Addresses user concern about technical debt: detection code exists only
to handle migration from SSH-in-container to proxy architecture, not to
serve functional purpose of the application.
Changes:
- Add PULSE_LEGACY_DETECTION env var to disable detection without redeployment
- Add explicit removal criteria: v5.0 or <1% detection rate for 30+ days
- Mark all detection code with "MIGRATION SCAFFOLDING" warnings
- Create MIGRATION_SCAFFOLDING.md to track temporary code across codebase
- Document removal instructions for when migration period ends
Backend:
- internal/api/router.go: detectLegacySSH() checks env var and has removal plan
- internal/api/types.go: HealthResponse fields documented as temporary
Frontend:
- src/components/LegacySSHBanner.tsx: Component marked with removal criteria
- src/App.tsx: Banner integration (will be removed with component)
This approach balances user safety during migration (auto-detection catches
rushed admins who skip changelogs) with long-term code cleanliness (explicit
removal plan prevents indefinite technical debt).
Added automatic detection to alert users when they're using the old
SSH-in-container method for temperature monitoring so they can upgrade
to the secure proxy architecture.
**Detection Logic:**
- Checks if Pulse is running in a container (Docker or LXC)
- Checks if SSH keys exist in data directory (/etc/pulse/.ssh)
- Checks if pulse-sensor-proxy socket is NOT available
- Sets legacySSHDetected and recommendProxyUpgrade flags in health endpoint
**API Changes:**
- Added fields to HealthResponse:
- legacySSHDetected: true when old method detected
- recommendProxyUpgrade: true when upgrade is recommended
- proxyInstallScriptAvailable: always true
**Use Case:**
Users who set up temperature monitoring before the proxy feature
won't know they should upgrade. This detection allows the frontend
to show a banner prompting them to re-run the setup script to
migrate to the secure proxy architecture.
**Frontend Integration (to be added):**
Frontend can poll /api/health and show a dismissible banner similar
to UpdateBanner when legacySSHDetected is true, with a button to
view the setup script.
Addresses #123
Codex caught an edge case in the authorized_keys removal logic:
**Problem:**
When authorized_keys contains ONLY pulse-managed keys, `grep -vF` returns
exit code 1 (no lines matched the inverse filter). The previous code only
executed the rewrite on exit 0, leaving managed keys in place when they
should have been removed.
**Solution:**
- Capture grep exit code explicitly
- Treat both exit 0 (lines remain) and exit 1 (all removed) as success
- Only treat exit codes > 1 as actual errors
- Properly handles the "remove all keys" scenario
This ensures complete removal works even when the file contains nothing
but Pulse-managed entries.
Addresses #123
Fixed three remaining issues from Codex's final review:
**1. nullglob State Management (line 3124)**
- Replaced shopt -s/u nullglob with compgen -G check
- Prevents changing global shell behavior that could affect later globs
- More explicit and safer pattern matching
**2. authorized_keys Permission Preservation (lines 3116-3117)**
- Now uses chmod/chown --reference to preserve original ownership/perms
- Falls back gracefully if --reference not available
- Proper cleanup on mv failure to prevent temp file leaks
- Aborts atomically if operations fail, leaving original untouched
**3. Multi-Address Container Detection (lines 3750-3761)**
- Iterates over ALL IPs from hostname -I, not just first one
- Handles dual-stack (IPv4 + IPv6) and multi-IP containers
- Uses break 2 to exit both loops when match found
- Prevents false negatives when Pulse IP is not the first address
All operations now handle edge cases properly: non-root accounts,
dual-stack networking, empty directories, and partial failures.
Addresses #123
Applied Codex's security and reliability recommendations:
**SSH Key Safety:**
- Added "pulse-managed-key" comment marker to all SSH keys
- Removal now targets only marked keys (prevents deleting operator keys)
- Uses atomic file replacement via mktemp for authorized_keys edits
**Idempotency Improvements:**
- LXC config glob now uses nullglob to handle empty directories
- pveum token removal handles missing users gracefully (|| printf '')
- All systemctl operations wrapped with || true for non-systemd hosts
- sed operations in loops protected with || true
**Container Detection:**
- Validates container is running before IP check (pct status)
- Confirms container exists with pct config before proceeding
- Uses printf '' instead of || true for command substitution
- Handles IPv6 and multi-IP scenarios more reliably
**Network Operations:**
- curl now uses --fail --show-error --silent --location
- Error messages visible to users instead of silenced
- Better diagnostics when download fails
**Migration Safety:**
- Verifies pulse-sensor-proxy service is active before key removal
- Fallback check for binary existence if systemd unavailable
- Preserves legacy SSH keys if proxy not confirmed healthy
- Clear messaging about deferred cleanup
All cleanup operations are now fully idempotent and safe for
repeated execution, even on partially-configured hosts.
