feat: initial commit — Ansible roles storage-01/gpu-01 opérationnels

ansible/ansible.cfg

@@ -0,0 +1,15 @@
+[defaults]
+inventory          = inventory.yml
+roles_path         = roles
+remote_user        = ansible
+private_key_file   = ~/.ssh/id_ed25519
+host_key_checking  = False
+retry_files_enabled = False
+stdout_callback    = ansible.builtin.default
+result_format      = yaml
+interpreter_python = auto_silent
+vault_password_file = ../.vault_pass
+
+[privilege_escalation]
+become        = True
+become_method = sudo

ansible/group_vars/all/vars.yml

@@ -0,0 +1,8 @@
+---
+dns_domain: lab.local
+ntp_servers:
+  - 0.fr.pool.ntp.org
+  - 1.fr.pool.ntp.org
+
+cluster_network: 192.168.10.0/24
+wan_gateway: 192.168.1.254

ansible/group_vars/all/vault.yml

@@ -0,0 +1,30 @@
+$ANSIBLE_VAULT;1.1;AES256
+33306234383266653631633563663630353937323132323336356161626533373832313530393064
+3130383235333530336531636665623737663261316332370a313739353336306130643461653635
+35353961333236636263613761393935373762336633396337346661373062636131373632366339
+6562383163316162390a393865383736623132653531643064663338393164366166303532633362
+38343131656238653562323632303362303566363635316132383335326462316432336330366230
+63663732356366356234646664646539316531613131313932313237663365643230343063613763
+63633061663833353337626433666363363336646136376334303166613764323634323463393464
+37393533333437386666653663323331666130656161646437663734363161653534666661383339
+38623934643636643361643766356664393335353034373532313262666537373330353762336361
+64343465333534313239323432656135626237333434653034333364663061383965653838663763
+63653832313934326163623831653839613330346238643630663539623436373738323837386162
+66326466623764346666613266663530363664343037643766653562613034363239346639326664
+30346337663530623732333264383839396534383762623338313437633266656665366538633564
+63363633343731636234623330393838313237633064333939663236616365326130323735623339
+34386635623033636166633536633661643133653263646333326363613362303461376432373466
+36613538393334626266396531616366383931636665633862373331343365623962643362306133
+66363730626639336232626430636239643261666661396664396661393635306466363338306662
+35343432346461306330623536363330653262323863343162623965633164653035373432316232
+36303630306465393837373033656130636261316535323464613563656235643136663038366536
+36343062623932346161616134653562346666333266393563623938376535323637376265353164
+65346663373331393561653030356163633038376234663231336332653161636662323537306633
+38636661613535336463616662363033313961613434323033316230653865356233636138343336
+35323438656564613931656538646561663432363234653134323933626236646237326636353833
+36383838666263626632613462646561393035663435373736653861383063623862613131653332
+66393731623630376463623533336133353036386234633361653237626335323432643332313036
+31363032323763656530313230346466343537663664623931623933616166333135326437666231
+30363135643632633962613631656339616665356463653162653832366336663165646264636638
+66663866303733633037313131353033393263346138643139626539326431383762646565613437
+32333462353765653931613431366639323434356365376636646237303166323762

ansible/group_vars/gateway/vars.yml

@@ -0,0 +1,5 @@
+---
+wan_interface: enp4s0          # NIC1 — port 2 switch — VLAN 1 (192.168.1.200)
+lan_interface: enp6s0f3u2c2    # NIC2 — port 7 switch — VLAN 10 (192.168.10.1)
+wan_ip: 192.168.1.200
+lan_ip: 192.168.10.1

ansible/group_vars/gpu_hosts/vars.yml

@@ -0,0 +1,2 @@
+---
+hsa_override_gfx_version: "10.3.0"

ansible/host_vars/gpu-01/vars.yml

@@ -0,0 +1,14 @@
+---
+ansible_host: 192.168.10.20
+
+# llama-server — modèle servi au démarrage
+llama_model_path: "/mnt/models/bartowski/Qwen3-8B-GGUF/Qwen3-8B-Q4_K_M.gguf"
+llama_model_alias: "qwen3-8b"
+llama_ctx_size: 32768
+
+# Montages NFS (surcharge les defaults du rôle nfs_client)
+nfs_client_mounts:
+  - local_path: /mnt/nfs
+    remote_export: /srv/data/nfs/k8s
+  - local_path: /mnt/models
+    remote_export: /srv/data/models

ansible/host_vars/storage-01/vars.yml

@@ -0,0 +1,21 @@
+---
+ansible_host: 192.168.1.200
+
+# RAID5
+raid_uuid: "3add8360-fa01-47eb-8aa1-0e84bacbbc15"
+raid_mount: /srv/data
+
+# NFS exports
+nfs_exports:
+  - path: /srv/data/nfs/k8s
+    network: "{{ cluster_network }}"
+    options: "rw,sync,no_subtree_check,no_root_squash"
+  - path: /srv/data/models
+    network: "{{ cluster_network }}"
+    options: "rw,sync,no_subtree_check,no_root_squash"
+
+# Chemins services (sous le RAID)
+hermes_data_dir: /srv/data/hermes
+postgres_data_dir: /srv/data/postgres
+qdrant_data_dir: /srv/data/qdrant
+minio_data_dir: /srv/data/minio

ansible/inventory.yml

@@ -0,0 +1,18 @@
+all:
+  vars:
+    ansible_user: ansible
+
+  children:
+    gateway:
+      hosts:
+        storage-01:
+          ansible_host: 192.168.1.200
+    gpu_hosts:
+      hosts:
+        gpu-01:
+          ansible_host: 192.168.10.20
+          ansible_ssh_common_args: '-o StrictHostKeyChecking=no -o ProxyJump=root@192.168.1.200'
+    almalinux:
+      children:
+        gateway:
+        gpu_hosts:

ansible/playbooks/gpu-01.yml

@@ -0,0 +1,8 @@
+---
+- name: Configure gpu-01
+  hosts: gpu_hosts
+  roles:
+    - { role: common,        tags: [common] }
+    - { role: rocm,          tags: [rocm] }
+    - { role: nfs_client,    tags: [nfs_client] }
+    - { role: llama_server,  tags: [llama_server] }

ansible/playbooks/site.yml

@@ -0,0 +1,3 @@
+---
+- import_playbook: storage-01.yml
+- import_playbook: gpu-01.yml

ansible/playbooks/storage-01.yml

@@ -0,0 +1,13 @@
+---
+- name: Configure storage-01
+  hosts: gateway
+  roles:
+    - { role: common,       tags: [common] }
+    - { role: gateway,      tags: [gateway] }
+    - { role: dnsmasq,      tags: [dnsmasq] }
+    - { role: nfs_server,   tags: [nfs_server] }
+    - { role: qdrant,       tags: [qdrant] }
+    - { role: postgresql,   tags: [postgresql] }
+    - { role: minio,        tags: [minio] }
+    - { role: litellm,      tags: [litellm] }
+    - { role: hermes_agent, tags: [hermes_agent] }

ansible/requirements.yml

@@ -0,0 +1,4 @@
+collections:
+  - name: ansible.posix
+  - name: community.general
+  - name: community.postgresql

ansible/roles/common/defaults/main.yml

@@ -0,0 +1,19 @@
+---
+timezone: Europe/Paris
+
+base_packages:
+  - vim
+  - curl
+  - wget
+  - jq
+  - git
+  - htop
+  - chrony
+  - sudo
+  - bash-completion
+  - tar
+  - unzip
+  - net-tools
+  - bind-utils
+
+ansible_service_user: ansible

ansible/roles/common/handlers/main.yml

@@ -0,0 +1,9 @@
+---
+- name: Restart sshd
+  ansible.builtin.systemd:
+    name: sshd
+    state: restarted
+
+- name: Reboot if kernel updated
+  ansible.builtin.reboot:
+    reboot_timeout: 300

ansible/roles/common/tasks/main.yml

@@ -0,0 +1,71 @@
+---
+- name: Set hostname
+  ansible.builtin.hostname:
+    name: "{{ inventory_hostname }}"
+
+- name: Set timezone
+  community.general.timezone:
+    name: "{{ timezone }}"
+
+- name: Enable EPEL repository
+  ansible.builtin.dnf:
+    name: epel-release
+    state: present
+
+- name: Upgrade all packages
+  ansible.builtin.dnf:
+    name: "*"
+    state: latest
+    update_cache: true
+  tags: update
+  notify: Reboot if kernel updated
+
+- name: Install base packages
+  ansible.builtin.dnf:
+    name: "{{ base_packages }}"
+    state: present
+
+- name: Enable and start chrony
+  ansible.builtin.systemd:
+    name: chronyd
+    enabled: true
+    state: started
+
+- name: Create ansible group
+  ansible.builtin.group:
+    name: "{{ ansible_service_user }}"
+    state: present
+
+- name: Create ansible user
+  ansible.builtin.user:
+    name: "{{ ansible_service_user }}"
+    group: "{{ ansible_service_user }}"
+    groups: wheel
+    shell: /bin/bash
+    create_home: true
+    state: present
+
+- name: Allow ansible user passwordless sudo
+  ansible.builtin.copy:
+    dest: /etc/sudoers.d/ansible
+    content: "{{ ansible_service_user }} ALL=(ALL) NOPASSWD:ALL\n"
+    mode: '0440'
+    validate: /usr/sbin/visudo -cf %s
+
+- name: Set up SSH authorized key for ansible user
+  ansible.posix.authorized_key:
+    user: "{{ ansible_service_user }}"
+    key: "{{ lookup('file', '~/.ssh/id_ed25519.pub') }}"
+    state: present
+
+- name: Harden sshd
+  ansible.builtin.lineinfile:
+    path: /etc/ssh/sshd_config
+    regexp: "{{ item.regexp }}"
+    line: "{{ item.line }}"
+    state: present
+  loop:
+    - { regexp: '^#?PermitRootLogin', line: 'PermitRootLogin no' }
+    - { regexp: '^#?PasswordAuthentication', line: 'PasswordAuthentication no' }
+    - { regexp: '^#?X11Forwarding', line: 'X11Forwarding no' }
+  notify: Restart sshd

ansible/roles/dnsmasq/defaults/main.yml

@@ -0,0 +1,16 @@
+---
+dns_upstream:
+  - 8.8.8.8
+  - 8.8.4.4
+
+dns_cluster_hosts:
+  - name: storage-01
+    ip: 192.168.10.1
+  - name: gpu-01
+    ip: 192.168.10.20
+  - name: compute-01
+    ip: 192.168.10.11
+  - name: compute-02
+    ip: 192.168.10.12
+  - name: compute-03
+    ip: 192.168.10.13

ansible/roles/dnsmasq/handlers/main.yml

@@ -0,0 +1,9 @@
+---
+- name: Restart dnsmasq
+  ansible.builtin.systemd:
+    name: dnsmasq
+    state: restarted
+
+- name: Reload systemd
+  ansible.builtin.systemd:
+    daemon_reload: true

ansible/roles/dnsmasq/tasks/main.yml

@@ -0,0 +1,29 @@
+---
+- name: Install dnsmasq
+  ansible.builtin.dnf:
+    name: dnsmasq
+    state: present
+
+- name: Deploy dnsmasq config
+  ansible.builtin.template:
+    src: dnsmasq.conf.j2
+    dest: /etc/dnsmasq.conf
+    mode: '0644'
+  notify: Restart dnsmasq
+
+- name: Deploy dnsmasq systemd override (wait for USB ethernet)
+  ansible.builtin.copy:
+    dest: /etc/systemd/system/dnsmasq.service.d/wait-for-network.conf
+    content: |
+      [Unit]
+      After=network-online.target
+      Wants=network-online.target
+    mode: '0644'
+  notify: Reload systemd
+
+- name: Enable and start dnsmasq
+  ansible.builtin.systemd:
+    name: dnsmasq
+    enabled: true
+    daemon_reload: true
+    state: started

