Is it normal that iotop consumes 98% CPU when running? It completely hogs one CPU core.

I'm using version 1.24.

I'd like to save my defaults for which column to sort, which columns to show or whether to show idle processes. This also opens the path to additional settings/preferences not everyone might like.

wouldn't it be useful to have iotop monitor iops instead of read/write throughput, too ?
(to be able to determine which process is an iops hog , which totally matters, especially with ordinary hdd)

from linux process info, this should be available (syscr/syscw)

cat /proc/2577/io

rchar: 231362397403320
wchar: 60299521079984
syscr: 8486645208
syscw: 2586473645
read_bytes: 191716084186159
write_bytes: 30517886778390
cancelled_write_bytes: 726028288

rchar: number of bytes the process read, using any read-like system call (from files, pipes, tty...).
wchar: number of bytes the process wrote using any write-like system call.
syscr: number of read-like system call invocations that the process performed.
syscr: number of write-like system call invocations that the process performed.
read_bytes: number of bytes the process directly read from disk.
write_bytes: number of bytes the process originally dirtied in the page-cache (assuming they will go to disk later).
cancelled_write_bytes: number of bytes the process "un-dirtied" - e.g. using an "ftruncate" call that truncated pages from the page-cache.  

http://www.haifux.org/lectures/254/alice_and_bob_in_io_land/node51.html

When running iotop on the Linux console (and, I assume, any terminal without braille support), we are limited to the characters provided in the console font - that somewhat restricts what can be displayed - the graphs show as diamonds. The Linux console, and most VT-like terminals have, however, some semigraphic mosaic characters that can be used to convey the same kind of information.

One such set is the scan 9-7-5-3-1 lines, ⎽⎼─⎻⎺, which can show one value per column (but 5 values rather than 4). Most physical terminals support these characters as well.

Linux console fonts can have 256 or 512 glyphs, as well as an extra Unicode table for more. With that, it'd be possible to support Braille and other symbols on the graphical console - this being a matter of font support for those. Unfortunately no font I tried bundled with Ubuntu, Debian, and Fedora did support them (I'll make feature requests for those on their respective bug trackers).

In the meantime, it'd be worth adding a theme command-line switch to select different graph character options.

With that done, we can start exploring different sets, such as the ITU T.100 set of "smoothed mosaic", which are in Unicode between U+1FB3C to U+1FB67. They offer 2 points at 4 heights per character cell, as a filled polygon, or a mix of fill rates (the 0-25-50% set from CP 437, which is commonly supported) and colors (as done in Sysdig's spectrogram display)

> docker run -it --rm registry.access.redhat.com/ubi8/ubi:8.4

# This doesn't work
# yum install epel-release -y
Updating Subscription Management repositories.
Unable to read consumer identity

This system is not registered to Red Hat Subscription Management. You can use subscription-manager to register.

Red Hat Universal Base Image 8 (RPMs) - BaseOS                                          192 kB/s | 798 kB     00:04
Red Hat Universal Base Image 8 (RPMs) - AppStream                                       2.8 MB/s | 2.6 MB     00:00
Red Hat Universal Base Image 8 (RPMs) - CodeReady Builder                                35 kB/s |  17 kB     00:00
No match for argument: epel-release
Error: Unable to find a match: epel-release

# yum -y install https://dl.fedoraproject.org/pub/epel/epel-release-latest-8.noarch.rpm
# yum install iotop-c -y
# iotop-c

or

> docker run -it --rm fedora

# yum install iotop-c -y
# iotop-c

# iotop-c --version
iotop-c 1.20

Currently, when you have multiple MiB/s of writes, you will often see values like 8000.00 K/s as it only switches to the M prefix in the 10s of MiB/s.

The original python iotop switches somewhere in the neighbourhood of 2048K which I personally find easier to interpret and more aesthetically pleasing.

I get a corrupt "graph" column display. There are also what appear to be unhandled terminal control sequences displayed at the end of the "title" line, and in front of some of the command names.
The xterm is with TERM=xterm-256color. and the curses library is ncurses-6.4

Testing iotop – using commit c10e657 (Update contributors) – there is a black bar in the rsync row, but the usage is shown as 0. What does that mean?

hi,
it's a very useful top, but lack a very important feature to be able to filter for a given partition or disk.
eg. i'd like to know hoe much data written a service. BUT suppose it write logs, db, user data etc. and I also would like to know how much data written into my system disk or root partition.
so a command line argument eg. / or /dev/sda2 would be very useful!
is it possible to add such a feature?