Addresses #123
Added a main menu at the beginning of the PVE setup script that gives users three options:
[I]nstall - Continue with normal setup (default)
[R]emove All - Complete uninstall of all Pulse components
[C]ancel - Exit without changes
The removal option comprehensively cleans up:
- pulse-sensor-proxy service, binary, and systemd unit
- pulse-sensor-proxy system user and data directories
- All SSH keys from authorized_keys (legacy and forced-command variants)
- LXC bind mounts from all container configs
- Pulse monitoring API tokens, user, and custom roles
This addresses user request for a clean removal path for everything
Pulse has installed on the host, including legacy components from
previous versions.
Made the setup and installation output more concise and reassuring for users. Less verbosity, clearer messaging.
**Setup script improvements:**
- Changed "Container Detection" → "Enhanced Security"
- Simplified prompts: "Enable secure proxy? [Y/n]"
- Cleaned up success messages: "✓ Secure proxy architecture enabled"
- Removed verbose status messages (node-by-node cleanup output)
- Only show essential information users need to see
**install-sensor-proxy.sh improvements:**
- Added --quiet flag to suppress verbose output
- In quiet mode, only shows: "✓ pulse-sensor-proxy installed and running"
- Full output still available when run manually
- Removed redundant "Installation complete!" banners
- Cleaner legacy key cleanup messaging
**Result:**
Users see a clean, professional installation flow that builds confidence. Technical details are hidden unless needed. Messages are clear and reassuring rather than verbose.
When pulse-sensor-proxy is installed, automatically remove old SSH keys that were stored in the container for security.
Changes:
**install-sensor-proxy.sh:**
- Checks container for SSH private keys (id_rsa, id_ed25519, etc.)
- Removes any found keys from container
- Warns user that legacy keys were cleaned up
- Explains proxy now handles SSH
**Setup script (config_handlers.go):**
- After successful proxy install, removes old SSH keys from all cluster nodes
- Cleans up authorized_keys entries that match the old container-based key
- Keeps only proxy-managed keys (pulse-sensor-proxy comment)
This provides a clean migration path from the old direct-SSH method to the secure proxy architecture. Users upgrading from pre-v4.24 versions get automatic cleanup of insecure container-stored keys.
The setup script now automatically detects when Pulse is running in an LXC container and offers to install pulse-sensor-proxy on the host for enhanced security.
What happens:
1. After temperature monitoring is configured
2. Script detects Pulse IP and finds matching container
3. Prompts: "Install pulse-sensor-proxy for container X? [Y/n]"
4. Downloads and runs install-sensor-proxy.sh automatically
5. Falls back gracefully if proxy install fails
Benefits:
- One-command setup for users (no manual proxy installation)
- SSH keys stay on host (not in container)
- Containerized Pulse gets the secure architecture automatically
- Native installs unaffected (still use direct SSH)
This solves the UX problem where users had to manually run install-sensor-proxy.sh as a separate step.
The name "temp-proxy" implied a temporary or incomplete implementation. The new name better reflects its purpose as a secure sensor data bridge for containerized Pulse deployments.
Changes:
- Renamed cmd/pulse-temp-proxy/ to cmd/pulse-sensor-proxy/
- Updated all path constants and binary references
- Renamed environment variables: PULSE_TEMP_PROXY_* to PULSE_SENSOR_PROXY_*
- Updated systemd service and service account name
- Updated installation, rotation, and build scripts
- Renamed hardening documentation
- Maintained backward compatibility for key removal during upgrades
Addresses #101
v4.23.0 introduced a regression where systems with only NVMe temperatures
(no CPU sensor) would display "No CPU sensor" in the UI. This was caused
by the Available flag being set to true when NVMe temps existed, even
without CPU data, triggering the error message in the frontend.
