The service_status_cache from discovery only has active_state with
all detailed metrics set to None. During collection, get_service_status()
was returning cached data instead of fetching fresh systemctl show data.
Now always fetch fresh data to populate memory_bytes, restart_count,
and uptime_seconds properly.
Shared:
- Add memory_bytes, restart_count, uptime_seconds to ServiceData
Agent:
- Add new fields to ServiceStatusInfo struct
- Fetch MemoryCurrent, NRestarts, ExecMainStartTimestamp from systemctl show
- Calculate uptime from start timestamp
- Parse and populate new fields in ServiceData
- Remove unused load_state and sub_state fields
Dashboard:
- Add memory_bytes, restart_count, uptime_seconds to ServiceInfo
- Update header: Service, Status, RAM, Uptime, ↻ (restarts)
- Format memory as MB/GB
- Format uptime as Xd Xh, Xh Xm, or Xm
- Show restart count with ! prefix if > 0 to indicate instability
All metrics obtained from single systemctl show call - zero overhead.
Changed header from 4 columns to 2 columns:
- Before: Service, Status, RAM, Disk
- After: Service, Status
Matches the removal of memory_mb and disk_gb fields.
Complete removal of service resource metrics:
Agent:
- Remove memory_mb and disk_gb fields from ServiceData struct
- Remove get_service_memory_usage() method
- Remove get_service_disk_usage() method
- Remove get_directory_size() method
- Remove unused warn import
Dashboard:
- Remove memory_mb and disk_gb from ServiceInfo struct
- Remove memory/disk display from format_parent_service_line
- Remove memory/disk parsing in legacy metric path
- Remove unused format_disk_size() function
Service resource metrics were slow, unreliable, and never worked
properly since structured data migration. Will be handled differently
in the future.
Service control migrated to SSH, command receiver no longer needed.
- Remove command_receiver Socket from ZmqHandler
- Remove try_receive_command method
- Remove AgentCommand enum
- Remove command_port from ZmqConfig
- Remove zbus dependency from agent
- Replace D-Bus Connection calls with systemctl show commands
- Fix agent hang by eliminating blocking D-Bus operations
- get_unit_property now uses systemctl show with property flags
- Memory, disk usage, and nginx config queries use systemctl
- Simpler, more reliable service monitoring
The D-Bus ListUnits call in discover_services_internal() was causing
the agent to hang on startup.
**Root cause:**
- D-Bus ListUnits call with complex tuple destructuring hung indefinitely
- Agent never completed first collection cycle
- No collector output in logs
**Fix:**
- Revert discover_services_internal() to use systemctl list-units/list-unit-files
- Keep D-Bus-based property queries (WorkingDirectory, MemoryCurrent, ExecStart)
- Hybrid approach: systemctl for discovery, D-Bus for individual queries
**External commands still used:**
- systemctl list-units, list-unit-files (service discovery)
- smartctl (SMART data)
- sudo du (directory sizes)
- nginx -T (config fallback)
Version bump: 0.1.198 → 0.1.199
Complete migration from systemctl subprocess calls to native D-Bus communication:
**Removed systemctl commands:**
- systemctl is-active (fallback) - use D-Bus cache from ListUnits
- systemctl show --property=LoadState,ActiveState,SubState - use D-Bus cache
- systemctl show --property=WorkingDirectory - use D-Bus Properties.Get
- systemctl show --property=MemoryCurrent - use D-Bus Properties.Get
- systemctl show nginx --property=ExecStart - use D-Bus Properties.Get
**Implementation details:**
- Added get_unit_property() helper for D-Bus property access
- Made get_nginx_site_metrics() async to support D-Bus calls
- Made get_nginx_sites_internal() async
- Made discover_nginx_sites() async
- Made get_nginx_config_from_systemd() async
- Fixed RwLock guard Send issues by using scoped locks
**Remaining external commands:**
- smartctl (disk.rs) - No Rust alternative for SMART data
- sudo du (systemd.rs) - Directory size measurement
- nginx -T (systemd.rs) - Nginx config fallback
- timeout hostname (nixos.rs) - Rare fallback only
Version bump: 0.1.197 → 0.1.198
CRITICAL FIX: The previous cached collector architecture still had ZMQ sending
in the main event loop, where it could block waiting for RwLock when collectors
were writing. This caused the 3-8 second delays you observed.
Changes:
- Move ZMQ publisher to dedicated std::thread (ZMQ sockets aren't thread-safe)
- Use try_read() instead of read() to avoid blocking on write locks
- Send previous data if cache is locked by collector
- ZMQ now sends every 2s regardless of collector timing
- Remove publisher from ZmqHandler (now only handles commands)
Architecture:
- Collectors: Independent tokio tasks updating shared cache
- ZMQ Sender: Dedicated OS thread with its own publisher socket
- Main Loop: Only handles commands and notifications
This ensures ZMQ transmission is NEVER blocked by slow collectors.