ansible/roles/dnsmasq/templates/dnsmasq.conf.j2

@@ -0,0 +1,31 @@
+# Géré par Ansible — ne pas modifier manuellement
+
+# Écoute uniquement sur le LAN cluster
+interface={{ lan_interface }}
+bind-interfaces
+
+# Domaine local
+domain={{ dns_domain }}
+local=/{{ dns_domain }}/
+expand-hosts
+
+# Ne pas lire /etc/resolv.conf pour les upstream
+no-resolv
+
+# Upstream DNS pour tout ce qui n'est pas lab.local
+{% for server in dns_upstream %}
+server={{ server }}
+{% endfor %}
+
+# Enregistrements statiques des machines du cluster
+{% for host in dns_cluster_hosts %}
+address=/{{ host.name }}.{{ dns_domain }}/{{ host.ip }}
+address=/{{ host.name }}/{{ host.ip }}
+{% endfor %}
+
+# Cache
+cache-size=1000
+
+# Logs (désactiver en prod si trop verbeux)
+log-queries
+log-facility=/var/log/dnsmasq.log

ansible/roles/gateway/handlers/main.yml

@@ -0,0 +1,5 @@
+---
+- name: Reload nftables
+  ansible.builtin.systemd:
+    name: nftables
+    state: reloaded

ansible/roles/gateway/tasks/main.yml

@@ -0,0 +1,32 @@
+---
+- name: Stop and disable firewalld
+  ansible.builtin.systemd:
+    name: firewalld
+    enabled: false
+    state: stopped
+
+- name: Enable IP forwarding
+  ansible.posix.sysctl:
+    name: net.ipv4.ip_forward
+    value: '1'
+    sysctl_set: true
+    state: present
+    reload: true
+
+- name: Install nftables
+  ansible.builtin.dnf:
+    name: nftables
+    state: present
+
+- name: Deploy nftables config
+  ansible.builtin.template:
+    src: nftables.conf.j2
+    dest: /etc/sysconfig/nftables.conf
+    mode: '0640'
+  notify: Reload nftables
+
+- name: Enable and start nftables
+  ansible.builtin.systemd:
+    name: nftables
+    enabled: true
+    state: started

ansible/roles/gateway/templates/nftables.conf.j2

@@ -0,0 +1,63 @@
+#!/usr/sbin/nft -f
+# Géré par Ansible — ne pas modifier manuellement
+
+flush ruleset
+
+table inet filter {
+    chain input {
+        type filter hook input priority 0; policy drop;
+
+        # Loopback
+        iif lo accept
+
+        # Connexions établies
+        ct state established,related accept
+
+        # ICMP
+        ip protocol icmp accept
+        ip6 nexthdr icmpv6 accept
+
+        # SSH depuis le LAN domestique et le LAN cluster
+        tcp dport 22 ip saddr { 192.168.1.0/24, 192.168.10.0/24 } accept
+
+        # DNS (dnsmasq)
+        udp dport 53 ip saddr 192.168.10.0/24 accept
+        tcp dport 53 ip saddr 192.168.10.0/24 accept
+
+        # NFS
+        tcp dport { 111, 2049 } ip saddr 192.168.10.0/24 accept
+        udp dport { 111, 2049 } ip saddr 192.168.10.0/24 accept
+
+        # Services données (accès cluster uniquement)
+        tcp dport { 5432, 6333, 6334, 9000, 9001 } ip saddr 192.168.10.0/24 accept
+
+        # Hermes gateway (accès LAN domestique + cluster)
+        tcp dport 8080 ip saddr { 192.168.1.0/24, 192.168.10.0/24 } accept
+
+        # LiteLLM (localhost uniquement — pas d'accès réseau direct)
+        # Port 4000 non exposé ici intentionnellement
+
+        log prefix "nft-drop: " drop
+    }
+
+    chain forward {
+        type filter hook forward priority 0; policy drop;
+
+        # Forwarding LAN cluster → WAN
+        iif {{ lan_interface }} oif {{ wan_interface }} ct state new,established,related accept
+        iif {{ wan_interface }} oif {{ lan_interface }} ct state established,related accept
+    }
+
+    chain output {
+        type filter hook output priority 0; policy accept;
+    }
+}
+
+table ip nat {
+    chain postrouting {
+        type nat hook postrouting priority 100;
+
+        # NAT masquerade pour le LAN cluster
+        oif {{ wan_interface }} ip saddr 192.168.10.0/24 masquerade
+    }
+}

ansible/roles/hermes_agent/defaults/main.yml

@@ -0,0 +1,12 @@
+---
+hermes_user: hermes
+hermes_home: /opt/hermes
+hermes_data_dir: /srv/data/hermes
+
+hermes_lm_base_url: "http://127.0.0.1:4000/v1"
+hermes_lm_api_key: "lm-studio"
+hermes_model: "hermes-default"
+hermes_context_length: 65536
+
+hermes_gateway_port: 8080
+hermes_gateway_host: "0.0.0.0"

ansible/roles/hermes_agent/handlers/main.yml

@@ -0,0 +1,9 @@
+---
+- name: Reload systemd
+  ansible.builtin.systemd:
+    daemon_reload: true
+
+- name: Restart hermes-agent
+  ansible.builtin.systemd:
+    name: hermes-agent
+    state: restarted

ansible/roles/hermes_agent/tasks/main.yml

@@ -0,0 +1,157 @@
+---
+# --- Dépendances système ---
+
+- name: Install Hermes Agent dependencies
+  ansible.builtin.dnf:
+    name:
+      - curl
+      - git
+      - tar
+    state: present
+
+# --- Utilisateur système ---
+
+- name: Create hermes system user
+  ansible.builtin.user:
+    name: "{{ hermes_user }}"
+    home: "{{ hermes_home }}"
+    shell: /bin/bash
+    system: false
+    create_home: true
+    state: present
+
+# --- Environnement shell utilisateur ---
+
+- name: Set HERMES_HOME in hermes user bashrc
+  ansible.builtin.lineinfile:
+    path: "{{ hermes_home }}/.bashrc"
+    line: "export HERMES_HOME={{ hermes_data_dir }}"
+    create: true
+    owner: "{{ hermes_user }}"
+    group: "{{ hermes_user }}"
+    mode: '0644'
+
+- name: Add hermes venv to PATH in bashrc
+  ansible.builtin.lineinfile:
+    path: "{{ hermes_home }}/.bashrc"
+    line: "export PATH={{ hermes_data_dir }}/hermes-agent/venv/bin:$PATH"
+    create: true
+    owner: "{{ hermes_user }}"
+    group: "{{ hermes_user }}"
+    mode: '0644'
+
+# --- Répertoire données (RAID5) ---
+
+- name: Create HERMES_HOME on RAID5
+  ansible.builtin.file:
+    path: "{{ hermes_data_dir }}"
+    state: directory
+    owner: "{{ hermes_user }}"
+    group: "{{ hermes_user }}"
+    mode: '0750'
+
+# --- Installation ---
+
+- name: Check if hermes binary is already installed
+  ansible.builtin.stat:
+    path: "{{ hermes_home }}/.local/bin/hermes"
+  register: hermes_binary
+
+- name: Install Hermes Agent via installer script
+  ansible.builtin.shell:
+    cmd: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
+    executable: /bin/bash
+  become: true
+  become_user: "{{ hermes_user }}"
+  environment:
+    HOME: "{{ hermes_home }}"
+    HERMES_HOME: "{{ hermes_data_dir }}"
+  when: not hermes_binary.stat.exists
+  register: hermes_install_result
+
+- name: Show installer output
+  ansible.builtin.debug:
+    var: hermes_install_result.stdout_lines
+  when: hermes_install_result is changed
+
+# --- Packages Python pour les skills ---
+
+- name: Install Python packages for Hermes skills
+  ansible.builtin.pip:
+    name: "qdrant-client>=1.12.0"
+    executable: "{{ hermes_data_dir }}/hermes-agent/venv/bin/pip"
+    state: present
+  become: true
+  become_user: "{{ hermes_user }}"
+  when: hermes_binary.stat.exists
+
+# --- Configuration ---
+
+- name: Ensure LM_BASE_URL in .env points to LiteLLM
+  ansible.builtin.lineinfile:
+    path: "{{ hermes_data_dir }}/.env"
+    regexp: '^LM_BASE_URL='
+    line: "LM_BASE_URL={{ hermes_lm_base_url }}"
+    create: false
+  notify: Restart hermes-agent
+  when: hermes_binary.stat.exists
+
+- name: Ensure LM_API_KEY in .env is correct
+  ansible.builtin.lineinfile:
+    path: "{{ hermes_data_dir }}/.env"
+    regexp: '^LM_API_KEY='
+    line: "LM_API_KEY={{ hermes_lm_api_key }}"
+    create: false
+  notify: Restart hermes-agent
+  when: hermes_binary.stat.exists
+
+- name: Deploy Hermes config.yaml
+  ansible.builtin.template:
+    src: config.yaml.j2
+    dest: "{{ hermes_data_dir }}/config.yaml"
+    owner: "{{ hermes_user }}"
+    group: "{{ hermes_user }}"
+    mode: '0640'
+  notify: Restart hermes-agent
+
+# --- Service systemd ---
+
+- name: Deploy hermes-agent systemd service
+  ansible.builtin.template:
+    src: hermes-agent.service.j2
+    dest: /etc/systemd/system/hermes-agent.service
+    mode: '0644'
+  notify:
+    - Reload systemd
+    - Restart hermes-agent
+
+- name: Enable and start hermes-agent
+  ansible.builtin.systemd:
+    name: hermes-agent
+    enabled: true
+    daemon_reload: true
+    state: started
+
+# --- Wrapper global ---
+
+- name: Deploy hermes global wrapper (/usr/local/bin/hermes)
+  ansible.builtin.copy:
+    dest: /usr/local/bin/hermes
+    mode: '0755'
+    content: |
+      #!/bin/bash
+      exec sudo -u {{ hermes_user }} \
+        HOME={{ hermes_home }} \
+        HERMES_HOME={{ hermes_data_dir }} \
+        {{ hermes_data_dir }}/hermes-agent/venv/bin/hermes "$@"
+
+# --- Firewall ---
+
+- name: Open Hermes gateway port in firewall
+  ansible.posix.firewalld:
+    port: "{{ hermes_gateway_port }}/tcp"
+    permanent: true
+    state: enabled
+    immediate: true
+  when: ansible_facts.services['firewalld.service'] is defined
+        and ansible_facts.services['firewalld.service']['state'] == 'running'

ansible/roles/hermes_agent/templates/config.yaml.j2

@@ -0,0 +1,13 @@
+model:
+  provider: lmstudio
+  base_url: "{{ hermes_lm_base_url }}"
+  default: "{{ hermes_model }}"
+  context_length: {{ hermes_context_length }}
+
+auxiliary:
+  compression:
+    context_length: {{ hermes_context_length }}
+
+gateway:
+  host: "{{ hermes_gateway_host }}"
+  port: {{ hermes_gateway_port }}

