Jobs overview#
The jobs to be run in the workflow are defined in a set of configuration files stored in conf/common/jobs/. Including a file into the configuration by uncommenting the respective line in conf/<icon_case>.yml adds the jobs to the workflow.
Base workflow#
BUILD_ICON#
This job builds the ICON model. It will clone the model git repository, according to the values stored in conf/common/build.yml in the ICON section. Afterwards, the model is configured with the script specified by ICON.CONFIG_FILE (usually set in the platform-specific config file), then build and installed. Details on how to configure the job can be found on the build page.
BUILD_PYTHON#
This job sets up a Python virtual environment and installs all required packages there. A few configuration options are available in conf/common/build.yml.
PREPARE_EXPERIMENT#
This job prepares the remote HPC system for the current experiment. It copies all custom Python modules and the core experiment configuration file <expname>/experiment.yml to the HPC system.
PRE_FIND_FILES#
This job locates all input files on the system. For details see the page on input files.
PREPARE_MEMBER#
This job creates the working/output directory and links all input files there. The input files are specified in the DIRECTORIES.LINK_FILES section specified in <expname>/experiment.yml. See the comments there for details and the step-by-step guide. By default it requires the working directory to be non-existent or empty to not overwrite something. This behavior can be altered in the conf/art/simulation.yml file with the parameter SIMULATION.REQUIRE_CLEAN_OUTDIR.
PREPARE_NAMELIST#
This job creates the namelist for the experiment (a specific member and chunk if applicable). For details, see the namelist page.
RUN_ICON#
This job runs the actual simulation for a single chunk. It uses a so-called envmodules.sh file (platform-specific, in templates/common/platforms/) to set up the environment required for the run on a specific HPC system.
Additional jobs#
TRANSFER_PROJECT#
This job is only required for the cases real-from-ideal and real-from-dwd-ana and transfers auto-icon to the HPC system.
CLEAN_RESTART#
This job deletes the restart files regularly to save disk space. Per default it leaves the latest three files in place. It has to be included manually (in the include file or via the init script).
Note
Only available for the art ICON case.
TRANSFER#
With this job, one can transfer some or all output data to a different location on the HPC system or to the local platform.
Important
Do not copy any output data to the auto-icon server of KIT!
Preprocessing#
BUILD_DWD_ICON_TOOLS#
Build the DWD ICON tools required for some input data remapping tasks.
PRECHECK_REMAP_ERA2ICON#
This job checks for the existence of remapped ERA5 data. If the data is present, the PREP_REMAP_ERA2ICON job can be skipped.
PREP_REMAP_ERA2ICON / PREP_REMAP_IFS2ICON#
This job does the actual remapping of ERA5 or IFS input data to the ICON grid.
PRE_CREATE_GRID#
This job creates grids for the ICON model by using the DWD ICON tools grid generator.
PRE_REMAP#
This is a general job to remap input data to the target grid with CDO. Intended for all non-ERA5 and non-IFS input data.
Postprocessing#
GET_SAMOA#
Pull the SAMOA project from the git repository.
POST_SAMOA#
Run SAMOA checks on the output.
POST_CDO#
Do postprocessing with CDO. This can be configured in the main experiment configuration file <expname>/experiment.yml, section CDO.
VIS_GLOBAL_2D#
Plot some global fields in a 2D plot. Can be configured in <expname>/visualization.yml.
Note
Only available for the art ICON case.
VIS_ANIMATION_2D#
Make an animation of subsequent global field 2D plots. Can be configured in <expname>/visualization.yml.
Note
Only available for the art ICON case.
Testing#
This set of jobs defined in the file conf/common/jobs/testcase.yml provides a check of the output files against identical files in a respective reference set. On successful comparison it deletes all output data. This is especially useful for testing purposes.
PRE_CLEAN_REMOTE#
Clean all files on the remote HPC system. This job is for preparing a fresh test run.
POST_VALIDATE_BIN_EQUAL_TO_REF#
This job does the actual comparison. It uses CDO to compare each pair of files, whether the content matches exactly. This requires the reference to be run on the same hardware (how strict this condition is, is currently not fully established). The location of the reference set is defined via DIRECTORIES.REFDIR in conf/art/experiments/<EXPNAME>.yml.
POST_CLEAN_EXPERIMENT#
This job deletes all output data. Intended for clean up of a test run.