It would be a useful enhancement to add colours and even themes to the displayed output. Colours make highlighting or identifying things easier at a glance. A monochrome only switch may also be needed for some older or more basic terminals that don't support ansi colour codes.

It might be also nice to be able to customise the characters in the graph column to use Unicode block style characters instead of dot style ones.

Sometimes when I start iotop-c -o when a process is already running, that process will not be shown. It does show a Total DISK WRITE including that process' IO and a few btrfs threads but their small IO doesn't add up to the total.

This does not happen with the python iotop.

The issue is very spurious and hard to reproduce.

i want to use it in development board

I'm used to having graphs grow from right to left from many other tools including i.e. btop.

It feels more intuitive for me to have time go from left-to-right rather than right-to-left once the graph is written out.

It'd be great if this was configurable #47

Why is there no option for iotop for monitoring directory io? (thx)

I made a busy reading of disk in clickhouse, and notice the GRAPH[IO]▽ column in the view is quite different with the IO> column in iotop.

The main thread of clickhouse shows the GRAPH[IO] is 100% and IO> is zero.
Which I think iotop is correct.

Because the IO mainly happens in a background thread, the main thread should be idle.

I see the data is fetched from netlink instead of /proc; but the algorithm is similar.
I don't understand the result, is there something I missed that GRAPH[IO] is not equivalent to IO> ?

FYI when following repo instructions for install on debian:

make -j

gives:

cc: error: unrecognized command-line option ‘-mno-stackrealign’; did you mean ‘-mno-strict-align’

When using the --ascii option, the per-process graphs appear to be trying to express two values per character cell, unlike the graphs on top, which express one value per cell when not in unicode. This is observable in the master branch.

I see this:

And that sounds a bit strange to me. Does the tool itself read so many data from the disk?

Use with screen will cause alignment problems, a poison for patients with obsessive-compulsive disorder
this normal

and with screen

Anyway, thank the author for this good work。
the python version authorization for non-root operation is too troublesome

No executable or binary in the final package, just package meta files.

ref: https://forum.manjaro.org/t/config-task-delay-acct-error-in-iotop-kernel-5-14/84466

kernel config: https://gitlab.manjaro.org/packages/core/linux514/-/blob/master/config#L141
CONFIG_TASK_DELAY_ACCT=y

Using kernel 5.14.7-2-MANJARO
iotop shows an error

CONFIG_TASK_DELAY_ACCT not enabled in kernel, cannot determine SWAPIN and IO %

now showing SWAPIN column but flag is enabled.

Possibility to specify -p multiple times would be really nice.

I see on

I'd like if this tool would store all the occurrences of -p, and store it in a linked list or in an array, and walk that array/ll in the filter1 function.

This is a suggestion: Is it possible to change iotop to work as a C/C++ library? This would allow developer to add this feature on other aplications, like monitoring systems.

When building with clang for 32-bit arm:

src/xxxid_info.c:103:52: warning: comparison of integers of different signs: '__u32' (aka 'unsigned int') and 'ssize_t' (aka 'int') [-Wsign-compare]
        if (answ.n.nlmsg_type==NLMSG_ERROR||(rep_len<0)||!NLMSG_OK((&answ.n),rep_len))
                                                          ^~~~~~~~~~~~~~~~~~~~~~~~~~~
/usr/include/linux/netlink.h:100:24: note: expanded from macro 'NLMSG_OK'
                           (nlh)->nlmsg_len <= (len))
                           ~~~~~~~~~~~~~~~~ ^   ~~~
src/xxxid_info.c:157:38: warning: comparison of integers of different signs: '__u32' (aka 'unsigned int') and 'ssize_t' (aka 'int') [-Wsign-compare]
        if (msg.n.nlmsg_type==NLMSG_ERROR||!NLMSG_OK((&msg.n),rv)) {
                                            ^~~~~~~~~~~~~~~~~~~~~
/usr/include/linux/netlink.h:100:24: note: expanded from macro 'NLMSG_OK'
                           (nlh)->nlmsg_len <= (len))
                           ~~~~~~~~~~~~~~~~ ^   ~~~
2 warnings generated.

Besides 'accumulated', it would be nice to have a view that would present the amount of IO per unit of time (depending on how long iotop or process is running).