ansible/roles/hermes_agent/templates/hermes-agent.service.j2

@@ -0,0 +1,26 @@
+[Unit]
+Description=Hermes Agent Gateway
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+User={{ hermes_user }}
+Group={{ hermes_user }}
+
+Environment=HOME={{ hermes_home }}
+Environment=HERMES_HOME={{ hermes_data_dir }}
+Environment=LM_BASE_URL={{ hermes_lm_base_url }}
+Environment=LM_API_KEY={{ hermes_lm_api_key }}
+Environment=PATH={{ hermes_data_dir }}/hermes-agent/venv/bin:/usr/local/bin:/usr/bin:/bin
+
+ExecStart={{ hermes_home }}/.local/bin/hermes gateway run
+
+Restart=on-failure
+RestartSec=10s
+
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=hermes-agent
+
+[Install]
+WantedBy=multi-user.target

ansible/roles/litellm/defaults/main.yml

@@ -0,0 +1,15 @@
+---
+litellm_user: litellm
+litellm_venv: /opt/litellm/venv
+litellm_config_dir: /etc/litellm
+
+litellm_host: "127.0.0.1"   # loopback — Hermes est sur la même machine
+litellm_port: 4000
+litellm_master_key: "lm-studio"
+
+# Clé Anthropic — vient du vault
+litellm_anthropic_api_key: "{{ vault_anthropic_api_key }}"
+
+# Endpoint llama-server sur gpu-01
+litellm_local_llm_base_url: "http://192.168.10.20:1234/v1"
+litellm_local_llm_api_key: "lm-studio"

ansible/roles/litellm/files/hermes-switch

@@ -0,0 +1,76 @@
+#!/bin/bash
+# Switch le modèle de Hermes via LiteLLM (hermes-default alias)
+# Usage: hermes-switch [qwen|claude|status]
+
+CONFIG=/etc/litellm/config.yaml
+
+qwen_block='    litellm_params:\n      model: openai/qwen2.5-7b-instruct\n      api_base: http://192.168.10.20:1234/v1\n      api_key: lm-studio'
+claude_block='    litellm_params:\n      model: anthropic/claude-sonnet-4-6\n      api_key: os.environ/ANTHROPIC_API_KEY'
+
+current() {
+    grep -A3 "model_name: hermes-default" "$CONFIG" | grep "model:" | awk '{print $2}'
+}
+
+status() {
+    local cur=$(current)
+    echo "hermes-default → $cur"
+}
+
+switch_qwen() {
+    python3 - "$CONFIG" <<'EOF'
+import sys, re
+path = sys.argv[1]
+text = open(path).read()
+block = """  - model_name: hermes-default
+    litellm_params:
+      model: openai/qwen3-8b
+      api_base: http://192.168.10.20:1234/v1
+      api_key: lm-studio"""
+text = re.sub(
+    r'  - model_name: hermes-default\n    litellm_params:.*?(?=\n  -|\nlitellm_settings)',
+    block + '\n',
+    text, flags=re.DOTALL
+)
+open(path, 'w').write(text)
+print("OK")
+EOF
+}
+
+switch_claude() {
+    python3 - "$CONFIG" <<'EOF'
+import sys, re
+path = sys.argv[1]
+text = open(path).read()
+block = """  - model_name: hermes-default
+    litellm_params:
+      model: anthropic/claude-sonnet-4-6
+      api_key: os.environ/ANTHROPIC_API_KEY"""
+text = re.sub(
+    r'  - model_name: hermes-default\n    litellm_params:.*?(?=\n  -|\nlitellm_settings)',
+    block + '\n',
+    text, flags=re.DOTALL
+)
+open(path, 'w').write(text)
+print("OK")
+EOF
+}
+
+case "${1:-status}" in
+    qwen)
+        switch_qwen
+        systemctl restart litellm
+        echo "Hermes → Qwen3-8B (gpu-01)"
+        ;;
+    claude)
+        switch_claude
+        systemctl restart litellm
+        echo "Hermes → Claude Sonnet 4.6 (API Anthropic)"
+        ;;
+    status)
+        status
+        ;;
+    *)
+        echo "Usage: hermes-switch [qwen|claude|status]"
+        exit 1
+        ;;
+esac

ansible/roles/litellm/handlers/main.yml

@@ -0,0 +1,9 @@
+---
+- name: Reload systemd
+  ansible.builtin.systemd:
+    daemon_reload: true
+
+- name: Restart litellm
+  ansible.builtin.systemd:
+    name: litellm
+    state: restarted

ansible/roles/litellm/tasks/main.yml

@@ -0,0 +1,62 @@
+---
+- name: Install Python pip and devel
+  ansible.builtin.dnf:
+    name:
+      - python3-pip
+      - python3-devel
+    state: present
+
+- name: Create litellm system user
+  ansible.builtin.user:
+    name: "{{ litellm_user }}"
+    system: true
+    shell: /sbin/nologin
+    home: /opt/litellm
+    create_home: true
+    state: present
+
+- name: Install litellm[proxy] in venv
+  ansible.builtin.pip:
+    name: "litellm[proxy]"
+    virtualenv: "{{ litellm_venv }}"
+    virtualenv_command: python3 -m venv
+    state: present
+
+- name: Create config directory
+  ansible.builtin.file:
+    path: "{{ litellm_config_dir }}"
+    state: directory
+    owner: "{{ litellm_user }}"
+    group: "{{ litellm_user }}"
+    mode: '0750'
+
+- name: Deploy LiteLLM config
+  ansible.builtin.template:
+    src: config.yaml.j2
+    dest: "{{ litellm_config_dir }}/config.yaml"
+    owner: "{{ litellm_user }}"
+    group: "{{ litellm_user }}"
+    mode: '0640'
+  notify: Restart litellm
+
+- name: Deploy systemd service
+  ansible.builtin.template:
+    src: litellm.service.j2
+    dest: /etc/systemd/system/litellm.service
+    mode: '0644'
+  notify:
+    - Reload systemd
+    - Restart litellm
+
+- name: Deploy hermes-switch script
+  ansible.builtin.copy:
+    src: hermes-switch
+    dest: /usr/local/bin/hermes-switch
+    mode: '0755'
+
+- name: Enable and start litellm
+  ansible.builtin.systemd:
+    name: litellm
+    enabled: true
+    daemon_reload: true
+    state: started

ansible/roles/litellm/templates/config.yaml.j2

@@ -0,0 +1,36 @@
+model_list:
+
+  # Alias Hermes — changer uniquement ce bloc pour switcher le modèle de Hermes
+  # sans toucher à la configuration de Hermes lui-même
+  - model_name: hermes-default
+    litellm_params:
+      model: openai/qwen3-8b
+      api_base: {{ litellm_local_llm_base_url }}
+      api_key: {{ litellm_local_llm_api_key }}
+  # Pour passer Hermes sur Claude : remplacer les 3 lignes ci-dessus par :
+  #   model: anthropic/claude-sonnet-4-6
+  #   api_key: os.environ/ANTHROPIC_API_KEY
+
+  # Modèles accessibles directement par nom
+  - model_name: qwen3-8b
+    litellm_params:
+      model: openai/qwen3-8b
+      api_base: {{ litellm_local_llm_base_url }}
+      api_key: {{ litellm_local_llm_api_key }}
+
+  - model_name: claude-sonnet-4-6
+    litellm_params:
+      model: anthropic/claude-sonnet-4-6
+      api_key: os.environ/ANTHROPIC_API_KEY
+
+  - model_name: claude-opus-4-7
+    litellm_params:
+      model: anthropic/claude-opus-4-7
+      api_key: os.environ/ANTHROPIC_API_KEY
+
+litellm_settings:
+  drop_params: true
+  request_timeout: 600
+
+general_settings:
+  master_key: {{ litellm_master_key }}

ansible/roles/litellm/templates/litellm.service.j2

@@ -0,0 +1,25 @@
+[Unit]
+Description=LiteLLM Proxy — unified LLM gateway (Qwen local + Claude API)
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=simple
+User={{ litellm_user }}
+Group={{ litellm_user }}
+
+Environment=ANTHROPIC_API_KEY={{ litellm_anthropic_api_key }}
+
+ExecStart={{ litellm_venv }}/bin/litellm \
+    --config {{ litellm_config_dir }}/config.yaml \
+    --host {{ litellm_host }} \
+    --port {{ litellm_port }}
+
+Restart=on-failure
+RestartSec=10
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=litellm
+
+[Install]
+WantedBy=multi-user.target

ansible/roles/llama_server/defaults/main.yml

@@ -0,0 +1,15 @@
+---
+llama_server_port: 1234
+llama_server_host: "0.0.0.0"
+llama_amdgpu_targets: "gfx1030"
+rocm_path: "/opt/rocm"
+hsa_override_gfx_version: "10.3.0"
+
+# Modèle à charger au démarrage
+llama_model_path: ""
+llama_model_alias: ""
+llama_ctx_size: 32768
+llama_n_gpu_layers: 99
+llama_parallel: 1
+llama_embeddings: true
+llama_pooling: "mean"

ansible/roles/llama_server/handlers/main.yml

@@ -0,0 +1,9 @@
+---
+- name: Reload systemd
+  ansible.builtin.systemd:
+    daemon_reload: true
+
+- name: Restart llama-server
+  ansible.builtin.systemd:
+    name: llama-server
+    state: restarted

ansible/roles/llama_server/tasks/main.yml

@@ -0,0 +1,79 @@
+---
+- name: Install build dependencies
+  ansible.builtin.dnf:
+    name:
+      - cmake
+      - gcc-c++
+      - git
+      - patchelf
+      - hipblas-devel
+      - rocblas-devel
+      - hip-devel
+      - hipcc
+    state: present
+
+- name: Clone llama.cpp
+  ansible.builtin.git:
+    repo: https://github.com/ggerganov/llama.cpp
+    dest: /opt/llama.cpp
+    depth: 1
+    version: "{{ llama_server_commit | default('HEAD') }}"
+  register: llama_clone
+
+- name: Check if llama-server binary exists
+  ansible.builtin.stat:
+    path: /opt/llama.cpp/build/bin/llama-server
+  register: llama_server_binary_stat
+
+- name: Configure cmake build (ROCm HIP)
+  ansible.builtin.command:
+    cmd: >
+      cmake -B build
+      -DGGML_HIP=ON
+      -DAMDGPU_TARGETS={{ llama_amdgpu_targets }}
+      -DCMAKE_BUILD_TYPE=Release
+      -DROCM_PATH={{ rocm_path }}
+      -DCMAKE_PREFIX_PATH={{ rocm_path }}
+      -DCMAKE_HIP_COMPILER={{ rocm_path }}/llvm/bin/clang++
+    chdir: /opt/llama.cpp
+  environment:
+    PATH: "{{ rocm_path }}/bin:{{ ansible_env.PATH }}"
+  when: llama_clone is changed or not llama_server_binary_stat.stat.exists
+
+- name: Build llama-server
+  ansible.builtin.command:
+    cmd: cmake --build build --target llama-server -j{{ ansible_processor_nproc }}
+    chdir: /opt/llama.cpp
+  environment:
+    PATH: "{{ rocm_path }}/bin:{{ ansible_env.PATH }}"
+  when: llama_clone is changed or not llama_server_binary_stat.stat.exists
+
+- name: Deploy systemd service
+  ansible.builtin.template:
+    src: llama-server.service.j2
+    dest: /etc/systemd/system/llama-server.service
+    mode: '0644'
+  notify:
+    - Reload systemd
+    - Restart llama-server
+
+- name: Enable llama-server service
+  ansible.builtin.systemd:
+    name: llama-server
+    enabled: true
+    daemon_reload: true
+    state: started
+
+- name: Open port in firewall
+  ansible.posix.firewalld:
+    port: "{{ llama_server_port }}/tcp"
+    permanent: true
+    state: enabled
+    immediate: true
+
+- name: Disable lm-studio service (replaced by llama-server)
+  ansible.builtin.systemd:
+    name: lm-studio
+    enabled: false
+    state: stopped
+  ignore_errors: true

