Prev: Mounting a hard drive from another computer
Next: Writing shell scripts to automate
From: Dan Stromberg on 5 Apr 2008 00:29
On Fri, 04 Apr 2008 02:51:10 +0200, Martin Jørgensen wrote:

> Martin Jørgensen wrote:
>> Martin Jørgensen wrote:
>>> Dear all,
>>>
>>> I'm using a linux pc which runs some large calculations that take
>>> about 1-2 days. I would like to use a script that every fifth minute
>>> checks if the cpu-usage is above, say 30% (or whatever). If it aint
>>> above 30% I assume that perhaps the calculation is done. Once done, I
>>> want the script to send me an e-mail using standard linux commands.
>>> However, to be absolutely sure the computation is done the cpu-usage
>>> for that process should be above 30% twice in a row (at polls 2*5
>>> minutes).
>>
>> Oops, .... should be _below_ 30% twice in a row, ofcourse... Then I
>> assume the computation is done...
>>
>>> Is this possible? How? I'm pretty bad at scripting but know a few
>>> basics. I was wondering about using "top" or similar. Ofcourse I would
>>> need the program to automatically figure out the PID. But I guess I
>>> can use some kind of "ps | grep -i "program_name" "...
>>
>> All script suggestions are welcome... I hope I don't need root access
>> (else I might talk to the sysadm). System: linux running redhat
>> (something - about 1-1,5 year old installation I think).
>
> Erh, sorry. About the mail-thing:
>
> I just googled a bit around and I think a mail-server should be
> installed which I don't think is the case. However:
>
> Is it possible to automate the following?
>
> -------
> $ ssh -X mylogin(a)myother_server.dk
> (password entered)
> (wait about 10 seconds)
> $ mail my_email(a)address.com
> Hi there, your computation using "program_name" is finished. Thank you
> and goodbye.
> .
> (mail should now be sended)
> $ exit
> ------
>
> Server should now respond with:
> logout
> Connection to myother_server.dk closed.
>
>
> I hope the above mail-thing is possible to automate....
>
>
> Best regards
> Martin Jørgensen

It's a simple program, but I use it a lot so I can run off and work on
task n+1 while as I'm waiting for tasks 1..n to finish:

http://stromberg.dnsalias.org/~strombrg/notify-when-up2.html

It'll do the e-mail, give you a little X11 popup window, and handle the
ssh stuff for you if you set up passwordless authentication:

http://stromberg.dnsalias.org/~strombrg/ssh-keys.html

From: Dan Stromberg on 5 Apr 2008 00:46
On Sat, 05 Apr 2008 00:33:58 +0200, Martin Jørgensen wrote:

> Janis wrote:
>> On 4 Apr., 02:38, Martin Jørgensen <megaf...(a)hotmail.com> wrote:
>>> Dear all,
>>>
>>> I'm using a linux pc which runs some large calculations that take
>>> about 1-2 days. I would like to use a script that every fifth minute
>>> checks if the cpu-usage is above, say 30% (or whatever). If it aint
>>> above 30% I assume that perhaps the calculation is done. Once done, I
>>> want the script to send me an e-mail using standard linux commands.
>>> However, to be absolutely sure the computation is done the cpu-usage
>>> for that process should be above 30% twice in a row (at polls 2*5
>>> minutes).
>>
>> Polling is a bad approach and usually unnecessary. Moreover the CPU
>> load doesn't really tell you anything in any reliable way.
>
> What is a better alternative?
>
>>> Is this possible? How? I'm pretty bad at scripting but know a few
>>> basics. I was wondering about using "top" or similar. Ofcourse I would
>>> need the program to automatically figure out the PID. But I guess I
>>> can use some kind of "ps | grep -i "program_name" "...
>>
>> It's of course possible to intercept the PID and check that.
>>
>> But given the small time scale you posted above I suppose you don't
>> need that to be informed once about the actual job but rather for
>> future invocations? If that is the case you may try
>>
>> program_name program_args ; mailx -s finished you(a)your.domain
>
> Sorry, that doesn't work.
>
> The computation is started from inside a GUI and I cannot make the
> program exit automatically after the simulation is done....
>
> Also: I believe "mailx" requires a mail-server to be installed? I added
> something about ssh'ing in my last posting, AFAIR.
>
>> You can provide a text body to the mail command through standard input.
>> Or even pipe the whole output of your program to the mail command
>> (beware if the output is huge)
>>
>> program_name program_args 2>&1 | mailx -s ...
>
> Thank you, but that doesn't work for me since the simulation is started
> from a GUI and the simulation program is not text/console-based...
>
>> If you want your prompt back after invocation start that program
>> sequence within a subshell as a background process
>>
>> ( ... )&
>>
>> In case you insist on polling because your program is already running
>> something like the following can be done
>>
>> while ps ... | grep -s known_pid
>> do sleep 600 # 5 minutes
>> done
>
> I'm not really sure I understand what to do with this. What do I insert
> at the 3 dots?
>
> I need to check when the CPU falls below 30% twice in a row, as I wrote.
> You're assuming my simulation program is text/console-based, which it
> isn't.
>
> The process will not kill itself after simulation is done.
>
>> But mind that you typically cannot rely on the ps output format; see
>> whether it supports option -o to make the grep more reliable.
>
> Well, thank you for trying even though I really think you misunderstood
> the question...
>
> Again: Since the simulations are started from a GUI I don't believe I
> can use most of your suggestions...
>
> I welcome any other suggestions anyone has... I hope somebody here knows
> how to solve the problem and wants to share the knowledge about how to
> do it.
>
>
>
> Best regards
> Martin Jørgensen

This being a simulation that doesn't have a pid that disappears at the
end, you're probably best off checking how much CPU time is being
accumulated (whether by ps or top), or perhaps using strace and counting
syscalls.

I have lots of notify-when-up (the old version of the program which
doesn't do ssh for you) examples at http://stromberg.dnsalias.org/
~strombrg/notify-when-up.html including checking for when something drops
out of top for a while. The usage is mostly the same between notify-when-
up and notify-when-up2.

Machine polling isn't great typically, but it's generally better than
human polling :) notify-when-up2 and notify-when-up are both pollers
with most of their options. I used to avoid machine polling on some sort
of principle, but my life got more productive when I got over it.

From: Maxwell Lol on 7 Apr 2008 12:14
Janis <janis_papanagnou(a)hotmail.com> writes:

> Hmm.. - I've not seen a Linux box without programs mail or mailx
> installed, and sendmail running. Sorry to hear that. But then, how
> do you think your requirement to send a mail can be be fulfilled at
> all if you have no mail system available?


Well, you can telnet to the SMTP port and send text.
From: Janis on 8 Apr 2008 07:13
On 7 Apr., 18:14, Maxwell Lol <nos...(a)com.invalid> wrote:
> Janis <janis_papanag...(a)hotmail.com> writes:
> > Hmm.. - I've not seen a Linux box without programs mail or mailx
> > installed, and sendmail running. Sorry to hear that. But then, how
> > do you think your requirement to send a mail can be be fulfilled at
> > all if you have no mail system available?
>
> Well, you can telnet to the SMTP port and send text.

Good point. (I often forget about that; it's just so rare
that basic programs are missing on a system which makes it
necesary to resort to a lower-level generic telnet solution.)

Janis
 | 
Pages: 1
Prev: Mounting a hard drive from another computer
Next: Writing shell scripts to automate