This commit contains multiple fixes for temperature proxy registration,
but the core issue remains unresolved.
## What's Fixed:
1. Added config pointer and reloadFunc to TemperatureProxyHandlers
2. Added SetConfig method to keep handler in sync with router config changes
3. Added config reload after registration to prevent monitor from overwriting
4. Fixed installer port conflict detection and duplicate YAML key issues
5. Added comprehensive debug logging throughout registration flow
## What's Still Broken:
The TemperatureProxyURL, TemperatureProxyToken, and TemperatureProxyControlToken
fields are NOT persisting to nodes.enc after SaveNodesConfig is called.
Debug logs confirm:
- HandleRegister correctly updates nodesConfig.PVEInstances[matchedIndex]
- The correct data is passed to SaveNodesConfig (verified in logs)
- SaveNodesConfig completes without errors
- Config reload executes successfully
- BUT after Pulse restart, the fields are empty when loaded from disk
The bug is in SaveNodesConfig serialization or file writing logic itself.
Related files:
- internal/api/temperature_proxy.go: Registration handler
- internal/config/persistence.go: SaveNodesConfig implementation
- internal/config/config.go: PVEInstance struct definition
Fixed bug where config.yaml would end up with root:root 600 permissions
after the installer modified it, causing service startup failures with
"permission denied" errors.
Root cause: Two code paths modified config.yaml without resetting ownership:
1. ensure_control_plane_config() - used mktemp (creates root-owned file),
then mv'd it over config.yaml without chown/chmod
2. HTTP mode configuration - appended to config.yaml without resetting perms
Fix: Added chown/chmod after both modifications:
- Line 1601-1602: After control-plane config update
- Line 1860-1861: After HTTP mode config append
Now config.yaml maintains pulse-sensor-proxy:pulse-sensor-proxy 644
permissions after all modifications, allowing the service to start correctly.
This bug was discovered during repair logic testing - the service failed
to start after the installer ran, even though the fmt.Sprintf argument
alignment fix was working correctly.
Same turnkey UX issue as PVE quick setup: when local sensor-proxy socket
exists and validates, install-docker.sh would skip reinstallation (line 498).
This left stale control-plane tokens and URLs.
Fix: Always reinstall when LOCAL_PROXY_EXISTED_AT_START=true
- Track socket existence at script start
- Change message: 'will refresh to update tokens/config'
- Set SKIP_INSTALLATION=false to force reinstall
- Installer is idempotent (Phase 2), safe to rerun
This completes the turnkey repair fix across all entry points (PVE quick
setup and Docker installer).
- Update installer to use v4.32.0 Phase 2 binaries with file-based config
- Add automatic detection of Pulse service (systemd/hot-dev/docker)
- Add --restart-pulse flag for automatic Pulse restart in dev/test environments
- Default behavior shows clear instructions to restart Pulse manually (safe for production)
- Add prominent restart notice with command suggestions based on detected deployment
- Improve UX by making restart step impossible to miss
Related to Phase 2 sensor-proxy architecture improvements
Stop pulse-sensor-proxy service before modifying config.yaml to prevent
races with the running daemon or concurrent CLI commands.
The migration script now:
1. Stops service if running
2. Updates config atomically (temp + rename in same directory)
3. Restarts service if it was running
This achieves complete architectural isolation - ALL config file writers
are now coordinated (either through Phase 2 CLI locking or by ensuring
the service is stopped during modification).
Addresses final Codex review feedback.
Create temp file in same directory as config.yaml to ensure mv is truly
atomic (won't degrade to copy+unlink on different filesystems).
Added comments noting this is a legacy migration script with minor race
risk (no file locking) that should be deprecated once all users upgrade
to v4.32+.
1. Self-heal script: Add BINARY_PATH variable so CLI migration actually runs
- Previously logged "Binary not available" and skipped migration
2. migrate-sensor-proxy-control-plane.sh: Use atomic write (temp + rename)
- Prevents partial writes if script is interrupted
- Reduces race window with running service
These were the remaining gaps identified by Codex review.
NOTE: migrate-sensor-proxy-control-plane.sh still uses Python manipulation
instead of the Phase 2 CLI, but as a one-time migration script for upgrades
from v4.31, the atomic write provides sufficient protection. Future versions
can deprecate this script entirely.
