Skip to content

Open-EO/openeo_odc_driver

Repository files navigation

OpenEO ODC Driver

OpenEO processing engine written in Python based on OpenDataCube, Xarray and Dask.

Please note: this project is still under development and many changes will occur in the next months.

Currently it is based on openEO processes implemented in this repository together with the openEO process graph parser from https://github.com/Open-EO/openeo-pg-parser-python

The next phase of the project will use the process implementations available at openeo-processes-dask and the parser openeo-pg-parser-networkx.

Docker deployment:

Based on the cube-in-a-box project and on the Docker image from andriyreznik: https://github.com/andriyreznik/docker-python-gdal

It requires Docker and Compose plugin installed. Please check the official documentation if you don't know how to install them: https://docs.docker.com/compose/install/

It also requires make to use the Makefile. If you don't have it and/or you can't install it, you can just manually type in the command contained in the Makefile.

Step 1: Clone the repository

git clone https://github.com/SARScripts/openeo_odc_driver.git -b dev
cd openeo_odc_driver

Step 2: Create and run the dockers

To create and run the docker containers, run:

make setup

This step will run in a sequence the Makefile steps build up init product index explorer. The result will be 3 dockers running.

It will automatically index publicly available Sentinel-2 data. To change the area of interest, modify the BBOX in the Makefile: lon_min,lat_min,lon_max,lat_max

One will serve the postgres database containing the OpenDataCube index data.

Another will serve the datacube-explorer web application, exposing the OpenDataCube metadata.

The last one will serve the openeo_odc_driver, getting as input openEO proces graphs and processing them.

If everything went well, you should see them running using:

sudo docker ps -a
CONTAINER ID        IMAGE                               COMMAND                  CREATED             STATUS                    PORTS                    NAMES
220ebc8a1dfe        openeoodcdriver_openeo_odc_driver   "/tini -- gunicorn -…"   18 hours ago        Up 18 hours               0.0.0.0:5001->5000/tcp   openeoodcdriver_openeo_odc_driver_1
c6a419dfb99d        openeoodcdriver_explorer            "/tini -- gunicorn '…"   18 hours ago        Up 18 hours               0.0.0.0:9001->9000/tcp   openeoodcdriver_explorer_1
fa1b162c8d44        postgis/postgis:12-2.5              "docker-entrypoint.s…"   36 hours ago        Up 36 hours               0.0.0.0:5433->5432/tcp   openeoodcdriver_postgres_1

You can verify that the deployment was successfull visiting:

http://localhost:9001 for the datacube explorer web app

http://localhost:5001/collections/ for the openEO collections exposed via the openeo_odc_driver.

Troubleshooting

make setup : docker.errors.DockerException: Error while fetching server API version

After checking that the docker service is actually running, this might be then a permissions issue with the socket:

sudo chmod 666 /var/run/docker.sock

(credits: Mafei@SO)

Step 3: Test your environment with an openEO process graph:

python tests/test_process_graph.py ./tests/process_graphs/NDVI_Bolzano_median.json

image

image

Local installation instructions:

Step 1: Clone the repository

git clone https://github.com/SARScripts/openeo_odc_driver.git -b dev
cd openeo_odc_driver

Step 2: Prepare the python environment

New ad-hoc conda environment:

conda env create -f environment.yml
conda activate openeo_odc_driver

Step 3:

Modify the config.py file with your system's details:

  1. Set the datacube-explorer address. For local deployment it should be http://0.0.0.0:9000 and for the Docker deployment http://explorer:9000.
DATACUBE_EXPLORER_ENDPOINT = "http://0.0.0.0:9000"
  1. Set the OpenDatCube config file .datacube.conf path or leave it to None if ENV variables are set (like in the Docker deployment).
OPENDATACUBE_CONFIG_FILE = ~/.datacube.conf # or None
  1. Set the result folder path to write output files. If this application is used together with the openeo-sping-driver, used for serving the full openEO API, this folder should be the same as the one set in application.properties for org.openeo.tmp.dir, so that the openeo-spring-driver can read the result directly from there.
RESULT_FOLDER_PATH = ''
  1. The OPENEO_PROCESSES variable is used to retrieve the list of available openEO processes. It can be the path to a json file, a dictionaty or an http address. See here for detailed info. The dault value is the /processes endpoint of the Eurac openEO back-end.
OPENEO_PROCESSES = "https://openeo.eurac.edu/processes"

The other config parameters could be looked at later on and are not affecting the openeo_odc_driver functionality. In the config.py file there are comments explaining their usage.

Step 4: Start the web server:

gunicorn -c gunicorn.conf.py odc_backend:app

Implemented OpenEO processes (to be updated)

aggregate & resample

  • resample_cube_temporal
  • resample_cube_spatial
  • aggregate_spatial
  • aggregate_spatial_window
  • aggregate_temporal_period

arrays

  • array_element
  • array_interpolate_linear

comparison

  • if
  • lt
  • lte
  • gt
  • gte
  • eq
  • neq

cubes

  • load_collection
  • save_result (PNG,GTIFF,NETCDF,JSON)
  • reduce_dimension
  • add_dimension
  • apply_dimension
  • filter_bands
  • filter_temporal
  • filter_spatial
  • filter_bbox
  • rename_labels
  • merge_cubes
  • apply
  • fit_curve
  • predict_curve
  • resample_cube_spatial
  • resample_cube_temporal

logic

  • and
  • or

masks

  • mask

math

  • multiply
  • divide
  • subtract
  • add
  • sum
  • product
  • sqrt
  • normalized_difference
  • min
  • max
  • mean
  • median
  • power
  • absolute
  • linear_scale_range
  • log
  • ln
  • quantiles
  • clip

experimental processes (SAR2Cube)

  • coherence
  • geocoding
  • radar_mask

About

OpenEO processing engine written in Python based on OpenDataCube, Xarray and Dask.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published