Launcher framework for executing jobs in the UFDL framework.
Python 3.7 or 3.8 (including development headers)
sudo apt-get install python3.7 python3.7-dev libpython3.7-dev
or
sudo apt-get install python3.8 python3.8-dev libpython3.8-dev
essential build environment
sudo apt-get install build-essential
additional libraries
sudo apt-get install virtualenv
dev_init.sh
- sets up a development virtual environment, use -h
for outputting the help screenThe following command-line tools (i.e., entry points) are available:
usage: ufdl-joblauncher [-h] [-C FILE] [-c]
Starts the UFDL job-launcher.
optional arguments:
-h, --help show this help message and exit
-C FILE, --config FILE
The configuration to use if not the system wide one
(/etc/ufdl/job-launcher.conf). (default: None)
-c, --continuous For continuous polling for jobs rather than stopping
after executing the first one. (default: False)
usage: ufdl-hwinfo [-h] [-C FILE] [-F FORMAT] [-O FILE]
Outputs UFDL hardware information.
optional arguments:
-h, --help show this help message and exit
-C FILE, --config FILE
The configuration to use if not the system wide one
(/etc/ufdl/job-launcher.conf). (default: None)
-F FORMAT, --format FORMAT
The format to use for the output of the information.
(default: yaml)
-O FILE, --output FILE
The file to store the information in, otherwise stdout
is used. (default: None)
An example configuration is available from examples/job-launcher-example.conf.
Executors are the workhorses in the job-launcher framework that interpret and execute the jobs that are compatible with a worker node's hardware setup (like GPU available and capability).
The jobs that get executed are based on the job templates defined in the UFDL backend. Executors are therefore tightly coupled with the job-templates, as these reference the executor class and required packages for running the executor.
The following super classes are available
ufdl.joblauncher.AbstractJobExecutor
- ancestor for all executorsufdl.joblauncher.AbstractDockerJobExecutor
- for launching docker-image-based jobsYou can find executor implementations in the following repositories: