C H A P T E R 6 - Killing or Sending Signals to Programs With mpkill

C H A P T E R 6

Killing or Sending Signals to Programs With mpkill

What You Can Do

To Perform This Task	Use This Option
How to kill a running program	mpkill
How to remove all traces of a job	-C
How to display a list of signals	-l -d
How to send a signal to a job	-signal

Return Values

The mpkill command returns these values:

0 - The command executed successfully.

1 - An error occurred during execution. For example, the job was not known.

2 - The command was partially successful. This typically occurs when you send a signal to a job in which one or more of the processes has already exited and therefore could not receive the signal. Note that this is usually not an error, since the reason you are using mpkill is most likely to eliminate a job that has hung in this intermediate state.

To Kill a Running Program

To kill a running program, use the mpkill command and the program's job ID:

% mpkill jid

The mpkill command stops all the processes associated with the Job ID.

The job ID now begins with the name of the resource manager (cre, lsf, pbs, or sge). For example: lsf.1289. To obtain a program's job ID, use the mpps command, described in To Display Information About Individual Jobs (-J).

To Remove All Traces of a Job

If you have killed a job but it continues to appear in the output of the mpps command (described in Chapter 7), log in as root to the master node and invoke the mpkill command with the -C option and the jid.

% mpkill -C jid

The -C option purges the job from the CRE database, including unpublishing names associated with the job.

Note - Processes spawned in the ClusterTools Runtime Environment are not killed by the mpkill or kill commands so long as they have (spawned) child processes running. To remove the parent process, you must first remove all of its child processes.

To Display a List of Supported Signals
(`-l -d`)

To simply list the supported signals, use the -l option.

% mpkill -l

To display a list with brief descriptions, use the -d option.

% mpkill -d

To Send a Signal to a Job

To send a signal to a job, use this syntax:

% mpkill -signal jid

For example:

% mpkill -CONT sge.59

The example above sends a SIGCONT signal to the processes of the program whose job ID is sge.59.

Issuing mpkill without specifying a signal sends a SIGTERM to the job.