id_ed25519_ghostnet_ai_ansible_deploy (see AGENTS.md; no ssh-agent required):

# Preview only – no changes, no downloads, no service start:
../../../.venv/bin/ansible-playbook -i inventories/testing playbooks/ai.yml \
  --check --diff --private-key ~/.ssh/id_ed25519_ghostnet_ai_ansible_deploy

# Provision model weights only (tag models):
../../../.venv/bin/ansible-playbook -i inventories/testing playbooks/ai.yml \
  --tags models --private-key ~/.ssh/id_ed25519_ghostnet_ai_ansible_deploy

# Render quadlets and start/verify the selected profile (tag service):
../../../.venv/bin/ansible-playbook -i inventories/testing playbooks/ai.yml \
  --tags service --private-key ~/.ssh/id_ed25519_ghostnet_ai_ansible_deploy

inference_engine_active_profile selects which rendered quadlet is started; all other profiles stay rendered but stopped. Check mode downloads no model and starts no service. A second identical run must report changed=0 (GNAI-010 section 10.4).

The full technical acceptance (64k/128k start, VRAM distribution, prefill ladder, restart behaviour) is an operator-run protocol documented in GNAI-010; this role automates deployment and basic health only.