Skip to content

tloubrieu-jpl/incubator-sdap-nexus-ingestion-manager

Repository files navigation

SDAP manager for ingestion of datasets

Prerequisites

python 3

Install anaconda for python 3. From the graphic install for example for macos:

https://www.anaconda.com/distribution/#macos

git lfs (for development)

Git lfs for the deployment from git, see https://git-lfs.github.com/

If not available you have to get netcdf files for test, if you do need the tests.

Deployed nexus on kubernetes cluster

See project https://github.com/apache/incubator-sdap-nexus

$ helm install nexus .  --namespace=sdap --dependency-update -f ~/overridden-nexus-values.yml 

Install, Configure and run

Install

Stay logged in as user

$ pip install sdap-ingest-manager

Configure the ingestion system

Catch the message at the end of the installation output

--------------------------------------------------------------
Now, create configuration files in
***/<some path>>/.sdap_ingest_manager***
 Use templates and examples provided there
--------------------------------------------------------------

If the path does not show in the installation stdout, you can find it with the command:

python -c "import sys; print(f'{sys.prefix}/.sdap_ingest_manager')"

Use the path shown in the message and create your own configuration files:

$ cd /<some path>>/.sdap_ingest_manager
$ cp sdap_ingest_manager.ini.default sdap_ingest_manager.ini

Edit and update the newly created files by following instructions in the comments.

Note that the .ini.default file will be used if no value is configured in the .ini file. So you can have a simplified .ini file with only your specific configuration. Don't put your specific configuration in the .ini.default file, it will be replaced when you upgrade the package.

Example of a simplified .ini file:

[COLLECTIONS_YAML_CONFIG]
yaml_file = collections.yml

[OPTIONS]
# set to False to actually call the ingestion command for each granule
dry_run = False
# set to True to automatically list the granules as seen on the nfs server when they are mounted on the local file system.
deconstruct_nfs = True
# number of parallel ingestion pods on kubernetes (1 per granule)
parallel_pods = 2

[INGEST]
# kubernetes namespace where the sdap cluster is deployed
kubernetes_namespace = nexus-dev

Configure the collections

You can configure it in a local yaml file referenced in the sdap_ingest_manager.ini file.

It can also be in a google spreadsheet.

If both are configured, the local yaml file will be used.

Run the ingestion

On the list of the configured collections:

$ run_collections

The number of parallel jobs can be updated during the process in the sdap_ingest_manager.ini file.

If interrupted (killed) the process will restart where it was.

For developers

deploy project

$ bash
$ git clone ...
$ cd sdap_ingest_manager
$ python -m venv venv
$ source ./venv/bin/activate
$ pip install .

Note the command pip install -e . does not work as it does not deploy the configuration files.

Update the project

Update the code and the test with your favorite IDE (e.g. pyCharm).

Test and create the package

A package based on the dev branch is automatically published at github release when a push is made.

Change version in file setup.py

$ python setup.py test
$ git tag <version>
$ git push origin <version>

The release will be automatically pushed to pypi through github action.

Containerization

Docker

(development version)

cd containers/docker
docker build --no-cache --tag tloubrieu/sdap-ingest-manager:latest .    
docker run -it --name sdap-ingest-manager -v sdap_ingest_config:/usr/local/.sdap_ingest_manager tloubrieu/sdap-ingest-manager:latest
docker volume inspect sdap_ingest_config

You can see the configuration files in the directory of the named volume (for example /var/lib/docker/volumes/sdap_ingest_config/_data).

Note on macos, to access this directory, you need to go inside the Virtual Machine which runs docker service. To update the configuration on macos:

docker run --rm -it -v /:/vm-root alpine:edge /bin/ash
cd /vm-root/var/lib/docker/volumes/sdap_ingest_config/_data
cp sdap_ingest_manager.ini.default sdap_ingest_manager.ini
vi sdap_ingest_manager.ini

To publish the docker image on dockerhub do (step necessary for kubernetes deployment):

docker login
docker push tloubrieu/sdap-ingest-manager:latest

Kubernetes

Create the configMap for your deployment

Prepare a configMap from existing native config files:

kubectl create configmap collection-ingester-config --from-file=venv/.sdap_ingest_manager -n sdap

Optionally you can update the configMap manually if the one you started from is not what you needed:

kubectl get configmap collection-ingester-config -o yaml -n sdap > containers/kubernetes/sdap_ingester_config.yml

Manually edit the yml file to only keep the configuration which is specific to the deployment (if different from the current one)

Replace the configmap:

kubectl delete configmap collection-ingester-config -n sdap
kubectl apply -f containers/kubernetes/sdap_ingester_config.yml -n sdap

Launch the service

kubectl apply -f containers/kubernetes/job.yml -n sdap

Delete the service:

kubectl delete jobs --all -n sdap