ansible/roles/llama_server/templates/llama-server.service.j2

@@ -0,0 +1,28 @@
+[Unit]
+Description=llama-server (llama.cpp OpenAI-compatible API)
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=simple
+User=root
+Environment=HSA_OVERRIDE_GFX_VERSION={{ hsa_override_gfx_version }}
+Environment=LD_LIBRARY_PATH={{ rocm_path }}/lib
+ExecStart=/opt/llama.cpp/build/bin/llama-server \
+    --model {{ llama_model_path }} \
+    --host {{ llama_server_host }} \
+    --port {{ llama_server_port }} \
+    --ctx-size {{ llama_ctx_size }} \
+    --n-gpu-layers {{ llama_n_gpu_layers }} \
+    --parallel {{ llama_parallel }} \
+    --alias {{ llama_model_alias }} \
+{% if llama_embeddings %}
+    --embeddings \
+    --pooling {{ llama_pooling }} \
+{% endif %}
+    --log-disable
+Restart=on-failure
+RestartSec=10
+
+[Install]
+WantedBy=multi-user.target

ansible/roles/lm_studio/TROUBLESHOOTING.md

@@ -0,0 +1,369 @@
+# LM Studio sur AlmaLinux 9 — Déblocages et solutions
+
+## Contexte
+
+- **Machine** : gpu-01, AlmaLinux 9.7, GCC 11, libstdc++ → max `GLIBCXX_3.4.29`
+- **Problème initial** : LM Studio refuse de démarrer avec `GLIBCXX_3.4.30 not found`
+- **Cause racine** : `watcher.node` (module Node natif livré avec LM Studio) a été compilé avec GCC 12 et déclare une dépendance ELF vers `GLIBCXX_3.4.30`. AlmaLinux 9 ne fournit que GCC 11 → `GLIBCXX_3.4.29` maximum.
+
+---
+
+## Blocage 1 — LD_PRELOAD ne peut pas satisfaire DT_VERNEED
+
+### Ce qui a été tenté
+
+Construire un shim `.so` qui exporte `_ZNSt18condition_variable4waitERSt11unique_lockISt5mutexE` sous le tag de version `GLIBCXX_3.4.30`, puis l'injecter via `LD_PRELOAD`.
+
+```c
+// glibcxx_compat.c
+__asm__(".symver _ZNSt18condition_variable4waitERSt11unique_lockISt5mutexE_compat,"
+        "_ZNSt18condition_variable4waitERSt11unique_lockISt5mutexE@GLIBCXX_3.4.30");
+```
+
+```
+gcc -shared -fPIC -Wl,--version-script=glibcxx_compat.map \
+    -o libglibcxx_compat.so glibcxx_compat.c /usr/lib64/libstdc++.so.6
+```
+
+### Pourquoi ça échoue
+
+Dans ELF, `DT_VERNEED` lie une exigence de version à une bibliothèque **nominalement identifiée**. La section `.gnu.version_r` de `watcher.node` dit :
+
+> « J'ai besoin de `GLIBCXX_3.4.30` fourni par `libstdc++.so.6`. »
+
+Le linker dynamique (`ld.so`) vérifie cette version **directement dans `libstdc++.so.6`**, pas dans les bibliothèques chargées via `LD_PRELOAD`. Un shim préchargé sous un autre nom de fichier ne peut pas satisfaire cette exigence — même s'il exporte le bon symbole sous le bon nom de version.
+
+### Erreur observée
+
+Malgré `LD_PRELOAD`, le daemon plante toujours avec :
+```
+/lib64/libstdc++.so.6: version GLIBCXX_3.4.30 not found
+```
+
+---
+
+## Blocage 2 — Patch texte simple ne suffit pas (hash ELF non mis à jour)
+
+### Ce qui a été tenté
+
+Remplacer la chaîne `GLIBCXX_3.4.30` par `GLIBCXX_3.4.29` directement dans le binaire avec `sed` ou un patch Python simple.
+
+### Pourquoi ça échoue
+
+La section `.gnu.version_r` contient des structures `Elf64_Vernaux` :
+
+```c
+typedef struct {
+    Elf64_Word  vna_hash;   // hash ELF du nom de version
+    Elf64_Half  vna_flags;
+    Elf64_Half  vna_other;
+    Elf64_Word  vna_name;   // offset dans la string table
+    Elf64_Word  vna_next;
+} Elf64_Vernaux;
+```
+
+Chaque entrée contient **deux références** au nom de version :
+1. `vna_name` → offset vers la chaîne dans `.dynstr`
+2. `vna_hash` → hash ELF de cette chaîne (précalculé)
+
+Un patch texte change la chaîne mais laisse `vna_hash` intact. Le linker dynamique vérifie le hash **en premier** — il ne trouve pas `hash(GLIBCXX_3.4.29)` dans `libstdc++.so.6` et rejette le module.
+
+### Erreur observée
+
+```
+/lib64/libstdc++.so.6: version GLIBCXX_3.4.29 not found
+```
+(nouveau message — la chaîne a changé mais le hash est toujours celui de `GLIBCXX_3.4.30`)
+
+---
+
+## Blocage 3 — Mauvais offset dans le parseur ELF
+
+### Ce qui a été tenté
+
+Écrire `patch-glibcxx-compat.py` pour parser `.gnu.version_r` et patcher hash + chaîne.
+
+### Erreur de décodage
+
+Structure `Elf64_Verneed` :
+```
+offset +0  : vn_version (2 bytes)
+offset +2  : vn_cnt     (2 bytes)
+offset +4  : vn_file    (4 bytes)  ← oublié !
+offset +8  : vn_aux     (4 bytes)
+offset +12 : vn_next    (4 bytes)
+```
+
+Le premier jet lisait `vn_aux` à `+4` (en sautant `vn_file`) — le parseur pointait vers de mauvaises entrées, ne trouvait rien à patcher.
+
+### Fix appliqué
+
+Corriger les offsets dans le script Python :
+```python
+vn_aux  = struct.unpack_from(f'{endian}I', data, vneed_pos + 8)[0]   # +8, pas +4
+vn_next = struct.unpack_from(f'{endian}I', data, vneed_pos + 12)[0]  # +12, pas +8
+```
+
+---
+
+## Blocage 4 — Patcher vers GLIBCXX_3.4.29 mais le symbole n'y existe pas
+
+### Ce qui a été tenté
+
+Rediriger la dépendance de `GLIBCXX_3.4.30` vers `GLIBCXX_3.4.29` (version juste en dessous, disponible sur AlmaLinux 9).
+
+### Pourquoi ça échoue
+
+Le symbole `_ZNSt18condition_variable4waitERSt11unique_lockISt5mutexE` n'existe **pas** sous `GLIBCXX_3.4.29` dans la libstdc++ système. Il est exporté sous `GLIBCXX_3.4.11` :
+
+```bash
+$ nm -D /lib64/libstdc++.so.6 | grep condition_variable | grep wait
+...
+_ZNSt18condition_variable4waitERSt11unique_lockISt5mutexE@@GLIBCXX_3.4.11
+```
+
+Patcher vers `.4.29` donnait encore `version GLIBCXX_3.4.29 not found` (pas de ce symbole sous cette version).
+
+### Fix appliqué
+
+Cibler `GLIBCXX_3.4.11` dans le patcher — c'est là que le symbole vit réellement :
+
+```python
+old_ver  = b'GLIBCXX_3.4.30'
+new_ver  = b'GLIBCXX_3.4.11'   # même longueur → patch in-place sûr
+old_hash = elf_hash('GLIBCXX_3.4.30')
+new_hash = elf_hash('GLIBCXX_3.4.11')
+```
+
+Les deux chaînes font 14 octets → remplacement in-place sans décalage de la table de chaînes.
+
+---
+
+## Blocage 5 — `bytearray.index(0, ...)` lève ValueError
+
+### Ce qui a été tenté
+
+Lire le nom de version depuis `.dynstr` :
+```python
+name_end = data.index(0, name_off)  # cherche le '\0' terminal
+```
+
+### Pourquoi ça échoue
+
+`bytearray.index()` attend un entier ou une séquence d'octets selon le contexte, mais la forme `data.index(0, ...)` sur `bytearray` lève `ValueError: 0 is not in bytearray` (comportement différent de `bytes`).
+
+### Fix appliqué
+
+```python
+name_end = data.index(b'\x00', name_off)  # forme bytes, toujours valide
+```
+
+---
+
+## Blocage 6 — `gcc -lstdc++` introuvable
+
+### Ce qui a été tenté
+
+Compiler le shim LD_PRELOAD (approche abandonnée) avec `-lstdc++`.
+
+### Pourquoi ça échoue
+
+AlmaLinux 9 ne fournit pas le lien symbolique `libstdc++.so` (sans version) nécessaire au linker. Seul `libstdc++.so.6` (avec version) est présent.
+
+### Fix appliqué
+
+Passer le chemin complet :
+```bash
+gcc -shared -fPIC -Wl,--version-script=glibcxx_compat.map \
+    -o libglibcxx_compat.so glibcxx_compat.c /usr/lib64/libstdc++.so.6
+```
+
+---
+
+## Blocage 7 — Systemd tuait le daemon après démarrage (`lms server start` quitte après 0)
+
+### Ce qui a été tenté
+
+Service `Type=simple` avec `ExecStart=lms server start --port 1234 --bind 0.0.0.0`.
+
+### Comportement observé
+
+`lms server start` :
+1. Lance le daemon `llmster` en arrière-plan
+2. Attend que le serveur HTTP soit prêt
+3. Affiche `Success! Server is now running on port 1234`
+4. **Quitte avec code 0**
+
+Systemd détecte la fin du processus principal → `KillMode=control-group` (défaut) → tue tout le cgroup → `llmster` mort → service `inactive`.
+
+Le daemon survivait parfois grâce à un double-fork qui lui permettait d'échapper au cgroup, mais c'était non fiable et `systemctl status` montrait `inactive` même quand le serveur tournait.
+
+### Ce qui n'a pas fonctionné en intermédiaire
+
+Lancer `llmster` directement via `ExecStart` + `ExecStartPost=lms server start` :
+
+```ini
+ExecStart=/opt/lmstudio/.lmstudio/llmster/0.0.12-1/llmster
+ExecStartPost=/bin/bash -c 'sleep 5 && lms server start --port 1234 --bind 0.0.0.0'
+```
+
+Erreur :
+```
+Failed to authenticate: Invalid passkey for lms CLI client.
+```
+
+`lms server start` utilise un passkey stocké/généré lors du démarrage du daemon via `lms`. Démarrer `llmster` directement bypasse ce mécanisme → authentification impossible depuis `lms`.
+
+### Fix appliqué — wrapper `run-llmster.sh`
+
+```bash
+#!/bin/bash
+LMS="${HOME}/.lmstudio/bin/lms"
+
+# lms gère le passkey ET le démarrage du daemon
+"$LMS" server start --port "$LMS_PORT" --bind "$LMS_BIND"
+
+# Rester en vie tant que llmster tourne → systemd garde le service "active"
+while pgrep -u "$(id -un)" -x llmster > /dev/null 2>&1; do
+    sleep 15
+done
+
+exit 1  # déclenche Restart=on-failure si le daemon meurt
+```
+
+Résultat :
+- Le wrapper est le processus principal (`Type=simple`)
+- `lms server start` gère correctement passkey + démarrage du daemon
+- La boucle maintient le service `active (running)` dans systemd
+- Si `llmster` meurt, le wrapper sort avec code 1 → `Restart=on-failure` relance tout
+
+---
+
+## Blocage 8 — Port 1234 inaccessible depuis le réseau
+
+### Symptôme
+
+`curl http://192.168.10.20:1234/v1/models` depuis storage-01 → `No route to host`
+
+Le serveur écoutait bien sur `0.0.0.0:1234` mais le pare-feu AlmaLinux bloquait le port.
+
+### Fix appliqué
+
+```bash
+firewall-cmd --permanent --add-port=1234/tcp
+firewall-cmd --reload
+```
+
+Ajouté dans `tasks/main.yml` via `ansible.posix.firewalld` pour idempotence.
+
+---
+
+## Solution finale
+
+| Composant | Approche retenue |
+|---|---|
+| Dépendance GLIBCXX | Patch ELF de `watcher.node` : réécriture de `vna_hash` + chaîne `GLIBCXX_3.4.30` → `GLIBCXX_3.4.11` |
+| Service systemd | Wrapper shell : `lms server start` + boucle de monitoring `pgrep llmster` |
+| Firewall | `firewall-cmd --permanent --add-port=1234/tcp` |
+
+### Fichiers du rôle Ansible
+
+```
+roles/lm_studio/
+├── files/
+│   ├── patch-glibcxx-compat.py    # parseur ELF, patch hash+string VERNEED
+│   ├── install-lmstudio-compat.sh # télécharge bundle, patche watcher.node, bootstrap
+│   └── run-llmster.sh             # wrapper systemd
+└── templates/
+    └── lm-studio.service.j2       # Type=simple, ExecStart=run-llmster.sh
+```
+
+### Vérification
+
+```bash
+# Depuis storage-01
+curl http://192.168.10.20:1234/v1/models
+# → {"data":[...],"object":"list"}
+
+# Sur gpu-01
+systemctl status lm-studio
+# → active (running), MainPID=run-llmster.sh, llmster dans le cgroup
+```
+
+---
+
+## Blocage 9 — llmster rebind sur 127.0.0.1 après chargement Vulkan
+
+### Symptôme
+
+`lms server start --bind 0.0.0.0` démarre correctement, mais après
+`lms load <modele> --gpu max` avec le backend Vulkan, llmster rebind son
+serveur HTTP sur `127.0.0.1:1234`. Le port 1234 devient inaccessible depuis
+le réseau LAN.
+
+```bash
+# Sur gpu-01 après lms load :
+ss -tlnp | grep 1234
+# LISTEN 0 128 127.0.0.1:1234   ← loopback seulement
+```
+
+### Cause
+
+Comportement interne de llmster lors de l'initialisation du backend Vulkan :
+le processus recrée son socket HTTP et le bind sur loopback plutôt que
+sur l'adresse configurée au démarrage.
+
+### Fix appliqué — socat proxy dans run-llmster.sh
+
+Après le chargement du modèle, un proxy socat est lancé sur l'IP LAN pour
+forwarder vers le loopback. Les deux sockets sont non-conflictuels (IPs
+différentes) :
+
+```bash
+LAN_IP=$(ip -4 addr show scope global | grep -oP '(?<=inet\s)\d+(\.\d+){3}' | head -1)
+socat TCP-LISTEN:1234,fork,bind="${LAN_IP}",reuseaddr TCP:127.0.0.1:1234 &
+```
+
+**Pourquoi ne pas relancer `lms server start --bind 0.0.0.0` ?**
+Relancer `lms server start` après le chargement recharge le modèle avec le
+contexte par défaut (4096 tokens), écrasant le `--context-length 65536`
+passé à `lms load`.
+
+---
+
+## Blocage 10 — Gemma 4 crash Vulkan sur prompts longs (>800 tokens)
+
+### Symptôme
+
+Avec `google/gemma-4-e4b` ou `google/gemma-4-e2b`, la première inférence
+avec un prompt de plus de ~800 tokens provoque un crash du processus llmster :
+
+```
+HTTP 400 {"error":"The model has crashed without additional information. (Exit code: null)"}
+```
+
+Après le crash, llmster recharge le modèle avec le contexte par défaut (4096),
+et les requêtes suivantes échouent avec `n_keep >= n_ctx`.
+
+### Cause
+
+Bug dans l'implémentation Vulkan de llama.cpp pour l'architecture Gemma 4.
+Gemma 4 utilise une attention **interleaved global/local** avec un sliding
+window de 512 tokens. Quand le prompt dépasse 512 tokens, les layers
+d'attention globale crashent dans le shader Vulkan.
+
+Seuil confirmé par test binaire :
+- OK à 500 tokens (~100x "You are a helpful assistant.")
+- CRASH à ~900 tokens (~130x)
+
+La VRAM n'est pas en cause : avec GQA aggressif et sliding window, le KV
+cache de Gemma 4 à 65536 tokens est ~2 GB seulement (bien dans les 12GB VRAM).
+
+### Fix
+
+Utiliser un modèle avec attention GQA standard (sans sliding window interleaved) :
+- **Qwen3.5-9B** ✓ — support Vulkan mature, GQA standard
+- **Qwen2.5-7B-Instruct** ✓ — alternative
+
+Ne pas utiliser Gemma 4 E2B ou E4B avec le backend Vulkan de LM Studio jusqu'à
+correction du bug upstream dans llama.cpp.