Remove all code paths that manipulate config files without Phase 2 locking:
1. Installer: Remove ensure_allowed_nodes_file_reference() call (line 1674)
- Migration now handled exclusively by config migrate-to-file
2. Installer: Make migration failures fatal in update_allowed_nodes()
- Prevents fallback to unsafe Python manipulation
3. Daemon sanitizer: Remove os.WriteFile() call
- Now only sanitizes in-memory copy, doesn't write back to disk
- Logs warning instructing admin to run `config migrate-to-file`
4. Self-heal script: Replace 132 lines of Python with CLI call
- sanitize_allowed_nodes() now calls `config migrate-to-file`
- Eliminates uncoordinated Python-based config rewriting
All config mutations now flow exclusively through Phase 2 CLI with
atomic operations and file locking. No code paths remain that can
create duplicate allowed_nodes blocks.
Addresses Codex review feedback on Phase 2 gaps.
Fixes critical issues found by Codex code review:
**1. Fixed file locking race condition (CRITICAL)**
- Lock file was being replaced by atomic rename, invalidating the lock
- New approach: lock a separate `.lock` file that persists across renames
- Ensures concurrent writers (installer + self-heal timer) are properly serialized
- Without this fix, corruption was still possible despite Phase 2
**2. Fixed validation to honor configured allowed_nodes_file path**
- validate command now uses loadConfig() to read actual config
- Respects allowed_nodes_file setting instead of assuming default path
- Prevents false positives/negatives when path is customized
**3. Allow empty allowed_nodes lists**
- Empty lists are valid (admin may clear for security, or rely on IPC validation)
- validate no longer fails on empty lists
- set-allowed-nodes --replace with zero nodes now supported
- Critical for operational flexibility
**4. Installer error propagation**
- update_allowed_nodes failures now exit installer with error
- Prevents silent failures that leave stale allowlists
- Self-heal will abort instead of masking CLI errors
**Technical Details:**
- withLockedFile() now locks `<path>.lock` instead of target file
- Lock held for entire duration of read-modify-write-rename
- atomicWriteFile() completes while lock is still held
- Empty lists represented as `allowed_nodes: []` in YAML
**Testing:**
✅ Lock file created and persists across operations
✅ Empty list can be written with --replace
✅ Validation passes with empty lists
✅ Config path from allowed_nodes_file honored
✅ Concurrent operations properly serialized
These fixes ensure Phase 2 actually eliminates corruption by design.
Identified by Codex code review
Related to Phase 2 commit 3dc073a28
Implements bullet-proof configuration management to completely eliminate
allowed_nodes corruption by design. This builds on Phase 1 (file-only mode)
by replacing all shell/Python config manipulation with proper Go tooling.
**New Features:**
- `pulse-sensor-proxy config validate` - parse and validate config files
- `pulse-sensor-proxy config set-allowed-nodes` - atomic node list updates
- File locking via flock prevents concurrent write races
- Atomic writes (temp file + rename) ensure consistency
- systemd ExecStartPre validation prevents startup with bad config
**Architectural Changes:**
1. Installer now calls config CLI instead of embedded Python/shell scripts
2. All config mutations go through single authoritative writer
3. Deduplication and normalization handled in Go (reuses existing logic)
4. Sanitizer kept as noisy failsafe (warns if corruption still occurs)
**Implementation Details:**
- New cmd/pulse-sensor-proxy/config_cmd.go with cobra commands
- withLockedFile() wrapper ensures exclusive access
- atomicWriteFile() uses temp + rename pattern
- Installer update_allowed_nodes() simplified to CLI calls
- Both systemd service modes include ExecStartPre validation
**Why This Works:**
- Single code path for all writes (no shell/Python divergence)
- File locking serializes self-heal timer + manual installer runs
- Validation gate prevents proxy from starting with corrupt config
- CLI uses same YAML parser as the daemon (guaranteed compatibility)
**Phase 2 Benefits:**
- Corruption impossible by design (not just detected and fixed)
- No more Python dependency for config management
- Atomic operations prevent partial writes
- Clear error messages on validation failures
The defensive sanitizer remains active but now logs loudly if triggered,
allowing us to confirm Phase 2 eliminates corruption in production before
removing the safety net entirely.
This completes the fix for the recurring temperature monitoring outages.
Related to Phase 1 commit 53dec6010
Phase 1 hotfix to address recurring config file corruption that causes
99% of temperature monitoring failures. The root cause was the installer
oscillating between inline and file-based allowlist modes, creating
duplicate `allowed_nodes:` keys in config.yaml.
