Difference between revisions of "Processes scope and visibility"

From OpenVZ Virtuozzo Containers Wiki
Jump to: navigation, search
("Poor man's vzps in bash")
(fix using template:warning)
 
(4 intermediate revisions by 3 users not shown)
Line 6: Line 6:
 
From [[CT0]] one can see all the processes running on the system; that
 
From [[CT0]] one can see all the processes running on the system; that
 
includes all the processes of all [[container]]s and the processes of the
 
includes all the processes of all [[container]]s and the processes of the
[[host system]] itself. Sometimes you just want to see the processes from the
+
[[host system]] itself. Sometimes you just want to see the processes from the host system only. Sometimes you just want to see the processes from a
host system only. Sometimes you just want to see the processes from a
 
 
particular container.
 
particular container.
  
Line 13: Line 12:
  
 
== Solutions ==
 
== Solutions ==
 +
 +
=== Hide container processes from host completely ===
 +
 +
It is possible to hide other CT's processes from [[CT0]]. For this just enable kernel.pid_ns_hide_child sysctl parameter:
 +
 +
<pre>
 +
sysctl -w 'kernel.pid_ns_hide_child=1'
 +
</pre>
 +
 +
and restart all containers. To make setting permanent put into /etc/sysctl.conf following line:
 +
 +
<pre>
 +
kernel.pid_ns_hide_child=1
 +
</pre>
 +
 +
After this ps or htop or top will not show other container processes.
 +
 +
{{Warning|If you use checkpointing and/or live migration, note they are not compatible with this feature and will stop working.}}
  
 
=== "Poor man's vzps in bash" ===
 
=== "Poor man's vzps in bash" ===
Line 65: Line 82:
 
Take <code>vzprocps</code> tools from http://download.openvz.org/contrib/utils/.
 
Take <code>vzprocps</code> tools from http://download.openvz.org/contrib/utils/.
 
These are usual <code>ps</code> and <code>top</code> utilities (named <code>vztop</code> and <code>vzps</code> to not conflict with the standard ones) with an <code>-E</code> option added. You can use <code>-E <i>CTID</i></code> option to limit the output to the selected CTID (use 0 for the host system), or just <code>-E</code> without an argument to just add CTID column to output.
 
These are usual <code>ps</code> and <code>top</code> utilities (named <code>vztop</code> and <code>vzps</code> to not conflict with the standard ones) with an <code>-E</code> option added. You can use <code>-E <i>CTID</i></code> option to limit the output to the selected CTID (use 0 for the host system), or just <code>-E</code> without an argument to just add CTID column to output.
 +
 +
=== Use vzprocps-perl tools ===
 +
Take <code>vzprocps-perl</code> tools from http://sourceforge.net/p/vzprocpsperl/wiki/vzprocps-perl/.
 +
Write in Perl with basics functions.
 +
Can be used in x86_64 architecture.
  
 
== See also ==
 
== See also ==

Latest revision as of 20:41, 29 May 2013

This HOWTO shows how OpenVZ hardware node administrator can see a processes belonging to the host system only, or to a particular container.

Problem[edit]

From CT0 one can see all the processes running on the system; that includes all the processes of all containers and the processes of the host system itself. Sometimes you just want to see the processes from the host system only. Sometimes you just want to see the processes from a particular container.

There are many ways to achieve it.

Solutions[edit]

Hide container processes from host completely[edit]

It is possible to hide other CT's processes from CT0. For this just enable kernel.pid_ns_hide_child sysctl parameter:

sysctl -w 'kernel.pid_ns_hide_child=1'

and restart all containers. To make setting permanent put into /etc/sysctl.conf following line:

kernel.pid_ns_hide_child=1

After this ps or htop or top will not show other container processes.

Warning.svg Warning: If you use checkpointing and/or live migration, note they are not compatible with this feature and will stop working.

"Poor man's vzps in bash"[edit]

Use the following script by aistis, broken by Kir, fixed by Hvdkamer.

First argument is CT ID (0 for the host system), all the remaining arguments are passed to ps(1) utility.

#!/bin/bash
# Usage: ./ovzps CTID [ps flags ...]

function find_container_pids(){
       local pid
       local myctid=$1
       local ctpids=

       for pid in $ALLPIDS; do
               [ -f /proc/$pid/status ] || continue
               ctid=`grep envID /proc/$pid/status | awk -F: '{print $2}'`
               if [ ${ctid} = ${myctid} ]; then
                       ctpids="$ctpids $pid"
               fi
       done
       echo "$ctpids"
}

ALLPIDS=`ps -A -o pid --no-headers`
CTPIDS=`find_container_pids $1`
shift

if [ -n "${CTPIDS}" ]; then
        ps $* -p $CTPIDS
else
        exit 0
fi

A faster version:

#! /bin/bash
# Usage: ovzps <CTID> [ps flags ...]

ctid=${1:-0}
shift

ps $* -p $(grep -l "^envID:[[:space:]]*$ctid\$" /proc/[0-9]*/status | 
	sed -e 's=/proc/\([0-9]*\)/.*=\1=')

Use vzprocps tools[edit]

Take vzprocps tools from http://download.openvz.org/contrib/utils/. These are usual ps and top utilities (named vztop and vzps to not conflict with the standard ones) with an -E option added. You can use -E CTID option to limit the output to the selected CTID (use 0 for the host system), or just -E without an argument to just add CTID column to output.

Use vzprocps-perl tools[edit]

Take vzprocps-perl tools from http://sourceforge.net/p/vzprocpsperl/wiki/vzprocps-perl/. Write in Perl with basics functions. Can be used in x86_64 architecture.

See also[edit]