Utilities#
This section provides information on tools that are useful when developing ESMValTool. Tools that are specific to ESMValTool live in the esmvaltool/utils directory, while others can be installed using the usual package managers.
Pre-commit#
pre-commit is a handy tool that can run many tools for checking code quality with a single command. Usually it is used just before committing, to avoid accidentally committing mistakes. It knows knows which tool to run for each filetype, and therefore provides a convenient way to check your code!
To run pre-commit
on your code, go to the ESMValTool directory
(cd ESMValTool
) and run
pre-commit run
By default, pre-commit will only run on the files that have been changed,
meaning those that have been staged in git (i.e. after
git add your_script.py
).
To make it only check some specific files, use
pre-commit run --files your_script.py
or
pre-commit run --files your_script.R
Alternatively, you can configure pre-commit
to run on the staged files before
every commit (i.e. git commit
), by installing it as a git hook using
pre-commit install
Pre-commit hooks are used to inspect the code that is about to be committed. The commit will be aborted if files are changed or if any issues are found that cannot be fixed automatically. Some issues cannot be fixed (easily), so to bypass the check, run
git commit --no-verify
or
git commit -n
or uninstall the pre-commit hook
pre-commit uninstall
Note that the configuration of pre-commit lives in .pre-commit-config.yaml.
nclcodestyle#
A tool for checking the style of NCL code, based on pycodestyle.
Install ESMValTool in development mode (pip install -e '.[develop]'
) to make it available.
To use it, run
nclcodestyle /path/to/file.ncl
Colormap samples#
Tool to generate colormap samples for ESMValTool’s default Python and NCL colormaps.
Run
esmvaltool colortables python
or
esmvaltool colortables ncl
to generate the samples.
Running multiple recipes#
It is possible to run more than one recipe in one go.
This can for example be achieved by using rose
and/or cylc
, tools
that may be available at your local HPC cluster.
In the case in which neither rose
nor cylc
are available at your HPC cluster,
it is possible to automatically generate job submission scripts, as well as a summary of the
job outputs using the scripts available in
esmvaltool/utils/batch-jobs.
Using cylc#
A cylc suite for running all recipes is available in esmvaltool/utils/testing/regression. This suite is configured to work with versions of cylc older than 8.0.0 .
To prepare for using this tool:
Log in to a system that uses slurm
Make sure the required CMIP and observational datasets are available and their
rootpath
anddrs
is properly set up in the configurationMake sure the required auxiliary data is available (see recipe documentation)
Install ESMValTool
Next, get started with cylc:
Run
module load cylc
Register the suite with cylc
cylc register run-esmvaltool-recipes ~/ESMValTool/esmvaltool/utils/testing/regression
Edit the suite if needed, this allows e.g. choosing which recipes will be run
Validate the suite
cylc validate run-esmvaltool-recipes --verbose
, this will e.g. list the recipes in the suiteRun all recipes
cylc run run-esmvaltool-recipes
View progress
cylc log run-esmvaltool-recipes
, use e.g.cylc log run-all-esmvaltool-recipes examples-recipe_python_yml.1 --stdout
to see the log of an individual esmvaltool run. Once the suite has finished running, you will see the message “WARNING - suite stalled” in the log.Stop the cylc run once everything is done
cylc stop run-esmvaltool-recipes
.
To generate an overview page of the recipe runs, use the summarize.py
utility script.
Using the scripts in utils/batch-jobs#
In utils/batch-jobs, you can find a script to generate slurm submission scripts for all available recipes in ESMValTool, as well as a script to parse the job outputs.
Using generate.py#
The script generate.py, is a simple python script that creates slurm submission scripts, and if configured, submits them to the HPC cluster. It has been tested in DKRZ’s Levante cluster.
The following parameters have to be set in the script in order to make it run:
env
, str: Name of the conda environment in which esmvaltool is installed.mail
, bool: Whether or not to receive mail notifications when a submitted job fails or finishes successfully. Default isFalse
.submit
, bool: Whether or not to automatically submit the job after creating the launch script. Default value isFalse
.account
, str: Name of the DKRZ account in which the job will be billed.outputs
, str: Name of the directory in which the job outputs (.out and .err files) are going to be saved. The outputs will be saved in /home/user/<outputs>.conda_path
, str: Full path to the miniforge3/etc/profile.d/conda.sh executable.
Optionally, the following parameters can be edited:
config_dir
, str: Path to configuration directory, by default~/.config/esmvaltool/
.partition
, str: Name of the DKRZ partition used to run jobs. Default isinteractive
to minimize computing cost compared tocompute
for which nodes cannot be shared.memory
, str: Amount of memory requested for each run. Default is64G
to allow to run 4 recipes on the same node in parallel.time
, str: Time limit. Default is04:00:00
to increase the job priority. Jobs can run for up to 8 hours and 12 hours on the compute and interactive partitions, respectively.default_max_parallel_tasks
, int: Default is8
which works for most recipes. For other cases, an entry needs to be made to theMAX_PARALLEL_TASKS
dictionary (see below).
The script will generate a submission script for all recipes using by default the interactive
queue and with a time limit of 4h. In case a recipe
may require of additional resources, they can be defined in the SPECIAL_RECIPES
dictionary. The recipe name has to be given as a key
in which the
values are another dictionary.
The latter are used to specify the partition
in which to submit the recipe, the new time
limit and other memory
requirements
given by the slurm flags --mem
, --constraint
or --ntasks
. In general, an entry in SPECIAL_RECIPES
should be set as:
SPECIAL_RECIPES = {
'recipe_name': {
'partition': '#SBATCH --partition=<name_of_the_partition>',
'time': '#SBATCH --time=<custom_time_limit>',
'memory': '#SBATCH --mem=<custom_memory_requirement>' # --constraint or --nstasks can be used instead.