I tend to run iotop in -Pa mode in a background terminal, but after a while I forget how long is it since I started iotop, which makes it difficult to calculate how much IO a certain process had done per day.

Hi @bbonev,
Commit cea4a5e annuls the conditional cc flags of HAVEWDTI and HAVESREA.

E.g. hard code the make variable to either
HAVEWDTI = yes
or
HAVEWDTI = no

and test by
make -nB | grep -m1 'date-time'

It makes no difference.

I can see that iotop is working good on my hosts (AWS EC2), which have been configured half a year ago

But today I've tried to set up a new host (same on AWS EC2) with the same linux distro (Ubuntu 20.04) and now for some reason:

1. It asks me to launch iotop using sudo (due to vulnerability). It is ok for me to run it as sudo, no problem.
2. There was a warning CONFIG_TASK_DELAY_ACCT not enabled in kernel, cannot determine SWAPIN and IO %
   no problem, executed echo 1 > /proc/sys/kernel/task_delayacct and warning disappeared. But IOPS are still not swing

I'm confused since versions of iotop are the same 0.6 (installed via apt both) and hosts are the same (distro, instance type etc)

Even when building iotop version 1.22 from source showing me 0% IOPS

I've tried btop++ and it is showing IOPS load, same as iostat

What I'm doing wrong?

As noted in #21, SWAPIN and IO don't work if task_delayacct is not enabled.

While knowing this is important, it may not be crucial for the current use of iotop; and the message is distracting and continues to be displayed even if SWAPIN and IO are disabled, potentially overlapping useful information depending on screen width/height.

The user may also not wish to enable task_delayacct because it results in a performance overhead, or as it will be misleading, since delay for most services will only be available after they are restarted, which may be impractical for ad-hoc troubleshooting.

It would therefore be nice if it the message was only shown if either of the columns it relates to were enabled. Potentially, a brief notice could be shown on startup, and thereafter the columns would be disabled by default.

As a secondary matter, it might be worth allowing iotop to run without CONFIG_TASK_DELAY_ACCT enabled, not showing SWAPIN and IO. While I/O delay % can be useful information, just knowing reads/writes per process may be sufficient for the intended use.

root@vultr:~/iotop# make
Package ncursesw was not found in the pkg-config search path.
Perhaps you should add the directory containing `ncursesw.pc'
to the PKG_CONFIG_PATH environment variable
No package 'ncursesw' found
Package ncurses was not found in the pkg-config search path.
Perhaps you should add the directory containing `ncurses.pc'
to the PKG_CONFIG_PATH environment variable
No package 'ncurses' found
Package ncurses was not found in the pkg-config search path.
Perhaps you should add the directory containing `ncurses.pc'
to the PKG_CONFIG_PATH environment variable
No package 'ncurses' found
DE bld/xxxid_info.o
CC bld/xxxid_info.o
DE bld/checks.o
CC bld/checks.o
DE bld/view_curses.o
CC bld/view_curses.o
src/view_curses.c:24:20: fatal error: curses.h: No such file or directory
compilation terminated.
Makefile:67: recipe for target 'bld/view_curses.o' failed
make: *** [bld/view_curses.o] Error 1

src/curses.h file lost?

I am trying to run iotop on a ubuntu 22.04.3 minimal install. I think I am missing a dependency. I just have no way to figure out which one. This does work on a regular install of ubuntu.

Any help would be appreciated. Thanks

Brad

Traceback (most recent call last):
File "/usr/sbin/iotop", line 10, in
from iotop.ui import main
File "/usr/lib/python3/dist-packages/iotop/ui.py", line 46, in
from iotop.data import find_uids, TaskStatsNetlink, ProcessList, Stats
File "/usr/lib/python3/dist-packages/iotop/data.py", line 51, in
vmstat_f = VmStat()
File "/usr/lib/python3/dist-packages/iotop/vmstat.py", line 23, in init
self.vmstat = self.read()
File "/usr/lib/python3/dist-packages/iotop/vmstat.py", line 40, in read
return pgpgin, pgpgout
UnboundLocalError: local variable 'pgpgin' referenced before assignment

Would it be possible to have a brew package for this? It looks great!

