- Remove all Default implementations from agent configuration structs
- Make configuration file required for agent startup
- Update NixOS module to generate complete agent.toml configuration
- Add comprehensive configuration options to NixOS module including:
- Service include/exclude patterns for systemd collector
- All thresholds and intervals
- ZMQ communication settings
- Notification and cache configuration
- Agent now fails fast if no configuration provided
- Eliminates configuration drift between defaults and NixOS settings
- Use systemctl --user commands to discover user-level services
- Include both user unit files and loaded user units
- Gracefully handle cases where user commands fail (no user session)
- Treat user services same as system services in filtering
- Enables monitoring of user-level Docker, development servers, etc.
- Add ark-permissions to exclusion list (maintenance service)
- Add sunshine to service_name_filters (game streaming server)
- Improves service discovery for game streaming infrastructure
- Use systemctl list-unit-files and list-units --all to find inactive services
- Parse both outputs to ensure all services are discovered
- Remove special SSH detection logic since sshd is in service filters
- Rename interesting_services to service_name_filters for clarity
- Now detects services in any state: active, inactive, failed, dead, etc.
- Remove bloated last_meaningful_status tracking
- Treat any Unknown→Ok transition as recovery
- Reduce JSON persistence to only metric_statuses and metric_details
- Eliminate unnecessary status history complexity
Add comprehensive hysteresis support to prevent status oscillation near
threshold boundaries while maintaining responsive alerting.
Key Features:
- HysteresisThresholds with configurable upper/lower limits
- StatusTracker for per-metric status history
- Default gaps: CPU load 10%, memory 5%, disk temp 5°C
Updated Components:
- CPU load collector (5-minute average with hysteresis)
- Memory usage collector (percentage-based thresholds)
- Disk temperature collector (SMART data monitoring)
- All collectors updated to support StatusTracker interface
Cache Interval Adjustments:
- Service status: 60s → 10s (faster response)
- Disk usage: 300s → 60s (more frequent checks)
- Backup status: 900s → 60s (quicker updates)
- SMART data: moved to 600s tier (10 minutes)
Architecture:
- Individual metric status calculation in collectors
- Centralized StatusTracker in MetricCollectionManager
- Status aggregation preserved in dashboard widgets
- Remove proxy_pass backend checking
- All sites now checked using https://server_name format
- Maintains 10-second timeout for external site checks
- Simplifies monitoring to consistent external health checks
- Add support for both proxied and static nginx sites
- Proxied sites show 'P' prefix and check backend URLs
- Static sites check external HTTPS URLs
- Fix services panel column alignment for main services
- Keep 10-second timeout for all site checks
- Add full email notifications with lettre and Stockholm timezone
- Add status persistence to prevent notification spam on restart
- Change nginx monitoring to check backend proxy_pass URLs instead of frontend domains
- Increase nginx site timeout to 10 seconds for backend health checks
- Fix cache intervals: disk (5min), backup (10min), systemd (30s), cpu/memory (5s)
- Remove rate limiting for immediate notifications on all status changes
- Store metric status in /var/lib/cm-dashboard/last-status.json
Only the 5-minute load average should trigger warning/critical alerts.
1-minute and 15-minute load averages now always show Status::Ok.
Thresholds (Warning: 9.0, Critical: 10.0) apply only to cpu_load_5min metric.
- Replace complex multi-strategy detection with single deterministic method
- Remove estimate_service_disk_usage and all fallback strategies
- Use simple get_service_disk_usage method with clear logic:
* Defined path exists → use only that path
* Defined path fails → return None (shows as '-')
* No defined path → use systemctl WorkingDirectory
* No estimates or guessing ever
Fixes misleading 5MB estimates when defined paths fail due to permissions.
ARK service directories require elevated permissions to access. The NixOS
configuration already allows sudo du with NOPASSWD, so use sudo du instead
of direct du command to properly detect disk usage for restricted directories.
- Prioritize defined service directories over systemctl WorkingDirectory fallbacks
- Add ARK Survival Ascended server mappings to correct NixOS-configured paths
- Remove legacy get_service_disk_usage method to eliminate code duplication
- Ensure deterministic behavior with single-purpose detection logic
Fixes ARK service disk usage reporting on srv02 by using actual data paths
from NixOS configuration instead of systemctl working directory detection.