},
}
Some recipes can only be run with a number of tasks less than default_max_parallel_tasks
for various reasons (memory issues, diagnostic issues, CMIP3 data used).
These recipes need to be added to the MAX_PARALLEL_TASKS
dictionary with a specific max_parallel_tasks
value.
Note that the script has been optimized to use standard SLURM settings to run most recipes while minimizing the computational cost of the jobs and tailored runtime settings for resource-intensive recipes. It is only necessary to edit this script for recipes that have been added since the last release and cannot be run with the default settings.
In the case in which submit
is set to True
, but you want to exclude certain recipes from being submitted, their name can be added in the exclude
list:
exclude = ['recipe_to_be_excluded_1', 'recipe_to_be_excluded_2']
Using parse_recipes_outputs#
You can run this script (simply as a standalone Python script) after all recipes have been run, to gather a bird’s eye view
of the run status for each recipe; running the script provides you with a Markdown-formatted list of recipes that succeeded,
recipes that failed due to a diagnostic error, and recipes that failed due to missing data (the two most common causes for
recipe run failure). You should provide the location of the output log files from SLURM (*.out
and *.err
) to the
script as well as a list of all available recipes. To generate the list, run the command:
for recipe in $(esmvaltool recipes list | grep '\.yml$'); do echo $(basename "$recipe"); done > all_recipes.txt
To keep the script execution fast, it is recommended to use log_level: info
in the configuration so that SLURM
output files are rather small.
Overview of recipe runs#
To create overview webpages of a set of recipe runs, run:
python esmvaltool/utils/testing/regression/summarize.py ~/esmvaltool_output/
This will generate 2 html files:
index.html
that displays a summary of each recipe run, with a title and a representative plot, a short description of the aim of the recipe, and links to each individual run.debug.html
that provides an overview table of successful and failed runs with links to each individual run, and computing resources used for each run.
Comparing recipe runs#
A command-line tool is available for comparing one or more recipe runs to known good previous run(s). This tool uses xarray to compare NetCDF files and difference hashing provided by imagehash to compare PNG images. All other file types are compared byte for byte.
To use it, first install the package imagehash:
pip install imagehash
Next, go to the location where ESMValTool is installed and run
python esmvaltool/utils/testing/regression/compare.py ~/reference_output/ ~/output/recipe_python_20220310_180417/
where the first argument is a reference run or a directory containing such runs and the second and following arguments are directories with runs to compare to the reference run(s).
To compare all results from the current version to the previous version, use e.g.:
python esmvaltool/utils/testing/regression/compare.py /shared/esmvaltool/v2.4.0 /shared/esmvaltool/v2.5.0
To get more information on how a result is different, run the tool with the
--verbose
flag.
Testing recipe settings#
A tool for generating recipes with various diagnostic settings, to test of those work.
Install ESMValTool in development mode (pip install -e '.[develop]'
) to make it available.
To use it, run
test_recipe --help
draft_release_notes.py#
draft_release_notes.py is a script for drafting release notes based on the titles and labels of the GitHub pull requests that have been merged since the previous release.
To use it, install the package pygithub:
pip install pygithub
Create a GitHub access token (leave all boxes for additional
permissions unchecked) and store it in the file ~/.github_api_key
.
Edit the script and update the date and time of the previous release and run the script:
python esmvaltool/utils/draft_release_notes.py ${REPOSITORY}
REPOSITORY
can be either esmvalcore
or esmvaltool
depending on the
release notes you want to create.
Review the resulting output (in .rst
format) and if anything needs changing,
change it on GitHub and re-run the script until the changelog looks acceptable.
In particular, make sure that pull requests have the correct label, so they are
listed in the correct category.
Finally, copy and paste the generated content at the top of the changelog.
Converting Version 1 Namelists to Version 2 Recipes#
The xml2yml converter can turn the old xml namelists into new-style yml recipes. It is implemented as a xslt stylesheet that needs a processor that is xslt 2.0 capable. With this, you simply process your old namelist with the stylesheet xml2yml.xsl to produce a new yml recipe.
After the conversion you need to manually check the mip information in the variables! Also, check the caveats below!
Howto#
One freely available processor is the Java based saxon. You can download the free he edition here. Unpack the zip file into a new directory. Then, provided you have Java installed, you can convert your namelist simply with:
java -jar $SAXONDIR/saxon9he.jar -xsl:xml2yml.xsl -s:namelist.xml -o:recipe.yml
Caveats/Known Limitations#
At the moment, not all model schemes (OBS, CMIP5, CMIP5_ETHZ…) are supported. They are, however, relatively easy to add, so if you need help adding a new one, please let me know!
The documentation section (namelist_summary in the old file) is not automatically converted.
In version 1, one could name an exclude, similar to the reference model. This is no longer possible and the way to do it is to include the models with another
additional_models
tag in the variable section. That conversion is not performed by this tool.
Authored by Klaus Zimmermann, direct questions and comments to klaus.zimmermann@smhi.se
Extracting a list of input files from the provenance#
There is a small tool available to extract just the list of input files used to generate
a figure from the *_provenance.xml
files (see Recording provenance for more
information).
To use it, install ESMValTool from source and run
python esmvaltool/utils/prov2files.py /path/to/result_provenance.xml
The tool is based on the prov library, a useful library for working with provenance files. With minor adaptations, this script could also print out global attributes of the input NetCDF files, e.g. the tracking_id.
Recipe Test Workflow (RTW)#
The Recipe Test Workflow (RTW) is a workflow that is used to regularly run recipes so issues can be discovered during the development process sooner rather than later.