Changes:
- Force file-based allowlist mode exclusively (refuse versions < v4.31.1)
- Add automatic migration from inline to file-based config
- Remove inline mode code path from update_allowed_nodes()
- Migration runs on every install/self-heal to clean up existing corruption
The self-heal timer runs every 5 minutes and was the primary source of
corruption when version detection failed or encountered edge cases.
This eliminates the dual code paths and ensures config.yaml is never
edited for allowlist changes - only /etc/pulse-sensor-proxy/allowed_nodes.yaml
is modified.
Phase 2 (next release) will implement proper Go-based config management
with atomic writes, locking, and systemd validation to prevent corruption
by design.
Related to recurring temperature monitoring outages
The installer was adding node hostnames (and accidentally the header "Name")
to allowed_nodes in addition to IPs. This caused:
1. Invalid entries like "Name", "minipc", "delly" in config
2. These are not valid for SSH temperature collection
Only IPs should be in allowed_nodes since that's what the proxy uses for SSH.
Removed the loop that added CLUSTER_NODE_NAMES to the array.
Also fixed: Removed extraction of CLUSTER_NODE_NAMES since it's no longer used.
Problem:
Cleanup script uses systemd-run with both --wait and
--property="After=pulse-sensor-cleanup.service", creating a circular
dependency:
- cleanup.service runs and waits for uninstaller to complete
- uninstaller has After=cleanup.service, so it waits for cleanup to finish
- Result: Both services stuck waiting for each other
Fix:
Remove the --property="After=pulse-sensor-cleanup.service" line. The
Conflicts=pulse-sensor-proxy.service is sufficient to ensure the proxy
stops before uninstallation. The cleanup script doesn't need to finish
before the uninstaller starts.
Testing:
Cleanup now completes successfully, removing all artifacts:
- Systemd units removed
- Binaries deleted from /opt/pulse/sensor-proxy/
- Data directory /var/lib/pulse-sensor-proxy/ removed
- SSH keys cleaned from authorized_keys
- pulse-monitor user and API tokens deleted
- LXC bind mounts removed from container configs
Related to #605 (temperature monitoring cleanup)
**Missed Migration**:
- Line 2204 still used /usr/local/bin/pulse-sensor-wrapper.sh in fallback path
- Updated to use /opt/pulse/sensor-proxy/bin/pulse-sensor-wrapper.sh
**Backward Compatibility**:
- When pushing SSH keys to cluster nodes, installer now checks if remote node
has old installation (/usr/local/bin wrapper exists but /opt path doesn't)
- Automatically creates symlink on remote nodes to maintain compatibility
- Prevents temperature collection failures when cluster has mixed old/new installs
**Root Cause**:
When installer runs on upgraded node (delly), it pushes SSH keys with new forced
command path to all cluster nodes. If remote node (minipc) has old installation,
the forced command fails because wrapper doesn't exist at new path.
This fix ensures "it works straight out the box" by bridging old and new paths
automatically during SSH key deployment.
Rewrote AWK state machine to correctly handle:
- Multiple allowed_nodes sections (removes all of them)
- Comment lines immediately preceding allowed_nodes (discards them)
- Empty lines within allowed_nodes section
- Indented list items and comments
The function now:
1. Buffers comment lines that might precede allowed_nodes
2. When allowed_nodes: is detected, discards buffered comments
3. Skips all content until hitting a non-indented, non-comment line
4. Flushes buffered comments when hitting non-comment content
This ensures running the installer multiple times won't create duplicate
allowed_nodes sections in config.yaml.
Tested with script that verifies duplicate sections are removed correctly.
**LXC Bind Mount Removal**:
- Changed from sed to `pct set -delete` for safer mount removal
- Validates syntax and prevents breaking container configs
- Finds mount points by grepping for pulse-sensor-proxy, extracts mp number
**API Token Parsing** (three-tier fallback):
1. Try `pveum --output-format json` with python3 JSON parsing
2. Fall back to `pvesh get /access/users/pulse-monitor@pam/token` JSON API
3. Last resort: parse table output with improved filtering (handles more Unicode chars)
**Retry Logic**:
- Rename cleanup-request.json to .processing instead of deleting immediately
- Allows retry on failure (processing file persists if script crashes)
- Remove .processing file only on successful completion
- Prevents loops while enabling failure recovery
These complete all 8 issues identified by Codex review.
