DOKK / manpages / debian 10 / coop-computing-tools / work_queue_worker.1.en
work_queue_worker(1) Cooperative Computing Tools work_queue_worker(1)

work_queue_worker - worker process for executing tasks dispatched through Work Queue

work_queue_worker [options] <masterhost> <port>

work_queue_worker [options] <masterhost:port]>

work_queue_worker [options] "<masterhost:port;[masterhost:port;masterhost:port;...]>"

work_queue_worker [options] -M <projectname>

work_queue_worker is the worker process for executing tasks dispatched from a master application built using the Work Queue API. work_queue_worker connects to the master application, accepts, runs, and returns tasks dispatched to it.

The masterhost and port arguments specify the hostname and port number of the master application for work_queue_worker to connect. Several masterhosts and ports may be specified, separated with a semicolon (;), with the worker connecting to any of the masters specified (When specifying multiple masters, remember to escape the ; from shell interpretation, for example, using quotes.)

Alternatevely, the master may be specified by name, using the -M option.

work_queue_worker can be run locally or deployed remotely on any of the grid or cloud computing environments such as SGE, Amazon EC2, Condor using sge_submit_workers(1), ec2_submit_workers(1), condor_submit_workers(1) respectively.

 -v
Show version string.
 -h
Show this help message.
Set the name of the project this worker should work for. A worker can have multiple projects.
Set catalog server to <catalog>. Format: HOSTNAME:PORT
Enable debugging for the given subsystem. Try -d all as a start.
Write debugging output to this file. By default, debugging is sent to stderr (":stderr"). You may specify logs be sent to stdout (":stdout"), to the system syslog (":syslog"), or to the systemd journal (":journal").
 --debug-max-rotate <bytes>
Set the maximum file size of the debug log. If the log exceeds this size, it is renamed to "filename.old" and a new logfile is opened. (default=10M. 0 disables)
 --debug-release-reset
Debug file will be closed, renamed, and a new one opened after being released from a master.
 --foreman
Enable foreman mode.
Set the project name of this foreman to <project>. Implies --foreman.
 --foreman-port <port[:highport]>

Set the port for the foreman to listen on. If <highport> is specified the port is chosen from between <port> and <highport>. Implies --foreman.
Select port to listen to at random and write to this file. Implies --foreman.
Set the fast abort multiplier for foreman (default=disabled).
 --specify-log <logfile>
Send statistics about foreman to this file.
Password file for authenticating to the master.
Abort after this amount of idle time. (default=900s)
Set TCP window size.
Set initial value for backoff interval when worker fails to connect to a master. (default=1s)
Set maxmimum value for backoff interval when worker fails to connect to a master. (default=60s)
Minimum free disk space in MB. When free disk space is less than this value, the worker will clean up and try to reconnect. (default=100MB)
 --memory-threshold <size>
Set available memory threshold (in MB). When exceeded worker will clean up and reconnect. (default=100MB)
Set the architecture string the worker reports to its supervisor. (default=the value reported by uname)
Set the operating system string the worker reports to its supervisor. (default=the value reported by uname)
Set the location where the worker should create its working directory. (default=/tmp)
 --bandwidth <mbps>
Set the maximum bandwidth the foreman will consume in Mbps. (default=unlimited)
 --cores <n>
Set the number of cores this worker should use. Set it to 0 to have the worker use all of the available resources. (default=1)
 --gpus <n>
Set the number of GPUs this worker should use. (default=0)
 --memory <mb>
Manually set the amount of memory (in MB) reported by this worker.
 --disk <mb>
Manually set the amount of disk space (in MB) reported by this worker.
 --wall-time <s>
Set the maximum number of seconds the worker may be active.
 --feature <feature>
Specifies a user-defined feature the worker provides (option can be repeated).
 --docker <image>

Enable the worker to run each task with a container based on this image.
 --docker-preserve <image>

Enable the worker to run all tasks with a shared container based on this image.
 --docker-tar <tarball>

Load docker image from this tarball.
 --volatility <chance>
Set the percent chance per minute that the worker will shut down (simulates worker failures, for testing only).

work_queue_worker can also be run in foreman mode, in which it connects to a master as a worker while acting as a master itself. Any tasks the foreman receives from its master are sent to its subordinate worker processes.

Foreman mode is enabled by either specifying a port to listen on using the -f <port> option or by setting the mode directly with the --foreman option. The foreman can be directed to advertise its presence on the catalog_server() with the -N <project name> flag, which other workers can use to contact the foreman.

work_queue_worker can be run with container. Docker is the default management tool and docker deamon should be enabled in computing nodes. Tasks received from master can be run with container based on user specified docker image.

Container mode is enable by either specifying a image name using the --docker <image> option, which enable workers running each tasks with an independent container or by using the --docker-preserve <image> option, which enable workers running all tasks with a shared container. The default way to manage the image is using docker hub, which means user has to push the container image into the docker hub in advance. If the image is saved in a tarball and cached in the computing node, --docker-tar <tarball> option can be adopted to load the image from the tarball.

On success, returns zero. On failure, returns non-zero.

To run work_queue_worker to join a specific master process running on host master.somewhere.edu port 9123:

% work_queue_worker master.somewhere.edu 9123

To run work_queue_worker in auto mode with debugging turned on for all subsystems and to accept tasks only from a master application with project name set to project_A:

% work_queue_worker -a -d all -M project_A

To run work_queue_worker as a foreman working for project_A and advertising itself as foreman_A1 while listening on port 9123:

% work_queue_worker --foreman -M project_A -N foreman_A1 -f 9123

The Cooperative Computing Tools are Copyright (C) 2003-2004 Douglas Thain and Copyright (C) 2005-2015 The University of Notre Dame. This software is distributed under the GNU General Public License. See the file COPYING for details.

CCTools 7.0.9 FINAL