Map each ARK service to its specific data directory:
- ark-island -> /var/lib/ark-servers/island
- ark-scorched -> /var/lib/ark-servers/scorched
- ark-center -> /var/lib/ark-servers/center
- ark-aberration -> /var/lib/ark-servers/aberration
- ark-extinction -> /var/lib/ark-servers/extinction
- ark-ragnarok -> /var/lib/ark-servers/ragnarok
- ark-valguero -> /var/lib/ark-servers/valguero
Based on NixOS configuration in srv02/configuration.nix.
Replace df-based auto-discovery with UUID-based detection using NixOS
hardware configuration data. Each host now has predefined filesystem
configurations with predictable metric names.
- Add FilesystemConfig struct with UUID, mount point, and filesystem type
- Remove auto_discover and devices fields from DiskConfig
- Add host-specific UUID defaults for cmbox, srv01, srv02, simonbox, steambox
- Remove legacy get_mounted_disks() df-based detection method
- Update DiskCollector to use UUID resolution via /dev/disk/by-uuid/
- Generate predictable metric names: disk_root_*, disk_boot_*, etc.
- Maintain fallback for labbox/wslbox (no UUIDs configured yet)
Provides consistent metric names across reboots and reliable detection
aligned with NixOS deployments without dependency on mount order.
Removed unused widget subscription system, cache utilities, error variants,
theme functions, and struct fields. Replaced subscription-based widgets
with direct metric filtering. Build now completes with zero warnings.
- Add sudo to disk collector smartctl commands for proper SMART data access
- Add reqwest dependency with blocking feature for HTTP site checks
- Replace curl-based site latency with reqwest HTTP client implementation
- Maintain 2-second connect timeout and 5-second total timeout
- Fix disk health UNKNOWN status by enabling proper SMART permissions
- Fix nginx site timeout issues by using proper HTTP client with redirect support
- Add BackupCollector for reading TOML status files with disk space metrics
- Implement BackupWidget with disk usage display and service status details
- Fix backup script disk space parsing by adding missing capture_output=True
- Update backup widget to show actual disk usage instead of repository size
- Fix timestamp parsing to use backup completion time instead of start time
- Resolve timezone issues by using UTC timestamps in backup script
- Add disk identification metrics (product name, serial number) to backup status
- Enhance UI layout with proper backup monitoring integration
This commit addresses several key issues identified during development:
Major Changes:
- Replace hardcoded top CPU/RAM process display with real system data
- Add intelligent process monitoring to CpuCollector using ps command
- Fix disk metrics permission issues in systemd collector
- Optimize service collection to focus on status, memory, and disk only
- Update dashboard widgets to display live process information
Process Monitoring Implementation:
- Added collect_top_cpu_process() and collect_top_ram_process() methods
- Implemented ps-based monitoring with accurate CPU percentages
- Added filtering to prevent self-monitoring artifacts (ps commands)
- Enhanced error handling and validation for process data
- Dashboard now shows realistic values like "claude (PID 2974) 11.0%"
Service Collection Optimization:
- Removed CPU monitoring from systemd collector for efficiency
- Enhanced service directory permission error logging
- Simplified services widget to show essential metrics only
- Fixed service-to-directory mapping accuracy
UI and Dashboard Improvements:
- Reorganized dashboard layout with btop-inspired multi-panel design
- Updated system panel to include real top CPU/RAM process display
- Enhanced widget formatting and data presentation
- Removed placeholder/hardcoded data throughout the interface
Technical Details:
- Updated agent/src/collectors/cpu.rs with process monitoring
- Modified dashboard/src/ui/mod.rs for real-time process display
- Enhanced systemd collector error handling and disk metrics
- Updated CLAUDE.md documentation with implementation details
Replace legacy SmartCache with MetricCollectionManager for granular control:
- RealTime tier (5s): CPU load, CPU temperature, Service CPU usage
- Fast tier (30s): Memory usage, top processes
- Medium tier (5min): Service status, C-states, users
- Slow tier (15min): Disk usage
All CPU-related metrics now update consistently every 5 seconds as requested,
eliminating the previous inconsistency where only CPU load was updating
at the correct frequency while service CPU usage was on 5-minute intervals.
The issue was that the metric-level system was sending individual
metric fragments (CPU load, temperature separately) instead of
complete System/Service messages that the dashboard expects.