Hi there, while using iotop in kubernetes pods I am getting an error as below.
Traceback (most recent call last):
File "/usr/sbin/iotop", line 17, in
main()
File "/usr/lib/python3/dist-packages/iotop/ui.py", line 737, in main
main_loop()
File "/usr/lib/python3/dist-packages/iotop/ui.py", line 727, in
main_loop = lambda: run_iotop(options)
File "/usr/lib/python3/dist-packages/iotop/ui.py", line 620, in run_iotop
return curses.wrapper(run_iotop_window, options)
File "/usr/lib/python3.7/curses/init.py", line 94, in wrapper
return func(stdscr, *args, **kwds)
File "/usr/lib/python3/dist-packages/iotop/ui.py", line 609, in run_iotop_window
taskstats_connection = TaskStatsNetlink(options)
File "/usr/lib/python3/dist-packages/iotop/data.py", line 151, in init
self.family_id = controller.get_family_id('TASKSTATS')
File "/usr/lib/python3/dist-packages/iotop/genetlink.py", line 76, in get_family_id
m = GeNlMessage.recv(self.conn)
File "/usr/lib/python3/dist-packages/iotop/genetlink.py", line 56, in recv
msg = conn.recv()
File "/usr/lib/python3/dist-packages/iotop/netlink.py", line 255, in recv
raise err
OSError: Netlink error: No such file or directory (2)

I already found out a solution for this. That is using below two configurations in k8s manifests
privileged : true
hostNetwork: true
But in our requirements it mandates to not use hostNetwork: true. So I just wanted to know why iotop demands hostNetwork : true ? And if you know any other way to run iotop without errors without using hostNetwork : true please let me know.

Thank You

I have XDG_CONFIG_HOME set to ~/.local/config instead of the default location of ~/.config.

I recently tested iotop -c and made the mistake of trying to run it as non root. After correcting that error I noticed that my home now has a ~/.config/iotop directory.

The directory is being created by iotop-c, likely because it has a hardcoded path to ~/.config instead of checking the xdg enviroment variables.

What's interesting is that iotop can no longer be run as regular user, so this should be easy to fix.

(This issue does not happen with the regular iotop btw).

Edit: The issue is on lines 53 and 23 of configfile.c.

I built this repo from source using make and running it gives a scrolling
fatal reply error, -1
The line changes color after some time and occasionally gibberish such as KB/S gets printed.

iotop enters an infinite wait on netlink, probably as a result of ignoring a previous error:

11:31:14.201289 socket(AF_NETLINK, SOCK_RAW, NETLINK_GENERIC) = 3
11:31:14.201393 bind(3, {sa_family=AF_NETLINK, nl_pid=0, nl_groups=00000000}, 12) = 0
11:31:14.201511 getpid()                = 46880
11:31:14.201610 sendto(3, {{len=36, type=nlctrl, flags=NLM_F_REQUEST, seq=0, pid=46880}, "\x03\x01\x00\x00\x0e\x00\x02\x00\x54\x41\x53\x4b\x53\x54\x41\x54\x53\x00\x00\x00"}, 36, 0, {sa_family=AF_NETLINK, nl_pid=0, nl_groups=00000000}, 12) = 36
11:31:14.202179 recvfrom(3, {{len=56, type=NLMSG_ERROR, flags=0, seq=0, pid=46880}, {error=-ENOENT, msg={{len=36, type=nlctrl, flags=NLM_F_REQUEST, seq=0, pid=46880}, "\x03\x01\x00\x00\x0e\x00\x02\x00\x54\x41\x53\x4b\x53\x54\x41\x54\x53\x00\x00\x00"}}}, 1044, 0, NULL, NULL) = 56
11:31:14.202340 rt_sigaction(SIGINT, {sa_handler=0x565546aa1060, sa_mask=[INT], sa_flags=SA_RESTORER|SA_RESTART, sa_restorer=0x7fc8a8c1cd60}, {sa_handler=SIG_DFL, sa_mask=[], sa_flags=0}, 8) = 0
11:31:14.202472 openat(AT_FDCWD, "/proc", O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_DIRECTORY) = 4
11:31:14.202597 fstat(4, {st_mode=S_IFDIR|0555, st_size=0, ...}) = 0
11:31:14.202738 getdents64(4, 0x56554726c380 /* 90 entries */, 32768) = 2632
11:31:14.202978 openat(AT_FDCWD, "/proc/1/task", O_RDONLY|O_NONBLOCK|O_CLOEXEC|O_DIRECTORY) = 5
11:31:14.203094 fstat(5, {st_mode=S_IFDIR|0555, st_size=0, ...}) = 0
11:31:14.203202 getdents64(5, 0x5655472743c0 /* 3 entries */, 32768) = 72
11:31:14.203309 sendto(3, {{len=28, type=0 /* NLMSG_??? */, flags=NLM_F_REQUEST, seq=0, pid=1}, "\x01\x01\x00\x00\x08\x00\x01\x00\x01\x00\x00\x00"}, 28, 0, {sa_family=AF_NETLINK, nl_pid=0, nl_groups=00000000}, 12) = 28
11:31:14.203458 recvfrom(3,   …

