Madagascar
/
CHANGELOG.md
1 contributor
# Madagascar Cluster Changelog All notable changes to the Madagascar cluster configuration and infrastructure are documented in this file. The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/). Each entry should reference related issues using the format `[ISSUE-YYYY-NNN]`. --- ## [Unreleased] ### Known Issues - [ISSUE-2025-001] Thunderbolt interfaces MTU resets to 1500 after networking restart (open) - PBS instances hosted inside VMs `301 is-anjohibe` and `302 is-andrafiabe` become temporarily unavailable during planned node reboot while `pgs suspend` freezes them; this is an accepted cluster-wide maintenance-window limitation that affects backup and restore availability only for that window ### Added - Added a central `cluster/projects/README.md` policy for current and future cluster-level projects ### Changed - Consolidated `pve-net-hang-watchdog` into its own project folder under `cluster/projects/pve-net-hang-watchdog` - Standardized project rules around well-known install paths, mandatory uninstall scripts, and uninstall-before-reinstall workflow - Anchored the central project policy in the existing `autoNAS` install/uninstall workflow and documented its known lessons and current path exception - Established `/usr/local/lib/xdev/<project-name>/uninstall.sh` as the canonical uninstall script location, with optional `/usr/local/sbin/xdev-<project-name>-uninstall` wrapper - Added standard namespaced locations for installed documentation, configuration, operational data, cache, and optional file-based logs - Removed the accidental empty `autoNAS/autoSMART` nested drop and kept `cluster/projects/autoSMART` as the canonical project location - Standardized `cluster/projects/pve-guests-state` with dedicated install/uninstall scripts, namespaced host paths, migrated state location, and cleaned legacy project artifacts - Standardized `cluster/projects/pve-net-hang-watchdog` with namespaced install paths, dedicated lifecycle scripts, and a defaults file under `/etc/default/xdev-pve-net-hang-watchdog` - Updated `pve-net-hang-watchdog` install behavior so deployment also starts the service immediately, not just enables it for boot - Added a standardized shared-runtime lifecycle for `cluster/projects/thunderbolts` that leaves network interface files untouched during reinstall/uninstall - Documented the cluster-wide deployment rule that required services/timers must be activated with `systemctl enable --now` during install, not left merely enabled - Standardized `cluster/projects/pve-backup-scheduler` around `/usr/local/lib/xdev/pve-backup-scheduler`, added canonical lifecycle scripts and `setup.sh`, and kept `/etc/pve/autobackup` as an explicit preserved config exception - Standardized `cluster/projects/autoNAS` around `/usr/local/lib/xdev/autonas` and `/usr/local/sbin/autonas`, while keeping `/etc/pve/autonas` and `/mnt/autonas` as explicit shared-state exceptions - Grouped cluster metadata and historical cache files under `cluster-context/` and moved legacy snapshots under `cluster-context/history/` - Added cluster-wide deployment orchestration in `scripts/deploy-project.sh`, driven by `cluster-context/madagascar.json`, while preserving one-node deploy paths for development and testing - Tightened lifecycle cleanup for `pve-guests-state` legacy systemd units and suppressed `thunderbolts` recovery noise on hosts without `bolt.service` --- ## [2025-10-30] ### Fixed - [ISSUE-2025-001] Thunderbolt interfaces MTU persistence issue resolved - **Root cause**: `systemctl restart networking` resets MTU because systemd services don't re-trigger - **Solution**: Hybrid approach with udev rule enhancement + post-up hooks - **Changes**: Updated udev rules and interfaces.d configs on all nodes (baobab, ebony, tapia) - **Testing**: Verified MTU 65520 persists after networking restart on all nodes ### Added - Issue tracking system in `cluster/issues/` directory - CHANGELOG.md for documenting all cluster changes with issue references - Template for issue documentation (`issues/TEMPLATE.md`) - First documented issue: ISSUE-2025-001 regarding thunderbolt MTU reset problem - Added `scripts/check_mcluster_network.sh` for cluster thunderbridge and network health checks (table output, ping tests from localhost and baobab). ### Changed - Removed codebase-specific references from `madagascar.json` to keep it cluster-focused --- ## [2025-10-19] ### Added - PBS (Proxmox Backup Server) configuration to `madagascar.json` - andrafiabe-AutoNAS (192.168.2.96) - anjothibe-AutoNAS (192.168.2.95) - Node roles (primary/secondary) to cluster configuration --- ## [2025-10-18] ### Added - Initial `madagascar.json` cluster cache file - Cluster network documentation (thunderbolt bridge configuration) - WAN configuration for all nodes (vmbr443, vmbr444) - Node-specific network information (baobab, ebony, tapia) - `madagascar-changelog.json` for automation-triggered changes - `README_madagascar_cache.md` with file contract documentation ### Infrastructure - Thunderbolt bridge (thunderbridge) on 192.168.10.0/24 with MTU 65520 - WAN bridges on 192.168.2.0/24 (vmbr443) and 192.168.4.0/24 (vmbr444) --- ## Format Guidelines ### Categories - **Added** - new features, files, or configurations - **Changed** - changes to existing functionality or configuration - **Deprecated** - features or configurations that will be removed - **Removed** - removed features or configurations - **Fixed** - bug fixes (always reference issue number) - **Security** - security-related changes ### Entry Format ``` - Brief description [ISSUE-YYYY-NNN] (optional details) ``` ### Issue References Always link changes to issues when applicable: - Bug fixes must reference the issue - New features should reference planning/feature issues - Configuration changes should reference related issues or RFCs