ansible/roles/lm_studio/defaults/main.yml

@@ -0,0 +1,18 @@
+---
+lm_studio_user: lmstudio
+lm_studio_home: /opt/lmstudio
+lm_studio_port: 1234
+lm_studio_host: "0.0.0.0"
+lm_studio_context_length: 65536
+lm_studio_gpu_offload: "max"
+
+# HSA override for unofficially supported GPUs (e.g. RX 6700XT = gfx1031 → 10.3.0)
+hsa_override_gfx_version: "10.3.0"
+
+# Modèle chargé au démarrage du service
+# Format HuggingFace : "organization/repo/fichier.gguf"
+# Laisser vide pour ne pas auto-charger (chargement manuel via lms load)
+lm_studio_default_model: ""
+
+# Chemin NFS où stocker les modèles (laisse vide = stockage local par défaut)
+lm_studio_models_nfs_path: ""

ansible/roles/lm_studio/files/install-lmstudio-compat.sh

@@ -0,0 +1,57 @@
+#!/usr/bin/env bash
+# Install LM Studio on AlmaLinux 9.
+#
+# AlmaLinux 9 ships GCC 11 → libstdc++ provides up to GLIBCXX_3.4.29.
+# LM Studio bundles watcher.node (compiled with GCC 12) which requires
+# _ZNSt18condition_variable4waitERSt11unique_lockISt5mutexE@GLIBCXX_3.4.30.
+# That symbol exists in the system libstdc++ under GLIBCXX_3.4.11.
+#
+# Fix: ELF-patch watcher.node to rewrite the VERNEED entry from
+# GLIBCXX_3.4.30 → GLIBCXX_3.4.11 (same symbol, correct version tag).
+# Both vna_hash and the version string are updated in-place.
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+
+# Ansible copy task places files in /tmp
+if [ -f "/tmp/patch-glibcxx-compat.py" ]; then
+    PATCHER="/tmp/patch-glibcxx-compat.py"
+else
+    PATCHER="${SCRIPT_DIR}/patch-glibcxx-compat.py"
+fi
+
+# --- Fetch installer metadata ---
+echo "[lmstudio-install] Fetching installer metadata..."
+INSTALL_SH=$(curl -fsSL https://lmstudio.ai/install.sh)
+VERSION=$(echo "$INSTALL_SH" | grep '^APP_VERSION=' | cut -d'"' -f2)
+ARTIFACT_URL_BASE=$(echo "$INSTALL_SH" | grep '^ARTIFACT_DOWNLOAD_URL=' | cut -d'"' -f2)
+
+ARTIFACT="${VERSION}-linux-x64.full.tar.gz"
+URL="https://${ARTIFACT_URL_BASE}/${ARTIFACT}"
+
+TMPDIR=$(mktemp -d)
+trap 'rm -rf "$TMPDIR"' EXIT
+
+echo "[lmstudio-install] Downloading ${ARTIFACT}..."
+curl -fsSL "$URL" -o "$TMPDIR/bundle.tar.gz"
+
+echo "[lmstudio-install] Extracting bundle..."
+tar xf "$TMPDIR/bundle.tar.gz" -C "$TMPDIR"
+
+# Patch watcher.node before bootstrap so the daemon never sees GLIBCXX_3.4.30
+echo "[lmstudio-install] Patching watcher.node (GLIBCXX_3.4.30 → GLIBCXX_3.4.11)..."
+find "$TMPDIR" -name 'watcher.node' | while read -r node; do
+    python3 "$PATCHER" "$node"
+done
+
+echo "[lmstudio-install] Running bootstrap..."
+LMS_BOOTSTRAP_INSTALL_SH=1 "$TMPDIR/llmster" bootstrap
+
+# Also patch the installed copy in ~/.lmstudio after bootstrap
+echo "[lmstudio-install] Patching installed watcher.node copies..."
+find "${HOME}/.lmstudio" -name 'watcher.node' 2>/dev/null | while read -r node; do
+    python3 "$PATCHER" "$node"
+done
+
+echo "[lmstudio-install] Installation complete."