**Host Detection**:
- Now detects localhost by hostname and FQDN, not just IP
- Fixes issue where nodes configured as https://hostname:8006 would skip
localhost cleanup (API tokens, bind mounts, service removal)
**Systemd Sandbox**:
- Added /etc/pve and /etc/systemd/system to ReadWritePaths
- Allows cleanup script to modify Proxmox configs and systemd units
**Uninstaller Improvements**:
- Use UUID for transient unit names (prevents same-second collisions)
- Added --purge flag for complete removal
- Added --wait and --collect flags to capture exit code
- Now fails cleanup if uninstaller exits non-zero
**Path Migration**:
- Fixed all /usr/local references to use /opt/pulse/sensor-proxy
- Updated forced command in SSH authorized_keys
- Updated self-heal script installer path
- Updated Go backend removal helpers (supports both new and legacy paths)
These fixes address Codex findings: hostname detection, sandbox permissions,
transient unit collisions, incomplete purging, and incomplete path migration.
Related to cleanup implementation testing.
After relocating binaries to INSTALL_ROOT, the SHARE_DIR variable was removed
but one reference remained in cache_installer_for_self_heal() causing
'unbound variable' error.
Changed to use INSTALL_ROOT directly since that's where the cached installer
is stored (STORED_INSTALLER=${INSTALL_ROOT}/install-sensor-proxy.sh).
The installer was trying to write binaries to /opt/pulse/sensor-proxy/bin/
before creating the directory structure, causing 'No such file or directory'
errors on fresh installs.
Moved directory creation for INSTALL_ROOT and bin/ to before binary installation
section (before line 657), ensuring directories exist before use.
Related to cleanup implementation testing.
Extends cleanup script to completely remove Pulse footprint from hosts
when nodes are removed, not just SSH keys. Now removes: SSH keys, proxy
service, binaries, API tokens, pulse-monitor user, and LXC bind mounts.
Key improvements:
1. **flock Serialization**: Prevents concurrent cleanup runs
- Acquires exclusive lock on cleanup.lock file
- Prevents race conditions and cleanup loops
2. **Immediate Request File Deletion**: Delete cleanup-request.json
before any long-running operations to prevent re-triggering
3. **API Token Cleanup**: Removes all pulse-monitor@pam API tokens
- Tries JSON output first (Proxmox 7.0+)
- Falls back to table parsing with proper filtering (no decoration chars)
- Deletes pulse-monitor@pam user after removing all tokens
4. **LXC Bind Mount Removal**: Scans all container configs and removes
pulse-sensor-proxy bind mount entries
5. **Process Isolation for Uninstaller**: Uses systemd-run to spawn
isolated transient unit that won't be killed when proxy service stops
- Unit name: pulse-uninstall-{timestamp}
- Properties: Type=oneshot, Conflicts=pulse-sensor-proxy.service
- Runs non-blocking so cleanup service can exit cleanly
- Falls back to direct call if systemd-run unavailable
6. **Complete Service/Binary Removal**: Calls installer's --uninstall
- Stops and disables pulse-sensor-proxy.service
- Removes all systemd units
- Deletes all binaries from /opt/pulse/sensor-proxy/
- Removes configuration files
- Cleans up directories
Changes to cleanup script logic:
- Added LOCKFILE and INSTALLER_PATH configuration
- Acquire flock before processing (prevents concurrent runs)
- Delete request file immediately after reading
- Full localhost cleanup: SSH keys → API tokens → bind mounts → uninstall
- Remote cleanup still SSH-key-only (can't orchestrate uninstall remotely)
- Better error handling with appropriate log levels
Updated cleanup service unit:
- ExecStart now uses ${CLEANUP_SCRIPT_PATH} variable (new /opt location)
- Changed heredoc from 'SERVICE_EOF' to SERVICE_EOF for variable expansion
Addresses all issues documented in CLEANUP_TODO.md:
- ✅ Read-only filesystem (binaries now in /opt, removable)
- ✅ Process isolation (systemd-run transient unit)
- ✅ Cleanup loops (flock + immediate file deletion)
- ✅ API token parsing (JSON first, filtered table fallback)
The UI message is now accurate: "Removing this proxmox ve node also
scrubs the Pulse footprint on the host — the proxy service, SSH key,
API token, and bind mount are all cleaned up automatically."
Part of: CLEANUP_TODO.md Phase 2-4
Supersedes: ed65fda74 (original cleanup attempt with process issues)
Depends on: b192c60e9 (binary relocation to /opt)