Now aggregates individual metrics into complete messages:
- CPU load + temperature -> complete System message
- Memory + processes -> complete System message
- Service metrics remain as complete messages
This should resolve 'No data received' on srv01 while maintaining
the 5-second CPU metric update frequency.
The MetricCollector implementation was returning JSON with null values
because it was incorrectly extracting Option<&Value> instead of the
actual values. Fixed by using .cloned().unwrap_or() to properly
extract and default the JSON values.
This should resolve the 'No data received' issue as the dashboard
will now receive properly formatted metric data instead of null values.
Added detailed logging for ZMQ data sending to see exactly what
data is being transmitted and whether sends are successful.
This will help identify if the issue is in data format, sending,
or dashboard reception.
Added comprehensive error logging to identify why metrics are not being
collected successfully. This will help diagnose the 'No data received'
issue on srv01 by showing exactly which metrics are failing and why.
The handle_commands() function was being called continuously in the main
tokio::select! loop, causing thousands of ZMQ state errors that prevented
the agent from functioning properly.
Temporarily disabled command handling to restore basic functionality.
Agent now properly collects and sends metrics without ZMQ errors.
Fixes 'No data received' issue on hosts running the new metric-level agent.
Replace legacy SmartCache with MetricCollectionManager for precise control
over individual metric refresh intervals. CPU load and Service CPU usage
now update every 5 seconds as required, while other metrics use optimal
intervals based on volatility.
Key changes:
- ServiceCollector/SystemCollector implement MetricCollector trait
- Metric-specific cache tiers: RealTime(5s), Fast(30s), Medium(5min), Slow(15min)
- SmartAgent main loop uses metric-level scheduling instead of tier-based
- CPU metrics (load, temp, service CPU) refresh every 5 seconds
- Memory and processes refresh every 30 seconds
- Service status and C-states refresh every 5 minutes
- Disk usage refreshes every 15 minutes
Performance optimized architecture maintains <2% CPU usage while ensuring
dashboard responsiveness with precise metric timing control.
Implements ZMQ command protocol for dashboard-to-agent communication:
- Agents listen on port 6131 for REQ/REP commands
- Dashboard sends "refresh" command when 'r' key is pressed
- Agents force immediate collection of all metrics via force_refresh_all()
- Fresh data is broadcast immediately to dashboard
- Updated help text to show "r: Refresh all metrics"
Also includes metric-level caching architecture foundation for future
granular control over individual metric update frequencies.
Root cause: Smart agent only sent data when tier intervals triggered:
- System (5s): sent data frequently ✓
- Services (5min): sent data only every 5 minutes ✗
- SMART (15min): sent data only every 15 minutes ✗
Dashboard needs continuous data flow every ~5 seconds.
Solution: Add broadcast_all_data() method that sends all available
cached data every 5 seconds, separate from collection intervals.
This ensures dashboard receives all collector data continuously while
maintaining smart caching benefits (reduced CPU from tier-based collection).
Expected result: All widgets (System/Services/SMART/Backup) should
populate immediately after agent restart and stay updated.
Cache storage was using keys like 'hostname_service' but lookup was using
'hostname_CollectorName', causing all non-System collectors to fail.
Changes:
- Standardize cache keys to use collector names ('SystemCollector', 'ServiceCollector', etc.)
- Add cache_key() getter method to CachedCollector
- Fix cache lookup to use consistent keys
This should resolve the issue where srv01 only shows System data but no
Services/SMART/Backup data in the dashboard.
- Update quota from 200GB to 500GB (matches NixOS config)
- Fix disk usage path: /var/lib/immich-server -> /var/lib/immich
- Add service-to-directory mapping for accurate disk usage detection
This should resolve the "<1MB disk usage of 200GB" issue -
immich should now correctly show usage of /var/lib/immich with 500GB quota.
Replace traditional 5-second polling with tiered collection strategy:
- RealTime (5s): CPU load, memory usage
- Medium (5min): Service status, disk usage
- Slow (15min): SMART data, backup status
Key improvements:
- Reduce CPU usage from 9.5% to <2%
- Cache warming for instant dashboard responsiveness
- Background refresh at 80% of tier intervals
- Thread-safe cache with automatic cleanup
Remove legacy polling code - smart caching is now the default and only mode.
Agent startup enhanced with parallel cache population for immediate data availability.