ansible/roles/lm_studio/files/patch-glibcxx-compat.py

@@ -0,0 +1,147 @@
+#!/usr/bin/env python3
+"""
+Patch ELF .so/.node files that require GLIBCXX_3.4.30 to use GLIBCXX_3.4.29
+instead. Required for AlmaLinux 9 / RHEL 9 which ships GCC 11 (max GLIBCXX_3.4.29).
+
+Both the version string AND the vna_hash in .gnu.version_r are updated so the
+dynamic linker finds the version correctly at runtime.
+"""
+import sys
+import struct
+
+
+def elf_hash(name: str) -> int:
+    """ELF version symbol hash (not GNU hash — used in .gnu.version_r)."""
+    h = 0
+    for c in name:
+        h = ((h << 4) + ord(c)) & 0xFFFFFFFF
+        g = h & 0xF0000000
+        if g:
+            h ^= g >> 24
+        h &= (~g) & 0xFFFFFFFF
+    return h
+
+
+def patch_file(path: str) -> bool:
+    with open(path, 'rb') as f:
+        data = bytearray(f.read())
+
+    old_ver = b'GLIBCXX_3.4.30'
+    new_ver = b'GLIBCXX_3.4.11'
+
+    if old_ver not in data:
+        print(f'  skip (no GLIBCXX_3.4.30): {path}')
+        return False
+
+    # ----- ELF header -----
+    if data[:4] != b'\x7fELF':
+        print(f'  skip (not ELF): {path}')
+        return False
+
+    ei_class = data[4]  # 1=32-bit, 2=64-bit
+    ei_data  = data[5]  # 1=LE, 2=BE
+    endian   = '<' if ei_data == 1 else '>'
+
+    if ei_class == 2:
+        # 64-bit ELF
+        e_shoff_fmt = f'{endian}Q'
+        ehdr_size   = 64
+        shdr_size   = 64
+        shdr_fmt    = f'{endian}IIQQQQIIQQ'  # 64-byte shdr
+        (e_shoff,)  = struct.unpack_from(f'{endian}Q', data, 40)
+        (e_shentsize, e_shnum, e_shstrndx) = struct.unpack_from(f'{endian}HHH', data, 58)
+        shdr_type_off = 4
+        shdr_off_off  = 24   # sh_offset within shdr
+        shdr_link_off = 48   # sh_link within shdr
+    else:
+        # 32-bit ELF
+        (e_shoff,)  = struct.unpack_from(f'{endian}I', data, 32)
+        (e_shentsize, e_shnum, e_shstrndx) = struct.unpack_from(f'{endian}HHH', data, 46)
+        shdr_type_off = 4
+        shdr_off_off  = 16
+        shdr_link_off = 28
+
+    SHT_GNU_verneed = 0x6FFFFFFE
+
+    patched = False
+    # Target GLIBCXX_3.4.11 — that's where condition_variable::wait actually lives
+    # in AlmaLinux 9's libstdc++.so.6 (same byte length: 14 chars each)
+    old_hash = elf_hash('GLIBCXX_3.4.30')
+    new_hash = elf_hash('GLIBCXX_3.4.11')
+    old_hash_bytes = struct.pack(f'{endian}I', old_hash)
+    new_hash_bytes = struct.pack(f'{endian}I', new_hash)
+
+    for i in range(e_shnum):
+        shdr_base = e_shoff + i * e_shentsize
+        sh_type   = struct.unpack_from(f'{endian}I', data, shdr_base + shdr_type_off)[0]
+
+        if sh_type != SHT_GNU_verneed:
+            continue
+
+        # sh_offset and sh_link (index of associated string table section)
+        if ei_class == 2:
+            sh_offset = struct.unpack_from(f'{endian}Q', data, shdr_base + 24)[0]
+            sh_size   = struct.unpack_from(f'{endian}Q', data, shdr_base + 32)[0]
+            sh_link   = struct.unpack_from(f'{endian}I', data, shdr_base + 40)[0]
+            sh_info   = struct.unpack_from(f'{endian}I', data, shdr_base + 44)[0]
+        else:
+            sh_offset = struct.unpack_from(f'{endian}I', data, shdr_base + 16)[0]
+            sh_size   = struct.unpack_from(f'{endian}I', data, shdr_base + 20)[0]
+            sh_link   = struct.unpack_from(f'{endian}I', data, shdr_base + 24)[0]
+            sh_info   = struct.unpack_from(f'{endian}I', data, shdr_base + 28)[0]
+
+        # String table section referenced by sh_link
+        strtab_shdr = e_shoff + sh_link * e_shentsize
+        if ei_class == 2:
+            strtab_off = struct.unpack_from(f'{endian}Q', data, strtab_shdr + 24)[0]
+        else:
+            strtab_off = struct.unpack_from(f'{endian}I', data, strtab_shdr + 16)[0]
+
+        # Walk Elf64_Verneed / Elf64_Vernaux entries
+        # Elf64_Verneed: vn_version(2)+vn_cnt(2)+vn_file(4)+vn_aux(4)+vn_next(4) = 16 bytes
+        # Elf64_Vernaux: vna_hash(4)+vna_flags(2)+vna_other(2)+vna_name(4)+vna_next(4) = 16 bytes
+        vneed_pos = sh_offset
+        for _ in range(sh_info):  # sh_info = number of Verneed entries
+            vn_cnt  = struct.unpack_from(f'{endian}H', data, vneed_pos + 2)[0]
+            vn_aux  = struct.unpack_from(f'{endian}I', data, vneed_pos + 8)[0]   # +8 (after vn_file)
+            vn_next = struct.unpack_from(f'{endian}I', data, vneed_pos + 12)[0]
+
+            vaux_pos = vneed_pos + vn_aux
+            for _ in range(vn_cnt):
+                vna_hash  = struct.unpack_from(f'{endian}I', data, vaux_pos)[0]
+                vna_name  = struct.unpack_from(f'{endian}I', data, vaux_pos + 8)[0]
+                vna_next  = struct.unpack_from(f'{endian}I', data, vaux_pos + 12)[0]
+
+                # Read the version name from the string table
+                name_off  = strtab_off + vna_name
+                name_end  = data.index(b'\x00', name_off)
+                ver_name  = bytes(data[name_off:name_end])
+
+                if ver_name == old_ver:
+                    # Patch hash
+                    struct.pack_into(f'{endian}I', data, vaux_pos, new_hash)
+                    # Patch string (same byte length — safe in-place)
+                    data[name_off:name_off + len(old_ver)] = new_ver
+                    patched = True
+
+                if vna_next == 0:
+                    break
+                vaux_pos += vna_next
+
+            if vn_next == 0:
+                break
+            vneed_pos += vn_next
+
+    if patched:
+        with open(path, 'wb') as f:
+            f.write(data)
+        print(f'  patched (hash+string): {path}')
+        return True
+    else:
+        print(f'  skip (GLIBCXX_3.4.30 not in verneed): {path}')
+        return False
+
+
+if __name__ == '__main__':
+    for path in sys.argv[1:]:
+        patch_file(path)

ansible/roles/lm_studio/files/run-llmster.sh

@@ -0,0 +1,45 @@
+#!/bin/bash
+# LM Studio service wrapper.
+#
+# Lets lms server start handle daemon launch (passkey setup, etc.),
+# optionally auto-loads a model (LMS_MODEL + LMS_CONTEXT_LENGTH),
+# then monitors the llmster process and exits when it dies so systemd
+# can restart the service.
+set -e
+
+LMS="${HOME}/.lmstudio/bin/lms"
+LMS_PORT="${LMS_PORT:-1234}"
+LMS_BIND="${LMS_BIND:-0.0.0.0}"
+LMS_MODEL="${LMS_MODEL:-}"
+LMS_CONTEXT_LENGTH="${LMS_CONTEXT_LENGTH:-65536}"
+LMS_GPU_OFFLOAD="${LMS_GPU_OFFLOAD:-max}"
+
+echo "Starting LM Studio server (port ${LMS_PORT}, bind ${LMS_BIND})..."
+"$LMS" server start --port "$LMS_PORT" --bind "$LMS_BIND"
+
+if [[ -n "$LMS_MODEL" ]]; then
+    echo "Unloading any previously loaded models..."
+    "$LMS" unload --all 2>/dev/null || true
+    echo "Auto-loading model ${LMS_MODEL} (context ${LMS_CONTEXT_LENGTH}, gpu ${LMS_GPU_OFFLOAD})..."
+    "$LMS" load "$LMS_MODEL" --context-length "$LMS_CONTEXT_LENGTH" --gpu "$LMS_GPU_OFFLOAD"
+fi
+
+# llmster bind sur 127.0.0.1 après chargement du backend Vulkan.
+# socat proxy sur l'IP LAN (non-conflictuel avec le loopback) → llmster.
+if [[ "$LMS_BIND" != "127.0.0.1" ]]; then
+    LAN_IP=$(ip -4 addr show scope global | grep -oP '(?<=inet\s)\d+(\.\d+){3}' | head -1)
+    if [[ -n "$LAN_IP" ]]; then
+        echo "Starting socat proxy ${LAN_IP}:${LMS_PORT} → 127.0.0.1:${LMS_PORT}..."
+        socat TCP-LISTEN:"${LMS_PORT}",fork,bind="${LAN_IP}",reuseaddr TCP:127.0.0.1:"${LMS_PORT}" &
+        SOCAT_PID=$!
+        echo "socat PID: ${SOCAT_PID}"
+    fi
+fi
+
+echo "Server started. Monitoring llmster daemon..."
+while pgrep -u "$(id -un)" -x llmster > /dev/null 2>&1; do
+    sleep 15
+done
+
+echo "llmster daemon stopped unexpectedly, exiting..."
+exit 1

ansible/roles/lm_studio/handlers/main.yml

@@ -0,0 +1,9 @@
+---
+- name: Reload systemd
+  ansible.builtin.systemd:
+    daemon_reload: true
+
+- name: Restart lm-studio
+  ansible.builtin.systemd:
+    name: lm-studio
+    state: restarted

ansible/roles/lm_studio/tasks/main.yml

@@ -0,0 +1,144 @@
+---
+- name: Install LM Studio dependencies
+  ansible.builtin.dnf:
+    name:
+      - fuse
+      - fuse-libs
+      - curl
+      - libatomic
+      - python3
+      - mesa-vulkan-drivers
+      - vulkan-tools
+      - socat
+    state: present
+
+- name: Create lmstudio system user
+  ansible.builtin.user:
+    name: "{{ lm_studio_user }}"
+    home: "{{ lm_studio_home }}"
+    shell: /bin/bash
+    groups:
+      - video
+      - render
+    append: true
+    system: false
+    create_home: true
+    state: present
+
+- name: Check if lms CLI is already installed
+  ansible.builtin.stat:
+    path: "{{ lm_studio_home }}/.lmstudio/bin/lms"
+  register: lms_binary
+
+- name: Copy patch-glibcxx-compat.py
+  ansible.builtin.copy:
+    src: patch-glibcxx-compat.py
+    dest: /tmp/patch-glibcxx-compat.py
+    mode: '0755'
+
+- name: Copy LM Studio installer files
+  ansible.builtin.copy:
+    src: "{{ item.src }}"
+    dest: "{{ item.dest }}"
+    mode: "{{ item.mode }}"
+  loop:
+    - {src: install-lmstudio-compat.sh, dest: /tmp/install-lmstudio-compat.sh, mode: '0755'}
+  when: not lms_binary.stat.exists
+
+- name: Run LM Studio installer
+  ansible.builtin.shell:
+    cmd: /tmp/install-lmstudio-compat.sh
+    executable: /bin/bash
+  become: true
+  become_user: "{{ lm_studio_user }}"
+  environment:
+    HOME: "{{ lm_studio_home }}"
+  when: not lms_binary.stat.exists
+  register: lms_install_result
+
+- name: Show installer output
+  ansible.builtin.debug:
+    var: lms_install_result.stdout_lines
+  when: lms_install_result is changed
+
+- name: Deploy llmster wrapper script
+  ansible.builtin.copy:
+    src: run-llmster.sh
+    dest: /opt/lmstudio/run-llmster.sh
+    mode: '0755'
+
+- name: Patch all GLIBCXX_3.4.30-dependent binaries (watcher.node, libllm_engine.so, ...)
+  ansible.builtin.shell:
+    cmd: |
+      find "{{ lm_studio_home }}/.lmstudio" \( -name '*.node' -o -name 'lib*.so' \) \
+        -exec python3 /tmp/patch-glibcxx-compat.py {} \;
+    executable: /bin/bash
+  register: patch_result
+  changed_when: "'patched' in patch_result.stdout"
+
+- name: Show patch output
+  ansible.builtin.debug:
+    var: patch_result.stdout_lines
+  when: patch_result is changed
+
+- name: Force Vulkan backend (AMD GPU — CPU backend selected by default)
+  ansible.builtin.copy:
+    dest: "{{ lm_studio_home }}/.lmstudio/.internal/backend-preferences-v1.json"
+    owner: "{{ lm_studio_user }}"
+    group: "{{ lm_studio_user }}"
+    mode: '0644'
+    content: |
+      [
+        {
+          "model_format": "gguf",
+          "name": "llama.cpp-linux-x86_64-vulkan-avx2",
+          "version": "2.13.0"
+        }
+      ]
+  notify:
+    - Reload systemd
+    - Restart lm-studio
+
+- name: Deploy LM Studio systemd service
+  ansible.builtin.template:
+    src: lm-studio.service.j2
+    dest: /etc/systemd/system/lm-studio.service
+    mode: '0644'
+  notify:
+    - Reload systemd
+    - Restart lm-studio
+
+- name: Enable lm-studio service
+  ansible.builtin.systemd:
+    name: lm-studio
+    enabled: true
+    daemon_reload: true
+    state: started
+
+- name: Redirect models dir vers NFS (symlink)
+  ansible.builtin.file:
+    src: "{{ lm_studio_models_nfs_path }}"
+    dest: "{{ lm_studio_home }}/.lmstudio/models"
+    state: link
+    force: true
+    owner: "{{ lm_studio_user }}"
+    group: "{{ lm_studio_user }}"
+  when: lm_studio_models_nfs_path is defined and lm_studio_models_nfs_path != ""
+
+- name: Deploy lms global wrapper (/usr/local/bin/lms)
+  ansible.builtin.copy:
+    dest: /usr/local/bin/lms
+    mode: '0755'
+    content: |
+      #!/bin/bash
+      exec sudo -u {{ lm_studio_user }} \
+        HOME={{ lm_studio_home }} \
+        HSA_OVERRIDE_GFX_VERSION={{ hsa_override_gfx_version }} \
+        {{ lm_studio_home }}/.lmstudio/bin/lms "$@"
+
+- name: Open LM Studio port in firewall
+  ansible.posix.firewalld:
+    port: "{{ lm_studio_port }}/tcp"
+    permanent: true
+    state: enabled
+    immediate: true

