Linux > #bash Scripting Help

Tommis · 09-12-2007, 06:09 AM

Hello,

First off i would like to say that i am not a linux newbie, been using it for 2-3 years now mostly console based. I love scripting in bash, making my own scripts for my server, but i am quite stuck on this one. I am making a #bash script that runs on a sleep timer every 5 seconds and my problem is acually getting the information (checking). I want it to check the "top -b -n 1" (to get one list of the top command) and to see if there is anything using over 80% cpu or 80% memory and if so, kill it. I know this is possible using the grep command, but i am very puzzeled by this :P

How would i get it to only say the PID of things that are using over 80% cpu or 80% memory? "top -b -n 1 | grep ?" Any help would be appreciated! Thanks again!

Cya,
Tommis

jschiwal · 09-12-2007, 06:41 AM

Awk would work better at extracting certain fields. Look in the top manpage. There is an option to output to a program for file, so you don't have the header.

You might want to look at other programs such as "ps". Using an output format you can display what you want.

What kind of server is this? If the program using up the CPU cycles is important you may not want to kill it. Doing so blindly may not be a good idea.

bigrigdriver · 09-12-2007, 06:47 AM

Since grep doesn't do numerical value comparisons such as less than, equal to, or greater than, you will have a tough time getting it to work. This sounds like it is taylor-made for bash and awk.

In your script, use bash to run the top command and pipe the output to awk. Use awk to inspect the contents of the %CPU and %MEM fields (in awk you can do it with field numbers where $1 would be the PID column from top output, and cpu and mem are something like fileds 9 and 10, on my system an any rate). Using greater than comparison, if either cpu or mem exceeds 80%, pass the PID back to bash to kill the process.

You can download a copy of "Effective AWK Programming" for more info.

Tommis · 09-12-2007, 07:19 AM

Hey,

Thanks for the help. Ye, never used awk before, but it sounds pretty usefull in this situation. ALso, this is a Centos 4.4 server, running just some game servers where you can script IN them. This means that if a client scripts a gamemode and does some sortof mistake, i just want it to close it. kill the pid. There are only a couple names for hte application that i want to be checked though, i probably could do that with awk. Well, anyhow, when i get back from school ill check it out. Thanks again for the help.

Cya,
Tommis

wjevans_7d1@yahoo.co · 09-12-2007, 07:48 AM

Quoth the highly esteemed jschiwal:

Quote:

Look in the top manpage. There is an option to output to a program for file, so you don't have the header.

I'm running an ancient distribution whose top doesn't have that option. If that's true for you, it will complicate things slightly.

Does the following script work for you? It's designed to be portable and paranoid, so you don't have to worry about which field is the %CPU field. It also skips past the header, so it will run on systems whose top always outputs the header. The sed bit near the end is there to get around the annoyance that sometimes "<" appears in top's output surrounded by spaces, so that it qualifies (from awk's viewpoint) as an extra field and throws everything off.

I have tested this, but you may wish to also.

Please be patient. This is my first non-trivial use of awk. Normally I'd use Perl.

Code:

#!/bin/bash

sleep_interval=5

threshold=90

field_name='%CPU'

# Get the field number within the line of the field of interest.

top -b -n 1 > tmp.$$

# Get the number of fields.

nf=$(awk '/PID/{print NF}' tmp.$$)

# Get the relevant field number.

fn=0

field_number=0

while [[ $fn -lt $nf ]]
do
  fn=$(($fn+1))
  if [ $(awk "/PID/{print \$$fn}" tmp.$$) = "$field_name" ]
  then
    field_number=$fn
  fi
done

rm tmp.$$

if [[ $field_number = "0" ]]
then
  echo field $field_name is undefined
  exit 1
fi

while true
do
  bad_guys=$(top -b -n 1 | sed -e 's/<//g' | awk "/PID/{x=1};x>0{x++};x>2&&\$$field_number>=$threshold{print \$1}")

  if [[ "$bad_guys" != "" ]]
  then
    for one_guy in $bad_guys
    do
      kill -9 $one_guy
    done
  fi

  sleep $sleep_interval
done

Tommis · 09-12-2007, 03:57 PM

Hey,

Bad news, it outputs how to use awk first, then outputs how to use gawk. After that it says "./monitor: line 44: /PID/{x=1};x>0{x++};x>2&&$9>=90{print $1}: No such file or directory" and quits. It then restarts every 5 seconds. Dam :P

Cya,
Tommis

wjevans_7d1@yahoo.co · 09-13-2007, 05:48 AM

Please don't be angry with me about what I'm about to say, but:

Did you do a cut and paste of the code I suggested,, or did you enter the code by hand?

And ...

Could you cut and paste the code back here the code exactly as it was run?

(If you're pissed at these questions, I'll understand.)

Tommis · 09-13-2007, 06:41 AM

Hey,

Nah man, why would i get angry, i appreciate your help very much. And i did cut and paste into a document opened with nano through ssh, but i made sure i have everything. This is what i ran CHMODDED to 777..

Code:

#!/bin/bash

sleep_interval=5

threshold=90

field_name='%CPU'

# Get the field number within the line of the field of interest.

top -b -n 1 > tmp.$$

# Get the number of fields.

nf=$(awk '/PID/{print NF}' tmp.$$)

# Get the relevant field number.

fn=0

field_number=0

while [[ $fn -lt $nf ]]
do
  fn=$(($fn+1))
  if [ $(awk "/PID/{print \$$fn}" tmp.$$) = "$field_name" ]
  then
    field_number=$fn
  fi
done

rm tmp.$$

if [[ $field_number = "0" ]]
then
  echo field $field_name is undefined
  exit 1
fi

while true
do
  bad_guys=$(top -b -n 1 | sed -e 's/<//g' | awk
"/PID/{x=1};x>0{x++};x>2&&\$$field_number>=$threshold{print \$1}")

  if [[ "$bad_guys" != "" ]]
  then
    for one_guy in $bad_guys
    do
      kill -9 $one_guy
    done
  fi

  sleep $sleep_interval
done

wjevans_7d1@yahoo.co · 09-14-2007, 07:03 AM

Quoth Tommis:

Quote:

Nah man, why would i get angry

There are people out there with awfully thin skins. ("Of course I copied and pasted your code accurately! GRRRRRRRRR!")

I'm glad you're not one of them!

But I see the problem. One of the lines got split into two.

At approximately line 42, there should be this long line:

Code:

  bad_guys=$(top -b -n 1 | sed -e 's/<//g' | awk "/PID/{x=1};x>0{x++};x>2&&\$$field_number>=$threshold{print \$1}")

Your copy of the code has split that line into two:

Code:

  bad_guys=$(top -b -n 1 | sed -e 's/<//g' | awk
"/PID/{x=1};x>0{x++};x>2&&\$$field_number>=$threshold{print \$1}")

I'm guessing that nano has some sort of line wrap option.

Tommis · 09-15-2007, 06:49 AM

Hey,

Yes, my bad, nano does that very often, argh ;p Now it is copied correctly, and i get this error,

"

[root@GERMANY home]# ./monitor
: bad interpreter: No such file or directory

"

(It is copied correctly now)

Cya,
Tommis

wjevans_7d1@yahoo.co · 09-16-2007, 06:02 PM

If that's the error you're getting, then at the shell prompt, do this please:

Code:

od -t -x1 thenameofyourscript | head

and post the results back here.