[SOLVED] Need help on bash script process selection

[ampra2]

Hi all,

I need help on bash script to do some specific process selection.
I know that we can actually get pid from process name by using

ps -C processname -o pid=

but I don't know how to deal with my condition below:


I have some processes running on my system:

Damn, I can't past my ps result here as it keeps giving me error when I try to submit it "You are only allowed to post URLs to other sites after you have made 15 posts or more."

I need to get a script which can give me only the pid (in this case = 17005) of the process "827715".

If I used "-C bash" then it will return not only the one which I need but also another process (pid = 1704 which I don't need. If I specify "-c bash /opt/sge-6.2u4/default/spool/east-25/job_scripts/827715" it doesn't give any output to me.

Any idea how do deal with it?
Thanks.

[ampra2]

Hope this time it works:

This is the ps result:

ampra2   17005  0.0  0.0  65932  1332 ?        Ss   22:02   0:00 -bash /opt/sge-6.2u4/default/spool/east-25/job_scripts/827715
ampra2   17044  100  0.0  33984  1556 ?        R    22:02   0:15 openssl speed
root     17045  0.0  0.0  94448  3520 ?        Ss   22:02   0:00 sshd: ampra2 [priv]
ampra2   17047  0.0  0.0  94448  1764 ?        S    22:02   0:00 sshd: ampra2@pts/0
ampra2   17048  0.1  0.0  66200  1600 pts/0    Ss   22:02   0:00 -bash
ampra2   17089  0.0  0.0  65596   968 pts/0    R+   22:02   0:00 ps aux
ampra2   17090  0.0  0.0  61156   736 pts/0    S+   22:02   0:00 grep ampra2

[Irithori]

May I suggest to let the called script write a pidfile into /var/run/<process>/<process>.pid?

This way you avoid parsing ps output (and hope for a match)
and you also gain some portability, as ps is not the same on each unix platform.

[ampra2]

Hi Irithori,

Thanks for your reply, but the called script is actually a job submitted into a grid, so I can do nothing with it.

On that case, 827715 is the JobID of the job submitted on the head node, this is what I can use to search for the right process when the job is executed on the computing node.

Any idea how to get its pid with this condition?
Thanks.

[Kloschüssel]

I'm not sure if I grasp this. You submit a job to a grid and that job executes somewhere .. and you want to get the pid for it on your pc? Isn't that ... impossible unless you ssh into the machine that really executes the job? When I last dealed with grid computing there were several well defined tools to place jobs, retrieve the status of a job and finally also retrieve the results.

Anyway, you might consider taking a closer look at the manual pages of ps (man ps), especially the options -C and -o.

[ampra2]

Hi Kloschüssel,

It's a bit complicated to explain. To make it short, I'm in the middle of building a process monitoring tool to work on grid. To do that, I need to get the pid of the job when it's running on the node.

You don't have to care about the grid thing actually, I just need to know the way I can get pid of a specific process with only knowing the name of the script being executed.

I can do multiple grep like this which will direct me to the correct process.

ps ax | grep -v "grep" | grep 827899 | grep job_scripts
17005 ?        Ss     0:00 -bash /opt/sge-6.2u4/default/spool/east-26/job_scripts/827715

but it gives me a long output, as I can't use "-o pid=" option on "ps" which can be used to limit the output only on pid.
It means I have to do another extra job by parsing this output to get only the pid.

I'm wondering whether there is another easy way to get only the PID.

[Irithori]

I also miss the big picture here.
But given, that one would locally look for a pid, that has a process name of bash and "827715" somewhere in the comandline, then this should do.
And maybe can be matched much closer.

pgrep -f "bash(.*)827715"

[ampra2]

pgrep -f "bash(.*)827715"

It works. Thanks Irithori

[Kloschüssel]

Please read the manual pages, there's an example telling "Print only the process IDs of syslogd".

EDIT: pgrep is new to me .. cool!

[ampra2]

Please read the manual pages, there's an example telling "Print only the process IDs of syslogd".

EDIT: pgrep is new to me .. cool!

I did read it before I posted here, but I still can't find any way doing it with ps.

Fortunately, there is pgrep which is -similar with you- also new to me