Backend changes:
- Add HasCPU and HasNVMe boolean fields to Temperature model
- Extend CPU sensor detection to support more chip types: zenpower,
k8temp, acpitz, it87 (case-insensitive matching)
- HasCPU is set based on CPU chip detection (coretemp, k10temp, etc.),
not value thresholds
- This prevents false negatives when sensors report 0°C during resets
- CPU temperature values now accepted even when 0 (checked with !IsNaN
instead of > 0)
- extractTempInput returns NaN instead of 0 when no data found
- Available flag means "any temperature data exists" for backward compatibility
- Update mock generator to properly set the new flags
- Add unit tests for NVMe-only and 0°C scenarios to prevent regression
- Removed amd_energy from CPU chip list (power sensor, not temperature)
Frontend changes:
- Add hasCPU and hasNVMe optional fields to Temperature interface
- Update NodeSummaryTable to check hasCPU flag with fallback to available
for backward compatibility with older API responses
- Update NodeCard temperature display logic with same fallback pattern
- Systems with only NVMe temps now show "-" instead of error message
- Fallback ensures UI works with both old and new API responses
Testing:
- All unit tests pass including NVMe-only and 0°C test cases
- Fix prevents false "no CPU sensor" errors when sensors temporarily report 0°C
- Fix eliminates false "no CPU sensor" errors for NVMe-only systems
Fixes LXC bind mount issue where socket-level mounts break when the
socket is recreated by systemd. Following Codex's recommendation to
bind mount the directory instead of the file.
Changes:
- Socket path: /run/pulse-temp-proxy/pulse-temp-proxy.sock
- Systemd: RuntimeDirectory=pulse-temp-proxy (auto-creates /run/pulse-temp-proxy)
- Systemd: RuntimeDirectoryMode=0770 for group access
- LXC mount: Bind entire /run/pulse-temp-proxy directory
- Install script: Upgrades old socket-level mounts to directory-level
- Install script: Detects and handles bind mount changes
This survives socket recreations and container restarts. The directory
mount persists even when systemd unlinks/recreates the socket file.
Related to #528
Replaced outdated security warnings with accurate information about
the pulse-temp-proxy architecture:
- Removed scary 'legacy feature' and 'compromised container' warnings
- Explains secure proxy architecture for containerized deployments
- Notes that SSH keys are stored on Proxmox host (not in container)
- Clarifies container compromise does not expose credentials
- Includes information for both containerized and native installs
- More factual and less alarmist tone
The old message implied temperature monitoring was insecure for
containers, which is no longer true with pulse-temp-proxy.
Related to #528
The setup script generator was passing sshPublicKey twice but only
using it once, causing a Go fmt.Sprintf formatting error that leaked
into the generated bash script as '%!(EXTRA string=...)'.
This resulted in bash syntax errors when running the setup script.
Fixes#528
Addresses #528
Introduces pulse-temp-proxy architecture to eliminate SSH key exposure in containers:
**Architecture:**
- pulse-temp-proxy runs on Proxmox host (outside LXC/Docker)
- SSH keys stored on host filesystem (/var/lib/pulse-temp-proxy/ssh/)
- Pulse communicates via unix socket (bind-mounted into container)
- Proxy handles cluster discovery, key rollout, and temperature fetching
**Components:**
- cmd/pulse-temp-proxy: Standalone Go binary with unix socket RPC server
- internal/tempproxy: Client library for Pulse backend
- scripts/install-temp-proxy.sh: Idempotent installer for existing deployments
- scripts/pulse-temp-proxy.service: Systemd service for proxy
**Integration:**
- Pulse automatically detects and uses proxy when socket exists
- Falls back to direct SSH for native installations
- Installer automatically configures proxy for new LXC deployments
- Existing LXC users can upgrade by running install-temp-proxy.sh
**Security improvements:**
- Container compromise no longer exposes SSH keys
- SSH keys never enter container filesystem
- Maintains forced command restrictions
- Transparent to users - no workflow changes
**Documentation:**
- Updated TEMPERATURE_MONITORING.md with new architecture
- Added verification steps and upgrade instructions
- Preserved legacy documentation for native installs
When Pulse runs in a container (LXC/Docker), the setup script would claim
temperature monitoring was enabled on cluster nodes, but Pulse couldn't
actually SSH to them. The script ran on the Proxmox host which could SSH
fine, but didn't verify connectivity from Pulse itself.
Changes:
- Added /api/system/verify-temperature-ssh endpoint that tests SSH from Pulse
- Setup script now calls this endpoint after configuring cluster nodes
- Detects when Pulse is containerized and provides ProxyJump config instructions
- Shows clear success/failure status for each node
Addresses #528
Add no-cache headers to both the install script and agent binary download endpoints to prevent browsers and curl from serving stale cached versions. This ensures users always get the latest install script with URL normalization fixes for trailing slash issues.
