Performance Bug: RTKBase Web Service Creates D-Bus Message Flood (400+ messages/sec)

[kiczko]

Summary

The RTKBase web service (rtkbase_web.service) creates an excessive D-Bus message flood that can overwhelm system resources on Raspberry Pi devices, causing high CPU usage and system instability.

Environment

Device: Raspberry Pi 2 Model B OS: Raspberry Pi OS (Debian 12 bookworm) RTKBase Version: 2.6.3 Python Version: 3.11.2 D-Bus Message Bus Daemon Version: 1.14.10 Affected Component: web_app/server.py and web_app/ServiceController.py

Problem Description

The RTKBase web application polls systemd service status every second using pystemd, creating thousands of new D-Bus connections per second. This causes:

1,000-3,000+ actual D-Bus messages per second (normal is 10-50/sec) 15,000-20,000+ lines of D-Bus output per second (due to multi-line message structure) High CPU usage from dbus-daemon (40%+) and polkitd (10%+) System performance degradation on resource-constrained devices Potential system instability due to D-Bus bus saturation

Evidence

Quantitative Analysis sudo timeout 10 dbus-monitor --system | grep -E '^(method call|method return|signal|error)' | wc -l

Qualitative Analysis The D-Bus flood shows a clear pattern:

1. Rapid connection creation: Connection IDs increment from :1.4533994 to :1.4578224 in seconds
2. Repetitive service queries: Same systemd units queried continuously
3. UnitNew/UnitRemoved cycles: Services created and immediately destroyed

The attached script dbus_analysis.sh runs a comprehensive analysis.

System Impact

# htop showing high CPU usage
PID  USER      PRI  NI  VIRT   RES   SHR S CPU% MEM%   TIME+  Command
358  messagebus 20   0  8608  3936  3424 S 41.6  0.4  19:41.25 dbus-daemon --system
366  polkitd    20   0 48120  7624  6512 S 11.7  0.7   4:50.54 polkitd --no-debug

Root Cause Analysis

Primary Issue: Inefficient D-Bus Usage in ServiceController.py The ServiceController class creates new D-Bus connections for every query:

# Current problematic code
class ServiceController(object):
    def __init__(self, unit):
        self.unit = Unit(bytes(unit, 'utf-8'), _autoload=True)  # New connection each time
    
    def isActive(self):
        # Creates new D-Bus connection
        if self.unit.Unit.ActiveState == b'active':
            return True

Secondary Issue: Excessive Polling in server.py The manager() function polls services every second:

# Current problematic code in manager()
while True:
    if connected_clients > 0:
        updated_services_status = getServicesStatus(emit_pingback=False)  # Every 1 second
        # ...
    time.sleep(1)

Why This Creates a D-Bus Flood

1. 11 services × 3 D-Bus calls per service (isActive, status, get_result) × every second = 33+ new connections/sec
2. pystemd library doesn't reuse connections efficiently
3. Each connection requires multiple D-Bus messages (Hello, NameAcquired, property queries, NameLost)
4. Broken / misconfigured services cause additional UnitNew/UnitRemoved cycles

dbus_analysis.sh.txt

[kiczko]

Proposed Solution

1. Implement Connection Reuse and Caching in ServiceController.py Replace the current ServiceController.py with an improved version that:

• Reuses D-Bus connections via shared Manager instance
• Caches service status for 5 seconds to eliminate redundant queries
• Handles errors gracefully with safe defaults

2. Reduce Polling Frequency in server.py Modify the manager() function to:

• Check services every 10 seconds instead of every second
• Send system info every 2 seconds (separate from service checks)
• Clear cache after service state changes

3. Add Better Error Handling

• Graceful fallbacks for problematic services
• Proper exception handling in service queries
• Safe defaults when services are unavailable

Files to Modify:

1. web_app/ServiceController.py - Complete rewrite with connection pooling
2. web_app/server.py - Update manager() and getServicesStatus() functions

Will create a corresponding pull request.

[Stefal]

Thank you for this in depth report.

PR is welcome.

2. Reduce Polling Frequency in server.py Modify the manager() function to:

* Check services every 10 seconds instead of every second

But, I will keep the service check every second, because I don't want to confuse the end user. And don't forget that these calls with pystemd are stopped when they are no user connected to the web interface.