Bump version to v0.1.195
Critical bug fix: Collectors were appending to Vecs instead of replacing them,
causing duplicate entries with each collection cycle.
Fixed by adding .clear() calls before populating:
- Memory collector: tmpfs Vec (was showing 11+ duplicates)
- Disk collector: drives and pools Vecs
- Systemd collector: services Vec
- Network collector: Already correct (assigns new Vec)
This prevents the exponential growth of duplicate entries in the dashboard UI.
Major architectural refactor to eliminate false "host offline" alerts:
- Replace sequential blocking collectors with independent async tasks
- Each collector runs at configurable interval and updates shared cache
- ZMQ sender reads cache every 1-2s regardless of collector speed
- Collector intervals: CPU/Memory (1-10s), Backup/NixOS (30-60s), Disk/Systemd (60-300s)
All intervals now configurable via NixOS config:
- collectors.*.interval_seconds (collection frequency per collector)
- collectors.*.command_timeout_seconds (timeout for shell commands)
- notifications.check_interval_seconds (status change detection rate)
Command timeouts increased from hardcoded 2-3s to configurable 10-30s:
- Disk collector: 30s (SMART operations, lsblk)
- Systemd collector: 15s (systemctl, docker, du commands)
- Network collector: 10s (ip route, ip addr)
Benefits:
- No false "offline" alerts when slow collectors take >10s
- Different update rates for different metric types
- Better resource management with longer timeouts
- Full NixOS configuration control
Bump version to v0.1.193
Add architectural plan for separating ZMQ sending from data collection to prevent false 'host offline' alerts caused by slow collectors.
Key concepts:
- Shared cache (Arc<RwLock<AgentData>>)
- Independent async collector tasks with different update rates
- ZMQ sender always sends every 1s from cache
- Fast collectors (1s), medium (5s), slow (60s)
- No blocking regardless of collector speed
Emoji rendering in terminals can be very slow, especially when rendered in the hot path (every frame for every docker image). The whale emoji 🐋 was causing significant rendering delays.
Temporary change to ASCII 'D' to test if emoji was the performance issue.
Docker images now display with distinctive 🐋 whale icon in blue (highlight color) instead of status icons. This provides clear visual identification that these are docker images while not implying operational status.
Changes:
- Rename docker images from 'image_node:18...' to 'I node:18...' for conciseness
- Change image status from 'active' to 'inactive' for neutral informational display
- Images now show with gray empty circle ○ instead of green filled circle ●
Docker images are static artifacts without meaningful operational status, so using inactive status provides neutral gray display that won't trigger alerts or affect service status aggregation.
Fixes random host disconnections caused by blocking operations preventing timely ZMQ packet transmission.
Changes:
- Add run_command_with_timeout() wrapper using tokio for async command execution
- Apply 10s timeout to smartctl (prevents 30+ second hangs on failing drives)
- Apply 5s timeout to du, lsblk, systemctl list commands
- Apply 3s timeout to systemctl show/is-active, df, ip commands
- Apply 2s timeout to hostname command
- Use system 'timeout' command for sync operations where async not needed
Critical fixes:
- smartctl: Failing drives could block for 30+ seconds per drive
- du: Large directories (Docker, PostgreSQL) could block 10-30+ seconds
- systemctl/docker: Commands could block indefinitely during system issues
With 1-second collection interval and 10-second heartbeat timeout, any blocking operation >10s causes false "host offline" alerts. These timeouts ensure collection completes quickly even during system degradation.
Agent changes:
- Changed docker ps and docker images commands to run without sudo
- cm-agent user is already in docker group, so sudo is not needed
- Fixes "unable to change to root gid: Operation not permitted" error
- Systemd security restrictions were blocking sudo gid changes
This fixes Docker container and image collection on systems with
systemd security hardening enabled.
Updated to version 0.1.178
Agent changes:
- Log stderr output when docker images command fails
- This will show the actual error message (e.g., permission denied, docker not found)
- Helps diagnose why docker images collection is failing
Updated to version 0.1.177
Agent changes:
- Changed debug!() to info!() for Docker collection logs
- This allows logs to show with default RUST_LOG=info setting
- Added info import to tracing use statement
Now logs will be visible in journalctl without needing to change log level:
- "Collecting Docker sub-services for service: docker"
- "Found X Docker containers"
- "Found X Docker images"
- "Total Docker sub-services added: X"
Updated to version 0.1.176
Agent changes:
- Added debug logging to Docker images collection function
- Log when Docker sub-services are being collected for a service
- Log count of containers and images found
- Log total sub-services added
- Show command failure details instead of silently returning empty vec
This will help diagnose why Docker images aren't showing up as sub-services
on some hosts. The logs will show if the docker commands are failing or if
the collection is working but data isn't being transmitted properly.
Updated to version 0.1.175
