Package list atop / lintian-fixes/main

Tree @lintian-fixes/main (Download .tar.gz)


Install the following packages to be able to build atop (package name
can be different depending on the Linux distro):

* zlib-devel    or libz-dev or zlib1g-dev
* ncurses-devel or libncurses5-dev/libncursesw5-dev

Install the following packages to be able to execute atop (package name
can be different depending on the Linux distro):

* zlib    or zlib1g
* ncurses or libncurses5/libncursesw5


For interactive use, it is sufficient to install ATOP with the command
(as root):

	make systemdinstall


	make sysvinstall

For automatic logging in compressed binary format, see the
description in the manual-page.

The kernel module 'netatop' can be downloaded and installed separately
This module is optional and can be used to gather network statistics
per process/thread as described in


Newer upstream kernels (e.g. 4.8 and 4.9) have two issuess
with process accounting:

1) Sometimes process accounting does not work at all¹. Atopacctd tries to
work around this issue, by retrying to initialize process accounting
several times.

2) When using the NETLINK inface, the command TASKSTATS_CMD_GET
consequently returns -EINVAL². Atopacctd needs NETLINK to be able to
be triggered that some process in the system has finished. In this way,
atopacctd can be in a blocking state as long as no processes terminate.
When atopacctd detects that NETLINK fails, it switches into a polling
mode to periodically try if it can read process accounting records as
a workaround. This issue has to do with cpumasks and you can work-around
it by building a kernel that has CONFIG_NR_CPUS configured to exactly
the amount of CPUs (logical CPUs) in the system the kernel runs on. You
can find this kernel option under "Processor type and features" -->
"Maximum number of CPUs".

[1] Bug 190271 - process accounting sometimes does not work

[2] Bug 190711 - Process accounting: Using the NETLINK interface,
the command TASKSTATS_CMD_GET returns -EINVAL

Linux kernel mailing list thread:

[REGRESSION] Two issues that prevent process accounting (taskstats) from
working correctly

Gerlof Langeveld