This is an lxc container and it may restrict access to various things around. Unhandled errors may result in weird behaviours.

Hello there,

From the description of the project (I didn't delve deep into the code myself), it seems that this tool (as opposed to the Python version) assumes there is a Linux kernel running beneath the OS/POSIX layer. This may explain why no references are made to FreeBSD, OpenBSD, NetBSD etc. on the README.md.

I did a quick attempt to compile it under macOS/Darwin (which is vaguely BSD-ish); however, it seems that clang (the default compiler by Apple) and gcc 12.1 (installed through Homebrew) get utterly confused when trying to include something as simple as src/iotop.h — mostly because of the way #define is working:

In file included from /Library/Developer/CommandLineTools/SDKs/MacOSX11.sdk/usr/include/sys/types.h:75,
                 from src/iotop.h:27,
                 from src/arr.c:14:
/Library/Developer/CommandLineTools/SDKs/MacOSX11.sdk/usr/include/sys/cdefs.h:681:49: error: operator '&&' has no right operand
  681 | #if defined(_POSIX_C_SOURCE) && _POSIX_C_SOURCE == 1L
      |                                                 ^~
/Library/Developer/CommandLineTools/SDKs/MacOSX11.sdk/usr/include/sys/cdefs.h:687:49: error: operator '&&' has no right operand
  687 | #if defined(_POSIX_C_SOURCE) && _POSIX_C_SOURCE == 2L
      |                                                 ^~
In file included from /Library/Developer/CommandLineTools/SDKs/MacOSX11.sdk/usr/include/sys/cdefs.h:715:
/Library/Developer/CommandLineTools/SDKs/MacOSX11.sdk/usr/include/sys/_posix_availability.h:32:79:error: operator '&&' has no right operand
   32 | #if !defined(_DARWIN_C_SOURCE) && defined(_POSIX_C_SOURCE) && _POSIX_C_SOURCE >= 198808L
      |                                                                               ^~
/Library/Developer/CommandLineTools/SDKs/MacOSX11.sdk/usr/include/sys/_posix_availability.h:38:79:error: operator '&&' has no right operand
   38 | #if !defined(_DARWIN_C_SOURCE) && defined(_POSIX_C_SOURCE) && _POSIX_C_SOURCE >= 199009L
      |                                                                               ^~
/Library/Developer/CommandLineTools/SDKs/MacOSX11.sdk/usr/include/sys/_posix_availability.h:44:79:error: operator '&&' has no right operand
   44 | #if !defined(_DARWIN_C_SOURCE) && defined(_POSIX_C_SOURCE) && _POSIX_C_SOURCE >= 199209L
      |
[...]

My assumption is that the macOS/Darwin pre-processor has serious issues, after expanding the #defines, to deal with more complex expressions. This may perhaps be a result of an extra flag or a missing one; being an utter newbie to the complexities of compiling C/C++ (anything except the most basic things!), I wonder if anyone has a clue on what I ought to do to fix the above errors...

tried compile it under WSL2 ubuntu, display very odds.

i'm getting this error when trying to build on centos7

make

DE bld/delayacct.o
cc: error: unrecognized command line option '-Wdate-time'
make: *** [bld/delayacct.o] Error 1

after removing -Wdate-time for testing, i'm getting

CC bld/view_curses.o
In file included from src/view_curses.c:26:0:
src/view_curses.c: In function 'view_curses':
src/view_curses.c:964:30: error: 'A_ITALIC' undeclared (first use in this function)
attron(config.f.nocolor?A_ITALIC:COLOR_PAIR(RED_PAIR));
^
src/view_curses.c:964:30: note: each undeclared identifier is reported only once for each function it appears in
make: *** [bld/view_curses.o] Error 1

IO column is always zero, so

-o, --only only show processes or threads actually doing I/O

doesn't work.