Architecture: SmartCache + CachedCollector + tiered CollectionScheduler
- Implement proper quota detection for services with known systemd configurations
- Set gitea quota to 100GB (matches NixOS tmpfiles configuration)
- Add service-specific quotas: postgres/mysql 50GB, immich 200GB, unifi 10GB
- Fallback to service-appropriate defaults for other services
- Remove unreachable descriptions from failed nginx sites
- Show complete site URLs instead of truncating at first dot
- Implement service-specific disk quotas (docker: 4GB, immich: 4GB, others: 1-2GB)
- Truncate process names to show only executable name without full path
- Display only highest C-state instead of all C-states for cleaner output
- Format system RAM as xxxMB/GB (totalGB) to match services format
- Convert recursive async function to synchronous with return values
- Collect all status changes first, then process them asynchronously
- Resolves Rust compiler error E0733 for recursive async functions
- Maintains same functionality without boxing requirement
- Verified with full workspace build matching NixOS configuration
- Replace hardcoded agent-specific notification logic with generic scanner
- Automatically detect all '_status' fields across all collectors recursively
- Send email notifications from hostname@cmtec.se to cm@cmtec.se for any status changes
- Include agent name, component, and source path in notification description
- Works identically for System, Service, Smart, Backup, and future collectors
- Supports nested objects and arrays for comprehensive monitoring
- Remove check_site_accessibility function and filtering logic
- Monitor ALL nginx sites from config regardless of current status
- Site status determined by measure_site_latency, not accessibility filter
- Fixes missing git.cmtec.se when backend is down (502 errors)
- Sites with errors now show as failed instead of being filtered out
- Change site latency timeout from 5s to 2s for faster error detection
- Replace curl with reqwest for external connectivity checks (consistent timeouts)
- Remove unused gitea-specific monitoring functionality
- Update dashboard: show 'unreachable' for latency > 2000ms, add arrows (→) between site and latency
- Add percentage signs to CPU metrics display
- All HTTP requests now use reqwest with 2-second timeouts
- Move get_top_cpu_process() and get_top_ram_process() methods inside SystemCollector impl block
- Remove duplicate method definitions that were placed after trait implementation
- Ensures methods are properly accessible during compilation
- Implement get_top_cpu_process() and get_top_ram_process() functions in SystemCollector
- Add top_cpu_process and top_ram_process fields to SystemSummary data structure
- Update System widget to display top processes as description rows
- Show process name and percentage usage for highest CPU and RAM consumers
- Skip kernel threads and filter out processes with minimal usage (<0.1%)
- Fix false negatives for sites that don't handle HEAD requests properly
- Resolves photos.cmtec.se showing error when it actually works fine
- Improves compatibility with modern web applications
- Return error status for HTTP 502/5xx responses instead of success
- Show 'error' description for sites with connectivity but wrong status codes
- Show 'unreachable' description for complete connection failures
- Each nginx site now has independent status based on actual health
- Sites with timeouts or server errors will trigger notifications
- Show 'unreachable' status for nginx sites that fail connection tests
- Set service status to error (red) for unreachable sites
- Display latency in milliseconds for responsive sites
- Properly count failed sites in service summary statistics
- Improve nginx site monitoring reliability and visibility
Agent improvements:
- Add reqwest dependency for HTTP latency testing
- Implement measure_site_latency() function for nginx sites
- Add latency_ms field to ServiceData structure
- Measure response times for nginx sites using HEAD requests
- Handle connection failures gracefully with 5-second timeout
- Use HTTPS for external sites, HTTP for localhost
Dashboard improvements:
- Add latency_ms field to ServiceInfo structure
- Display latency for nginx sites: "docker.cmtec.se 134ms"
- Only show latency for nginx sub-services, not other services
- Change disk usage "0" to "<1MB" for better readability
The Services widget now shows:
- Nginx sites with response times when measurable
- Cleaner disk usage formatting for small values
- Improved user experience with meaningful latency data
Agent improvements:
- Add get_logged_in_users() function to SystemCollector using 'who' command
- Collect unique, sorted list of currently logged-in users
- Include logged_in_users field in system metrics JSON output
- Change C-state formatting to show 2 states per row instead of 4
Dashboard improvements:
- Update Backups widget to show "Archives: XX, ..." format
- System widget ready to display logged-in users with proper formatting
The System widget will now show:
- C-states formatted as 2 per row for better readability
- Logged-in users displayed as "Logged in: user" or "Logged in: X users (user1, user2)"
