High CPU Usage: Conduit binary/wrapper ignoring resource limits

[ab-a]

Description

I am experiencing unexpectedly high CPU usage on my Conduit instance. Despite setting a resource limit of 0.2 cores per container, the host system reports a conduit process (running via bash) consuming nearly 100% of a core.

Environment

• OS: Ubuntu 24.04.3
• Conduit Version: Latest 1.3
• Deployment: Docker 28.1.1
• System Specs: 4 CPU cores, ~24GB RAM

Observed Behavior

I have applied a CPU limit of 0.2 cores per container. While the child processes (conduit start) appear to stay within a ~20% range, the parent/wrapper process is spiking to over 90% CPU usage, often staying around 100%.

Resource Limits Applied:

═══ RESOURCE LIMITS ═══
Container    CPU Limit    Memory Limit
────────────────────────────────────────
conduit      0.2 cores    512m         
conduit-2    0.2 cores    512m         
conduit-3    0.2 cores    512m         
conduit-4    0.2 cores    512m 

Process monitor output:

ps aux --sort -%cpu | grep conduit
root     1941950 92.7  0.0  14256 10012 ?        R    Feb09 705:12 /bin/bash /usr/local/bin/conduit
opc      1892388 19.7  0.2 2578764 70296 ?       Ssl  Feb09 152:06 conduit start --max-clients 500 --bandwidth 5 --stats-file
opc      1892830 19.6  0.2 2505164 70124 ?       Ssl  Feb09 151:21 conduit start --max-clients 500 --bandwidth 5 --stats-file
opc      1892612 19.6  0.3 2505100 79988 ?       Ssl  Feb09 151:04 conduit start --max-clients 500 --bandwidth 5 --stats-file
opc      1893054 19.5  0.2 2726352 62912 ?       Ssl  Feb09 149:55 conduit start --max-clients 500 --bandwidth 5 --stats-file
root     1893204  1.4  0.0   8464  4372 ?        Ss   Feb09  11:18 /bin/bash /opt/conduit/conduit-tracker.sh
root     2860927  0.0  0.0   8464  2988 ?        S    12:36   0:00 /bin/bash /opt/conduit/conduit-tracker.sh

Steps Taken to Resolve

Applied CPUQuota=5% in /etc/systemd/system/conduit.service.

daemon-reload and restarted the service.

Result: The directive was ignored or bypassed by the binary/bash wrapper, and CPU usage remained at ~100%.

Additional Context

It appears the bash script at /usr/local/bin/conduit might be entering a tight loop or failing to pass signals correctly to the child processes, causing it to consume host resources outside of the intended container constraints.

[SamNet-dev]

Thanks for the detailed report.

The high CPU process (PID 1941950) is not a Conduit container — your Docker resource limits are working correctly (each conduit start process is at ~20%, matching your 0.2 core limit).

The 92.7% CPU process is the management script (interactive TUI menu) running without a terminal. Notice in your ps output it shows /bin/bash /usr/local/bin/conduit with no arguments — meaning it entered the interactive menu instead of conduit start. Running in a non-interactive context, the menu loop can spin.

Immediate fix — kill the stray process:

kill 1941950

Your containers will keep running normally.

Preventive fix (814084b):
Added a guard that detects non-interactive invocations and exits cleanly instead of entering the TUI. Run conduit update to get it.

About the CPUQuota in conduit.service:
The systemd service only runs conduit start --auto (starts containers then exits). The actual Conduit processes run inside Docker containers, so CPUQuota on the service won't affect them — Docker's --cpus flag (which you've already set to 0.2) is the correct way to limit container CPU, and it's working as expected.