Files
rick-infra/roles/valkey/templates/valkey.service.j2
Joakim 3506e55016 Migrate to rootful container architecture with infrastructure fact pattern
Major architectural change from rootless user services to system-level (rootful)
containers to enable group-based Unix socket access for containerized applications.

Infrastructure Changes:
- PostgreSQL: Export postgres-clients group GID as Ansible fact
- Valkey: Export valkey-clients group GID as Ansible fact
- Valkey: Add socket-fix service to maintain correct socket group ownership
- Both: Set socket directories to 770 with client group ownership

Authentik Role Refactoring:
- Remove rootless container configuration (subuid/subgid, lingering, user systemd)
- Deploy Quadlet files to /etc/containers/systemd/ (system-level)
- Use dynamic GID facts in container PodmanArgs (--group-add)
- Simplify user creation to system user with infrastructure group membership
- Update handlers for system scope service management
- Remove unnecessary container security options (no user namespace isolation)

Container Template Changes:
- Pod: Remove --userns args, change WantedBy to multi-user.target
- Containers: Replace Annotation with PodmanArgs using dynamic GIDs
- Remove /dev/shm mounts and SecurityLabelDisable (not needed for rootful)
- Change WantedBy to multi-user.target for system services

Documentation Updates:
- Add ADR-005: Rootful Containers with Infrastructure Fact Pattern
- Update ADR-003: Podman + systemd for system-level deployment
- Update authentik-deployment-guide.md for system scope commands
- Update service-integration-guide.md with rootful pattern examples
- Document discarded rootless approach and rationale

Why Rootful Succeeds:
- Direct UID/GID mapping preserves supplementary groups
- Container process groups match host socket group ownership
- No user namespace remapping breaking permissions

Why Rootless Failed (Discarded):
- User namespace UID/GID remapping broke group-based socket access
- Supplementary groups remapped into subgid range didn't match socket ownership
- Even with --userns=host and keep_original_groups, permissions failed

Pattern Established:
- Infrastructure roles create client groups and export GID facts
- Application roles validate facts and consume in container templates
- Rootful containers run as dedicated users with --group-add for socket access
- System-level deployment provides standard systemd service management

Deployment Validated:
- Services in /system.slice/ ✓
- Process groups: 961 (valkey-clients), 962 (postgres-clients), 966 (authentik) ✓
- Socket permissions: 770 with client groups ✓
- HTTP endpoint responding ✓
2025-12-14 16:56:50 +01:00

81 lines
1.7 KiB
Django/Jinja

# Valkey Systemd Service
# Generated by rick-infra Valkey role
#
# This service provides a secure, hardened Valkey instance with proper configuration loading
[Unit]
Description=Valkey (Redis-compatible) Key-Value Store
Documentation=https://valkey.io/
After=network.target
Wants=network-online.target
{% if valkey_unix_socket_enabled and valkey_client_group_create %}
Wants=valkey-socket-fix.service
{% endif %}
[Service]
Type=notify
User=valkey
Group=valkey
# Core service configuration - ensures config file is loaded
ExecStart=/usr/bin/valkey-server /etc/valkey/valkey.conf --supervised systemd
ExecReload=/bin/kill -USR2 $MAINPID
# Restart configuration
Restart=on-failure
RestartSec=5s
TimeoutStartSec=60
TimeoutStopSec=60
# Runtime directory
RuntimeDirectory=valkey
RuntimeDirectoryMode=755
# Resource limits
LimitNOFILE=10032
# Security hardening
NoNewPrivileges=yes
PrivateTmp=yes
PrivateDevices=yes
ProtectSystem=strict
ProtectHome=yes
ProtectKernelTunables=yes
ProtectKernelModules=yes
ProtectControlGroups=yes
RestrictRealtime=yes
RestrictSUIDSGID=yes
# Network security
RestrictAddressFamilies=AF_INET AF_INET6 AF_UNIX
# Filesystem permissions
ReadWritePaths=/var/lib/valkey
ReadOnlyPaths=/etc/valkey
# System call filtering
SystemCallFilter=@system-service
SystemCallFilter=~@privileged @resources @obsolete
# Memory and resource limits
MemoryDenyWriteExecute=yes
LockPersonality=yes
# Capabilities (remove all unnecessary capabilities)
CapabilityBoundingSet=
AmbientCapabilities=
# Process isolation
PrivateUsers=yes
RemoveIPC=yes
# Additional security
UMask=0027
# Ensure service stops cleanly
KillMode=mixed
KillSignal=SIGTERM
[Install]
WantedBy=multi-user.target
Alias=redis.service