30 releases

0.18.4 Dec 3, 2023
0.18.3 Aug 20, 2023
0.18.2 Feb 18, 2023
0.18.1 Oct 16, 2022
0.11.0 Jul 19, 2020

#6 in #proc


5.5K SLoC


Rust Crates.io Crates.io

Oprs is a process monitor for Linux. The name is an abbreviation for Observe Process ReSources.

It's based on procfs.


  • Monitor memory, execution time, input/output, page fault, file descriptors, mapped memory regions.
  • Optional minimum and maximum.
  • Select processes by PID, PID file or name.
  • Display in plain text or a terminal UI.
  • Export in CSV or RRDtool format.


To get the memory size, elapsed time and page fault of a process by PID (firefox), a process by pid file (pulseaudio) and a process by name (bash), run:

oprs -F human -p 6913 -f pulseaudio.pid -n dhclient mem:vm mem:rss+max time:elapsed fault:major

Screenshot of oprs

Without argument, the command prints the available metrics.

By default, the raw figure is printed unless -raw is added: mem:rss-raw+min+max.


Without argument, the command prints the list of available metrics.

Limited patterns are allowed for metrics: by prefix mem:*, suffix :call, both io::count.

A metric may be followed by a unit. For example: mem:vm/gi

Available units:

  • ki kibi
  • mi mebi
  • gi gibi
  • ti tebi
  • k kilo
  • m mega
  • g giga
  • t tera
  • sz the best unit in k, m, g or t.
  • du format duration as hour, minutes, seconds.

Metrics can be also aggregated using +min and/or +max. For example mem:vm+max/gi prints the virtual memory size and the peak size. To get only the max, use: mem:vm-raw+max. To get all: mem:vm+min+max.

For some metrics, min or max is meaningless.

Data type

Metrics are either:

  • Counters: values that always increase like the number of read call or the CPU time.
  • Gauges: positive values that may decrease such as the memory comsumptions.

CPU usage

Unlike other tools, the CPU usage of a process displayed by time:cpu+ratio is the percentage of the total CPU time. A process using all cores of a 4-cores system would be at 100%, not 400%.

The CPU usage is (stime + utime) / ((user - guest) + (nice - guest_nice) + system + idle + iowait + irq + softirq) where stime and utime comes from /proc/PID/stat and user, … from /proc/stat.


Processes are selected using one of the four following options:

  • Option --pid: the pid of the process.

  • Option --pid-file: file containing the pid of the process. It doesn't have to exists when the command starts.

  • Option --name: monitor all processes with the given name.

  • Option --merge: sum the metrics of all the processes with the same name. The number of processes is displayed between parenthesis instead of the pid.

Options can be specified more than once.



In CSV export, the first column is the number of seconds since the Unix Epoch.

The size of exported data can be limited with --export-size to set the maximum size of a CSV file and --export-count to set the maximum number of files.

Round Robin Database (RRD)

Creates one RRD database for each process by spawning a RRDtool process. Only raw values are written in the database.

The number of rows is set with option --export-count.

Option --graph creates one png file per metric in the export directory.

Screenshot of RRD graph


Configuration file name is settings.ini. It's located according to the XDG Base Directory Specification.

Example ~/.config/oprs/settings.ini:

mode = term
every = 10
format = human
theme = light
border = yes

kind = csv
dir = /tmp
size = 10m
count = 5

file = /var/log/oprs.log
level = info

system = yes
myself = yes


Copyright (c) 2020-2021 Laurent Pelecq

oprs is distributed under the terms of the GNU General Public License version 3.

See the LICENSE GPL3 for details.


~320K SLoC