A template repository for model configurations utilizing the ACCESS-NRI/model-config-tests CI infrastructure
Note
Feel free to replace this README with information on the model configurations once the TODOs have been ticked off. An example README.md is in the last section
Appropriate branch protections (or rulesets) should be set up on main, release-* and dev-* branches.
There are a few repository secrets and variables that need to be set:
secrets.GH_COMMIT_CHECK_TOKENused to allow checks to run on workflow-generated commits.secrets.GH_FORCE_PUSH_TOKENused to bypass branch protections for committing checksums torelease-*branches.vars.CI_JSON_SCHEMA_VERSION: the version of the schema for theconfig/ci.jsonconfiguration file. Possible versions are found in theACCESS-NRI/schemarepository here.
Important
These Environments MUST be created in your <model>-configs repository, otherwise the linked reusable workflows will not have the information required to connect to a given deployment target.
GitHub Environments are sets of variables and secrets that have more security requirements for their use. For example, they can require that a user or team must sign off on jobs that use a given environment, because it makes use of allocated supercomputer time, for example.
We at present require two environments per deployment target with the same information - one that is used with the ci.yml and schedule.yml workflows, and a protected one that is used by the user-dispatchable generate-initial-checksums.yml workflow. For a deployment target SUPERCOMPUTER, we usually call these SUPERCOMPUTER and SUPERCOMPUTER Initial Checksum respectively.
Regarding the secrets and variables that must be created:
secrets.SSH_HOST: The deployment location SSH Hostsecrets.SSH_HOST_DATA: The deployment location SSH Host for data transfer (may be the same asSSH_HOST)secrets.SSH_KEY: A SSH Key that allows access to the aboveHOST/HOST_DATAsecrets.SSH_USER: A Username to login to the aboveHOST/HOST_DATAvars.DEPLOYMENT_TARGET: Name of the deployment target for logging purposes (ex.Supercomputer)vars.EXPERIMENTS_LOCATION- directory on the deployment target that will contain all the experiments used during testing of reproducibility across multiple runs of this workflow (ex./scratch/some/directory/experiments)vars.MODULE_LOCATION- directory on the deployment target that contains module files for payu used during reproducibility testing (ex./g/data/vk83/modules)vars.PRERELEASE_MODULE_LOCATION- directory on the deployment target that contains module files for development version of payu (ex./g/data/vk83/prerelease/modules)
This file contains configuration information on the types of pytest markers to run for different styles of test, as well as the version of Python and the overall model-config-tests package. It contains sensible defaults in case one does not specify markers, python-version or model-config-tests-version.
There are two levels of defaults. There's a default at test type level which is useful for defining test markers - this selects certain pytests to run in model-config-tests. There is an outer global default, which is used if a property is not defined for a given branch/tag, and it is not defined for the test default. The parse-ci-config action applies the fall-back default logic.
Hence, one will need to modify this file as configuration tags are added (for scheduled checks) or new markers are added (for reproducibility/qa checks).
For example, if one wants to add scheduled checks with the default markers for the config tag release-1deg_jra55_iaf-2.0, you would add the following to the scheduled section in config/ci.json:
{
# ...
"scheduled": {
"release-1deg_jra55_iaf-2.0": {},
# ...
}
}If you wanted to add the tests from some new marker other, it would look like so:
{
# ...
"scheduled": {
"release-1deg_jra55_iaf-2.0": {
"markers": "other or checksum"
},
# ...
}
}Note that the default at any level can be modified to your own liking.
Check the on.schedule.cron trigger - currently it runs on the first day of every month. Adjust as required.
Check the GitHub Environment specified in jobs.generate-checksums.with.environment-name. It is currently Gadi Initial Checksum. Adjust the name depending on the deployment target.
As noted earlier, one must have created the appropriate GitHub Environments in order for all the linked reusable workflows to work correctly.
Note the jobs.*.secrets:inherit sections on almost all of the jobs in each workflow - it is explicitly inheriting the secrets and variables from the GitHub Environments that you have created in your repository, not the repository that contains the reusable workflows!
Note
Even though this is an example, there are some fill-in-the-blanks in <these>. Adjust to your specific flavour of model configuration repository.
This repo contains standard global configurations for <MODEL>.
This is an "omnibus repository": it contains multiple related configurations, and each configuration is stored in a separate branch.
Branches utilise a naming scheme of <some kind of scheme like{resolution}_{atmosforcing}_{forcingmode}[_{option}]>.
All available configurations are browsable under the list of branches and should also be listed below:
| Branch | Configuration Description |
|---|---|
release-example-configuration |
A model |
There are more detailed notes contained in the respective branches for each configuration.
More supported configurations will be added over time.
All configurations use payu to run the model.
This repository contains many related experimental configurations to make support and discovery easier. As a user it does not necessarily make sense to clone all the configurations at once.
In most cases only a single experiment is required. If that is the case choose which experiment and then run
git clone -b <experiment> https://github.com/ACCESS-NRI/model-configs/ <experiment>and replace <experiment> with the branch name or tag of the experiment you wish to run.
ACCESS-Hive contains detailed instructions for how to configure and run <MODEL> with payu.
This repository makes use of GitHub Actions to perform reproducibility checks on model config branches.
Config branches are branches that store model configurations of the form: release-<config> or dev-<config>, for example: release-1deg_jra55_iaf. For more information on creating your own config branches, or for understanding the PR process in this repository, see the CONTRIBUTING.md.
Config tags are specific tags on config branches, whose MAJOR.MINOR version compares the reproducibility of the configurations. Major version changes denote that a particular config tag breaks reproducibility with tags before it, and a minor version change does not. These have the form: release-<config>-<tag>, such as release-1deg_jra55_iaf-1.2.
So for example, say we have the following config tags:
release-1deg_jra55_iaf-1.0release-1deg_jra55_iaf-1.1release-1deg_jra55_iaf-2.0release-1deg_jra55_iaf-3.0
This means that *-1.0 and *-1.1 are configurations for that particular experiment type that are reproducible with each other, but not any others (namely, *-2.0 or *-3.0).
*-2.0 is not reproducible with *-1.0, *.1.1 or *-3.0 configurations.
Similarly, *-3.0 is not reproducible with *-1.0, *-1.1 or *-2.0.
These checks are in the context of:
- PR checks: In which a PR creator can modify a config branch, create a pull request, and have their config run and checked for reproducibility against a 'ground truth' version of the config.
- Scheduled checks: In which config branches and config tags that are deemed especially important are self-tested monthly against their own checksums.
More information on submitting a Pull Request and on the specifics of this pipeline can be found in the CONTRIBUTING.md and README-DEV.md respectively.
For more information on the manually running the pytests that are run as part of the reproducibility CI checks, see model-config-tests.
<NOTE CONDITIONS OF USE>