Netdata

My linux system regularly write data on disk, I don’t expect this activity, how to find which process does that? - ChatGPT

Netdata is a very common source of constant disk write

By default, netdata writes metrics to disk continuously (its internal database), even if you’re not actively viewing dashboards.

$ sudo iotop -oPa

-o → show only processes actually doing I/O
-P → show per-process (not per-thread)
-a → accumulate I/O over time

# Setup 

$ sudo systemctl restart netdata

# Netdata dashboard

# Ram only 

# /opt/netdata/netdata-configs/netdata.conf
[global]
    memory mode = ram
    
[health]
    enabled = no
    
[logging]
    debug log = none
    error log = none
    access log = none
    
[registry]
    enabled = no
    
[db]
    mode = ram

# Nvidia ⮺

Netdata consumes CPU, never VRAM.

# /opt/netdata/netdata-configs/netdata.conf
[plugins]
    nvidia_smi = yes

That [plugins] entry is dead config on netdata v2

nvidia_smi under [plugins] only loads plugins.d/nvidia_smi.sh, which no longer ships — the collector was rewritten in Go. Verified on netdata v2.8.5: no such file in plugins.d/. The GPU charts actually come from the go.d collector, configured somewhere else entirely.

The go.d job spawns a long-lived child that dumps every counter as full XML, every 5 s, forever:

$ pgrep -af nvidia-smi
netdata  2402  /usr/bin/nvidia-smi -q -x -l 5     # ppid 2000 = go.d.plugin

-q -x = full XML dump, -l 5 = loop it every 5 seconds. Much heavier than a dashboard needs.

# Fix

Override the job in /opt/netdata/etc/netdata/go.d/nvidia_smi.conf. Editing the packaged /opt/netdata/usr/lib/netdata/conf.d/go.d/nvidia_smi.conf would be lost on upgrade; the user dir wins and survives.

jobs:
  - name: nvidia_smi
    update_every: 10

go.d jobs are read at startup, so systemctl restart netdata is required — a reload is not enough.

Written on February 1, 2026, Last update on March 20, 2026
monitoring disk