ansible/roles/lm_studio/templates/lm-studio.service.j2

@@ -0,0 +1,39 @@
+[Unit]
+Description=LM Studio Server
+After=network.target
+Wants=network.target
+
+[Service]
+Type=simple
+User={{ lm_studio_user }}
+Group={{ lm_studio_user }}
+
+# Répertoire home pour que lms trouve ses fichiers
+Environment=HOME={{ lm_studio_home }}
+
+# ROCm — obligatoire pour RX 6700XT (gfx1031 non officiel)
+Environment=HSA_OVERRIDE_GFX_VERSION={{ hsa_override_gfx_version }}
+
+# PATH inclut le binaire lms
+Environment=PATH={{ lm_studio_home }}/.lmstudio/bin:/usr/local/bin:/usr/bin:/bin
+
+Environment=LMS_PORT={{ lm_studio_port }}
+Environment=LMS_BIND={{ lm_studio_host }}
+Environment=LMS_MODEL={{ lm_studio_default_model }}
+Environment=LMS_CONTEXT_LENGTH={{ lm_studio_context_length }}
+Environment=LMS_GPU_OFFLOAD={{ lm_studio_gpu_offload }}
+
+# Wrapper: runs lms server start, auto-loads model if LMS_MODEL is set, then monitors llmster
+ExecStart=/opt/lmstudio/run-llmster.sh
+
+Restart=on-failure
+RestartSec=15
+TimeoutStartSec=120
+
+# Journalisation
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=lm-studio
+
+[Install]
+WantedBy=multi-user.target

ansible/roles/minio/tasks/main.yml

@@ -0,0 +1 @@
+---

ansible/roles/nfs_client/defaults/main.yml

@@ -0,0 +1,7 @@
+---
+nfs_server_ip: 192.168.10.1
+nfs_server_export: /srv/data/nfs/k8s
+
+nfs_client_mounts:
+  - local_path: /mnt/nfs
+    remote_export: "{{ nfs_server_export }}"

ansible/roles/nfs_client/handlers/main.yml

@@ -0,0 +1,5 @@
+---
+- name: Reload systemd and activate automounts
+  ansible.builtin.systemd:
+    daemon_reload: true
+  listen: Reload systemd and activate automounts

ansible/roles/nfs_client/tasks/main.yml

@@ -0,0 +1,42 @@
+---
+- name: Install nfs-utils
+  ansible.builtin.dnf:
+    name: nfs-utils
+    state: present
+
+- name: Create NFS mount points
+  ansible.builtin.file:
+    path: "{{ item.local_path }}"
+    state: directory
+    mode: '0755'
+  loop: "{{ nfs_client_mounts }}"
+
+- name: Write NFS entries in fstab (sans monter immédiatement)
+  ansible.posix.mount:
+    src: "{{ nfs_server_ip }}:{{ item.remote_export }}"
+    path: "{{ item.local_path }}"
+    fstype: nfs
+    # Options pour extinctions fréquentes :
+    #   soft                         → échoue proprement (EIO) si serveur injoignable
+    #   timeo=30 / retrans=3         → 3s timeout, 3 essais avant EIO
+    #   _netdev                      → montage après réseau disponible seulement
+    #   nofail                       → n'empêche pas le boot si NFS absent
+    #   x-systemd.automount          → monté à la demande (premier accès), pas au boot
+    #   x-systemd.mount-timeout=30   → 30s max pour réussir le montage à la demande
+    opts: "nfsvers=4,soft,timeo=30,retrans=3,_netdev,nofail,x-systemd.automount,x-systemd.mount-timeout=30"
+    state: present
+    dump: 0
+    passno: 0
+  loop: "{{ nfs_client_mounts }}"
+  notify: Reload systemd and activate automounts
+
+- name: Reload systemd et activer les automounts NFS
+  ansible.builtin.systemd:
+    daemon_reload: true
+
+- name: Enable and start NFS automount units
+  ansible.builtin.systemd:
+    name: "{{ item.local_path | replace('/', '-') | regex_replace('^-', '') }}.automount"
+    enabled: true
+    state: started
+  loop: "{{ nfs_client_mounts }}"

ansible/roles/nfs_server/defaults/main.yml

@@ -0,0 +1,10 @@
+---
+raid_device: /dev/md127
+raid_uuid: "3add8360-fa01-47eb-8aa1-0e84bacbbc15"
+raid_fstype: ext4
+raid_mount: /srv/data
+
+nfs_exports:
+  - path: /srv/data/nfs/k8s
+    network: "{{ cluster_network }}"
+    options: "rw,sync,no_subtree_check,no_root_squash"

ansible/roles/nfs_server/handlers/main.yml

@@ -0,0 +1,14 @@
+---
+- name: Rebuild initramfs
+  ansible.builtin.command: dracut --force
+  listen: Rebuild initramfs
+
+- name: Reload exports
+  ansible.builtin.command: exportfs -ra
+  listen: Reload exports
+
+- name: Restart nfs-server
+  ansible.builtin.systemd:
+    name: nfs-server
+    state: restarted
+  listen: Restart nfs-server

ansible/roles/nfs_server/tasks/main.yml

@@ -0,0 +1,83 @@
+---
+- name: Install nfs-utils and mdadm
+  ansible.builtin.dnf:
+    name:
+      - nfs-utils
+      - mdadm
+    state: present
+
+# --- RAID persistence ---
+
+- name: Persist mdadm array configuration
+  ansible.builtin.shell:
+    cmd: mdadm --detail --scan
+  register: mdadm_scan
+  changed_when: false
+
+- name: Write /etc/mdadm.conf
+  ansible.builtin.copy:
+    content: "{{ mdadm_scan.stdout }}\n"
+    dest: /etc/mdadm.conf
+    mode: '0644'
+  notify: Rebuild initramfs
+
+# --- RAID mount ---
+
+- name: Create RAID mount point
+  ansible.builtin.file:
+    path: "{{ raid_mount }}"
+    state: directory
+    mode: '0755'
+
+- name: Mount RAID5 array (fstab + immediate)
+  ansible.posix.mount:
+    src: "UUID={{ raid_uuid }}"
+    path: "{{ raid_mount }}"
+    fstype: "{{ raid_fstype }}"
+    opts: defaults,noatime,nodiratime
+    state: mounted
+    dump: 0
+    passno: 2
+
+# --- Répertoires NFS ---
+
+- name: Create NFS export directories
+  ansible.builtin.file:
+    path: "{{ item.path }}"
+    state: directory
+    mode: '0755'
+    owner: nobody
+    group: nobody
+  loop: "{{ nfs_exports }}"
+
+# --- NFS server ---
+
+- name: Deploy /etc/exports
+  ansible.builtin.template:
+    src: exports.j2
+    dest: /etc/exports
+    mode: '0644'
+  notify:
+    - Reload exports
+    - Restart nfs-server
+
+- name: Enable and start nfs-server
+  ansible.builtin.systemd:
+    name: nfs-server
+    enabled: true
+    state: started
+
+# --- Firewall ---
+
+- name: Allow NFS through firewall (firewalld uniquement, pas nftables)
+  ansible.posix.firewalld:
+    service: "{{ item }}"
+    permanent: true
+    state: enabled
+    immediate: true
+  loop:
+    - nfs
+    - mountd
+    - rpc-bind
+  when: ansible_facts.services['firewalld.service'] is defined
+        and ansible_facts.services['firewalld.service']['state'] == 'running'

ansible/roles/nfs_server/templates/exports.j2

@@ -0,0 +1,4 @@
+# /etc/exports — géré par Ansible, ne pas éditer manuellement
+{% for export in nfs_exports %}
+{{ export.path }}  {{ export.network }}({{ export.options }})
+{% endfor %}

ansible/roles/postgresql/defaults/main.yml

@@ -0,0 +1,24 @@
+---
+postgresql_version: 16
+postgresql_port: 5432
+postgresql_data_dir: /srv/data/postgres
+
+# Écoute sur localhost + LAN cluster (pour les services k8s futurs)
+postgresql_listen_addresses: "localhost, 192.168.10.1"
+
+postgresql_max_connections: 100
+postgresql_shared_buffers: "256MB"
+
+# Bases et utilisateurs à créer
+# Mots de passe dans le vault : vault_pg_hermes_password, vault_pg_litellm_password
+postgresql_databases:
+  - name: hermes
+  - name: litellm
+
+postgresql_users:
+  - name: hermes
+    password: "{{ vault_pg_hermes_password }}"
+    db: hermes
+  - name: litellm
+    password: "{{ vault_pg_litellm_password }}"
+    db: litellm

ansible/roles/postgresql/handlers/main.yml

@@ -0,0 +1,10 @@
+---
+- name: Restart postgresql
+  ansible.builtin.systemd:
+    name: "postgresql-{{ postgresql_version }}"
+    state: restarted
+
+- name: Reload postgresql
+  ansible.builtin.systemd:
+    name: "postgresql-{{ postgresql_version }}"
+    state: reloaded