Fixes#528
The PUT /api/config/nodes/{id} endpoint was corrupting node configurations
when making partial updates (e.g., updating just monitorPhysicalDisks):
- Authentication fields (tokenName, tokenValue, password) were being cleared
when updating unrelated settings
- Name field was being blanked when not included in request
- Monitor* boolean fields were defaulting to false
Changes:
- Only update name field if explicitly provided in request
- Only switch authentication method when auth fields are explicitly provided
- Preserve existing auth credentials on non-auth updates
- Applied fix to all node types (PVE, PBS, PMG)
Also enables physical disk monitoring by default (opt-out instead of opt-in)
and preserves disk data between polling intervals.
Simplified host sidebar to show only essential monitoring info (name, status, alerts, container count). Moved detailed metrics (CPU, memory, uptime) to host detail view when selected to eliminate duplication. Added alert highlighting with visual indicators for hosts with container alerts. Fixed double 'v' prefix in agent version display.
Backend was loading system settings from disk on every HTTP request
causing massive log spam and UI responsiveness issues. System settings
are now cached at startup and reloaded only when updated.
Also fixed tab cursor showing text cursor instead of pointer cursor.
Changes:
- Cache system settings in Router struct with mutex protection
- Load settings once at startup instead of on every request
- Add reloadSystemSettings() method to refresh cache when settings change
- Hook up cache reload in config handlers after successful save
- Add cursor-pointer to tab CSS classes for proper hover cursor
Backend:
- Handle 'no releases found' error gracefully instead of returning 500
- Return proper response indicating no updates available for the channel
- Fixes the 500 error when checking for updates on development versions
Frontend:
- Detect when backend enters 'restarting' phase during updates
- Show 'Pulse is restarting...' message instead of getting stuck on 'Initializing...'
- Implement health check polling with exponential backoff (2s → 15s max)
- Auto-reload page when backend becomes healthy again
- Improve messaging: inform users page will reload automatically
This fixes the stuck update modal issue where users would see
'Initializing...' forever when the backend restarted during updates.
The update history infrastructure (history.go, UI panel) already exists
and is used by InstallShAdapter. However, for a home lab monitoring tool,
tracking every update in the Manager is overkill.
Removed:
- History field and initialization from Manager
- Event tracking in ApplyUpdate function
- Update history population on success/failure
The history.go file and UI panel remain for InstallShAdapter usage,
but the main Manager update path is now simpler and more focused.
Fixed 12 critical issues in the update system:
Critical bugs:
- Cache invalidation when channel auto-detected vs explicitly provided
- Data race on cache access (added proper RWMutex locking)
- Incorrect error handling when already on latest version
Security improvements:
- Command injection prevention with version string validation
- Symlink vulnerabilities in backup/restore (replaced cp with safe Go implementations)
- SHA256 checksum verification for downloads
- Improved backup completeness (now includes binary and VERSION file)
Quality improvements:
- Channel-keyed cache map instead of single cache value
- Update history tracking with event IDs and rollback support
- Cleanup of old temp directories (24+ hours)
- Proper resource cleanup with Close() method
- Auto-detect update channel from current version when config is empty
(RC users automatically get RC updates, stable users get stable)
- Add background update checker that runs on startup and hourly
- Add GetCachedUpdateInfo() method to return cached update info
- Update /api/version endpoint to include updateAvailable from cache
- Fixes issue where RC users on rc.4 weren't seeing rc.5 updates
This ensures RC testers continue receiving RC notifications while
stable users stay on the stable channel, all without explicit config.
- Fix update history using hardcoded /var/lib/pulse instead of configured data dir
- Skip mock.env watching in Docker environments to avoid 'no such file' errors
Fixes issue #529
Update History Path:
- Modified NewUpdateHandlers to accept dataDir parameter
- Pass r.config.DataPath from router (honors PULSE_DATA_DIR=/data in Docker)
- Maintains backward compatibility with default /var/lib/pulse when empty
Mock Env Watcher:
- Check PULSE_DOCKER env var and skip mock.env watching if true
- Only watch /opt/pulse/mock.env if directory exists (not in containers)
- Guard all mock.env operations to prevent errors when path is empty
- Clean up log messages to only show mock_env_path when actually watching
- Strip trailing slash from PULSE_URL in install script to prevent double-slash URLs
- Add path normalization in router for defense-in-depth on public endpoint matching
- Fixes issue #528 where users copying URLs with trailing slashes got 401 errors
The install script now normalizes PULSE_URL with ${PULSE_URL%/} before concatenating
with /download/pulse-docker-agent, preventing https://example.com//download URLs.
The router normalization provides additional resilience for path matching, though the
existing path traversal check already blocks double slashes at ServeHTTP level.
rcourtman