ansible/roles/postgresql/tasks/main.yml

@@ -0,0 +1,142 @@
+---
+# --- Dépôt officiel PostgreSQL ---
+
+- name: Install PostgreSQL repo
+  ansible.builtin.dnf:
+    name: "https://download.postgresql.org/pub/repos/yum/reporpms/EL-9-x86_64/pgdg-redhat-repo-latest.noarch.rpm"
+    state: present
+    disable_gpg_check: true
+
+- name: Disable built-in postgresql module
+  ansible.builtin.command:
+    cmd: dnf -qy module disable postgresql
+  changed_when: false
+
+# --- Installation ---
+
+- name: Install PostgreSQL {{ postgresql_version }}
+  ansible.builtin.dnf:
+    name:
+      - "postgresql{{ postgresql_version }}-server"
+      - "postgresql{{ postgresql_version }}"
+      - python3-psycopg2
+    state: present
+
+# --- Répertoire données sur RAID5 ---
+
+- name: Create PostgreSQL data directory on RAID5
+  ansible.builtin.file:
+    path: "{{ postgresql_data_dir }}"
+    state: directory
+    owner: postgres
+    group: postgres
+    mode: '0700'
+
+# --- Service override pour PGDATA custom ---
+
+- name: Create systemd override directory
+  ansible.builtin.file:
+    path: "/etc/systemd/system/postgresql-{{ postgresql_version }}.service.d"
+    state: directory
+    mode: '0755'
+
+- name: Deploy PGDATA override
+  ansible.builtin.copy:
+    dest: "/etc/systemd/system/postgresql-{{ postgresql_version }}.service.d/pgdata.conf"
+    content: |
+      [Service]
+      Environment=PGDATA={{ postgresql_data_dir }}
+    mode: '0644'
+  notify: Restart postgresql
+
+# --- Initialisation (une seule fois) ---
+
+- name: Check if database cluster is initialized
+  ansible.builtin.stat:
+    path: "{{ postgresql_data_dir }}/PG_VERSION"
+  register: pg_initialized
+
+- name: Initialize database cluster
+  ansible.builtin.command:
+    cmd: "/usr/pgsql-{{ postgresql_version }}/bin/initdb -D {{ postgresql_data_dir }} --auth-local=peer --auth-host=scram-sha-256"
+  become: true
+  become_user: postgres
+  environment:
+    PGDATA: "{{ postgresql_data_dir }}"
+  when: not pg_initialized.stat.exists
+
+# --- Configuration ---
+
+- name: Deploy pg_hba.conf
+  ansible.builtin.template:
+    src: pg_hba.conf.j2
+    dest: "{{ postgresql_data_dir }}/pg_hba.conf"
+    owner: postgres
+    group: postgres
+    mode: '0600'
+  notify: Reload postgresql
+
+- name: Configure listen_addresses
+  ansible.builtin.lineinfile:
+    path: "{{ postgresql_data_dir }}/postgresql.conf"
+    regexp: "^#?listen_addresses"
+    line: "listen_addresses = '{{ postgresql_listen_addresses }}'"
+  notify: Reload postgresql
+
+- name: Configure port
+  ansible.builtin.lineinfile:
+    path: "{{ postgresql_data_dir }}/postgresql.conf"
+    regexp: "^#?port"
+    line: "port = {{ postgresql_port }}"
+  notify: Reload postgresql
+
+- name: Configure shared_buffers
+  ansible.builtin.lineinfile:
+    path: "{{ postgresql_data_dir }}/postgresql.conf"
+    regexp: "^#?shared_buffers"
+    line: "shared_buffers = {{ postgresql_shared_buffers }}"
+  notify: Reload postgresql
+
+# --- Démarrage ---
+
+- name: Enable and start PostgreSQL
+  ansible.builtin.systemd:
+    name: "postgresql-{{ postgresql_version }}"
+    enabled: true
+    daemon_reload: true
+    state: started
+
+# --- Bases de données et utilisateurs ---
+
+- name: Create databases
+  community.postgresql.postgresql_db:
+    name: "{{ item.name }}"
+    encoding: UTF8
+    state: present
+  become: true
+  become_user: postgres
+  loop: "{{ postgresql_databases }}"
+
+- name: Create users
+  community.postgresql.postgresql_user:
+    name: "{{ item.name }}"
+    password: "{{ item.password }}"
+    state: present
+  become: true
+  become_user: postgres
+  loop: "{{ postgresql_users }}"
+  no_log: true
+
+- name: Grant user privileges on their database
+  community.postgresql.postgresql_privs:
+    db: "{{ item.db }}"
+    privs: ALL
+    type: database
+    role: "{{ item.name }}"
+    state: present
+  become: true
+  become_user: postgres
+  loop: "{{ postgresql_users }}"
+
+# Firewall géré centralement par le rôle gateway (nftables)
+# Port 5432 ouvert pour cluster_network dans nftables.conf.j2

ansible/roles/postgresql/templates/pg_hba.conf.j2

@@ -0,0 +1,9 @@
+# TYPE  DATABASE        USER            ADDRESS                 METHOD
+# Accès local superuser
+local   all             postgres                                peer
+# Accès local apps — scram-sha-256
+local   all             all                                     scram-sha-256
+host    all             all             127.0.0.1/32            scram-sha-256
+host    all             all             ::1/128                 scram-sha-256
+# Accès depuis le LAN cluster (services k8s futurs)
+host    all             all             {{ cluster_network }}   scram-sha-256

ansible/roles/qdrant/defaults/main.yml

@@ -0,0 +1,13 @@
+---
+qdrant_version: "1.13.4"
+qdrant_user: qdrant
+qdrant_data_dir: /srv/data/qdrant
+qdrant_install_dir: /opt/qdrant
+
+qdrant_http_port: 6333
+qdrant_grpc_port: 6334
+
+# 0.0.0.0 — accessible depuis le LAN cluster (firewall filtre)
+qdrant_host: "0.0.0.0"
+
+qdrant_log_level: INFO

ansible/roles/qdrant/handlers/main.yml

@@ -0,0 +1,5 @@
+---
+- name: Restart qdrant
+  ansible.builtin.systemd:
+    name: qdrant
+    state: restarted

ansible/roles/qdrant/tasks/main.yml

@@ -0,0 +1,92 @@
+---
+# --- Utilisateur système ---
+
+- name: Create qdrant system user
+  ansible.builtin.user:
+    name: "{{ qdrant_user }}"
+    system: true
+    shell: /sbin/nologin
+    home: "{{ qdrant_install_dir }}"
+    create_home: true
+    state: present
+
+# --- Répertoires données (RAID5) ---
+
+- name: Create Qdrant data directories on RAID5
+  ansible.builtin.file:
+    path: "{{ item }}"
+    state: directory
+    owner: "{{ qdrant_user }}"
+    group: "{{ qdrant_user }}"
+    mode: '0750'
+  loop:
+    - "{{ qdrant_data_dir }}"
+    - "{{ qdrant_data_dir }}/storage"
+    - "{{ qdrant_data_dir }}/snapshots"
+
+# --- Binaire ---
+
+- name: Check if qdrant binary is already installed
+  ansible.builtin.stat:
+    path: "{{ qdrant_install_dir }}/qdrant"
+  register: qdrant_binary
+
+- name: Check installed qdrant version
+  ansible.builtin.command:
+    cmd: "{{ qdrant_install_dir }}/qdrant --version"
+  register: qdrant_current_version
+  changed_when: false
+  failed_when: false
+  when: qdrant_binary.stat.exists
+
+- name: Download qdrant binary
+  ansible.builtin.get_url:
+    url: "https://github.com/qdrant/qdrant/releases/download/v{{ qdrant_version }}/qdrant-x86_64-unknown-linux-musl.tar.gz"
+    dest: /tmp/qdrant.tar.gz
+    mode: '0644'
+  when: >
+    not qdrant_binary.stat.exists or
+    qdrant_version not in (qdrant_current_version.stdout | default(''))
+
+- name: Extract qdrant binary
+  ansible.builtin.unarchive:
+    src: /tmp/qdrant.tar.gz
+    dest: "{{ qdrant_install_dir }}"
+    remote_src: true
+    owner: "{{ qdrant_user }}"
+    group: "{{ qdrant_user }}"
+  when: >
+    not qdrant_binary.stat.exists or
+    qdrant_version not in (qdrant_current_version.stdout | default(''))
+  notify: Restart qdrant
+
+# --- Configuration ---
+
+- name: Deploy Qdrant config
+  ansible.builtin.template:
+    src: config.yaml.j2
+    dest: "{{ qdrant_install_dir }}/config.yaml"
+    owner: "{{ qdrant_user }}"
+    group: "{{ qdrant_user }}"
+    mode: '0640'
+  notify: Restart qdrant
+
+# --- Service systemd ---
+
+- name: Deploy Qdrant systemd service
+  ansible.builtin.template:
+    src: qdrant.service.j2
+    dest: /etc/systemd/system/qdrant.service
+    mode: '0644'
+  notify:
+    - Restart qdrant
+
+- name: Enable and start Qdrant
+  ansible.builtin.systemd:
+    name: qdrant
+    enabled: true
+    daemon_reload: true
+    state: started
+
+# Firewall géré centralement par le rôle gateway (nftables)
+# Ports 6333 (HTTP) et 6334 (gRPC) ouverts pour cluster_network dans nftables.conf.j2

ansible/roles/qdrant/templates/config.yaml.j2

@@ -0,0 +1,11 @@
+storage:
+  storage_path: {{ qdrant_data_dir }}/storage
+  snapshots_path: {{ qdrant_data_dir }}/snapshots
+
+service:
+  host: {{ qdrant_host }}
+  http_port: {{ qdrant_http_port }}
+  grpc_port: {{ qdrant_grpc_port }}
+  enable_cors: true
+
+log_level: {{ qdrant_log_level }}

ansible/roles/qdrant/templates/qdrant.service.j2

@@ -0,0 +1,18 @@
+[Unit]
+Description=Qdrant — vector database
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=simple
+User={{ qdrant_user }}
+Group={{ qdrant_user }}
+ExecStart={{ qdrant_install_dir }}/qdrant --config-path {{ qdrant_install_dir }}/config.yaml
+Restart=on-failure
+RestartSec=10
+StandardOutput=journal
+StandardError=journal
+SyslogIdentifier=qdrant
+
+[Install]
+WantedBy=multi-user.target

ansible/roles/rocm/defaults/main.yml

@@ -0,0 +1,25 @@
+---
+rocm_rhel_version: "9.4"
+rocm_amdgpu_repo: "https://repo.radeon.com/amdgpu/latest/rhel/{{ rocm_rhel_version }}/main/x86_64/"
+rocm_repo: "https://repo.radeon.com/rocm/rhel9/latest/main/"
+rocm_gpg_key: "https://repo.radeon.com/rocm/rocm.gpg.key"
+
+rocm_packages:
+  - rocminfo
+  - rocm-hip-runtime
+  - rocm-opencl-runtime
+  - rocblas
+  - hipblaslt
+
+# Packages supplémentaires requis pour compiler llama.cpp from source (rôle llama_server)
+rocm_build_packages:
+  - hipblas-devel
+  - rocblas-devel
+  - hip-devel
+  - hipcc
+  - cmake
+  - gcc-c++
+  - git
+  - patchelf
+
+hsa_override_gfx_version: "10.3.0"

ansible/roles/rocm/handlers/main.yml

@@ -0,0 +1,6 @@
+---
+- name: Reload amdgpu module
+  ansible.builtin.shell: |
+    modprobe -r amdgpu || true
+    modprobe amdgpu
+  changed_when: true

ansible/roles/rocm/tasks/main.yml

@@ -0,0 +1,60 @@
+---
+- name: Import AMD GPG key
+  ansible.builtin.rpm_key:
+    key: "{{ rocm_gpg_key }}"
+    state: present
+
+- name: Add amdgpu repository
+  ansible.builtin.yum_repository:
+    name: amdgpu
+    description: AMD GPU drivers
+    baseurl: "{{ rocm_amdgpu_repo }}"
+    enabled: true
+    gpgcheck: true
+    gpgkey: "{{ rocm_gpg_key }}"
+
+- name: Add ROCm repository
+  ansible.builtin.yum_repository:
+    name: rocm
+    description: ROCm
+    baseurl: "{{ rocm_repo }}"
+    enabled: true
+    gpgcheck: true
+    gpgkey: "{{ rocm_gpg_key }}"
+
+- name: Install kernel headers and devel (requis pour amdgpu-dkms)
+  ansible.builtin.dnf:
+    name:
+      - kernel-headers
+      - kernel-devel
+    state: present
+
+- name: Install ROCm packages
+  ansible.builtin.dnf:
+    name: "{{ rocm_packages }}"
+    state: present
+  notify: Reload amdgpu module
+
+- name: Add ansible user to video and render groups
+  ansible.builtin.user:
+    name: "{{ ansible_service_user }}"
+    groups:
+      - video
+      - render
+    append: true
+
+- name: Set HSA_OVERRIDE_GFX_VERSION system-wide
+  ansible.builtin.lineinfile:
+    path: /etc/environment
+    regexp: '^HSA_OVERRIDE_GFX_VERSION='
+    line: "HSA_OVERRIDE_GFX_VERSION={{ hsa_override_gfx_version }}"
+    create: true
+    mode: '0644'
+
+- name: Set HSA_OVERRIDE_GFX_VERSION dans profile.d (sessions interactives)
+  ansible.builtin.copy:
+    dest: /etc/profile.d/rocm.sh
+    content: |
+      export HSA_OVERRIDE_GFX_VERSION={{ hsa_override_gfx_version }}
+      export PATH=$PATH:/opt/rocm/bin
+    mode: '0644'