Purpose:

The purpose of this program is to transfer the Unified Forecast System Land Data Assimilation Application (UFS Land DA) datasets residing within the RDHPCS to cloud data storage via chaining API calls to communicate with its cloud data storage bucket. The program will support the data required for the UFS Land DA Application release versions.

According to Amazon AWS, the following conditions need to be considered when transferring data to cloud data storage:

• Largest object that can be uploaded in a single PUT is 5 GB.
• Individual Amazon S3 objects can range in size from a minimum of 0 bytes to a maximum of 5 TB.
• For objects larger than 100 MB, Amazon recommends using the Multipart Upload capability.
• The total volume of data in a cloud data storage bucket are unlimited.

Tools which could be be utilized to perform data transferring & partitioning (Multipart Upload/Download) are:

• AWS SDK
• AWS CLI
• AWS S3 REST API

In this demontration, the framework will implement Python AWS SDK for transferring the UFS Land DA application fixed and input model datasets from the RDHPCS, Orion, to the cloud data storage with low latency.

The AWS SDK will be implemented for the following reasons: To integrate with other python scripts. AWS SDK carries addition capabilities/features for data manipulation & transferring compare to the aforementioned alternate tools.

Capabilities:

The framework will be able to perform the following actions:

Multi-threading & partitioning the datasets to assist in the optimization in uploading performance of the datasets from on-premise to cloud.

Future Capabilities:

Can make the framework customizable so that users can request the Land DA datasets that is applicable to their Land DA release version needs -- rather than the full datasets within the Land DA tar folders as Land DA development continues within this project program in the future.

• Setting up AWS CLI configurations for uploading to Cloud.
• Setting up conda environment w/in RDHPCS.
  • Refer to Environment Setup
• 

Datasets for the Land DA v1.0.0 Release:

• Model Input Data
• Observation Data
• Sample Forcing Data
• NOAH-MP Offline Data (Model Spec Data for Offline Vector)
• GFS Tile Space Data (Model Spec Data for FV3 Tiles)
• Land DA Singularity Container
• Data required to compare the validated dataset to the user's experiment output dataset (comps data)
• Data required to run the comparison step (inps data)

1. Install miniconda per "Environment Setup" section.
2. Establish AWS credentials configuration file via the "AWS Command Line Interface (AWS CLI) Credentials Setup" page in Confluence.
3. Save data of interest to migrate to cloud within the repository's "main" folder & structure the data on-premise as you would like it structured in cloud. For example, the relative directory of "landda-data-2016.tar.gz" should be saved under the "current_land_da_release_data" folder within the repository's "main" folder on-premise for its object's key to be set as "current_land_da_release_data/landda-data-2016.tar.gz" in cloud.
4. Execute the following command within the terminal to migrate the data to cloud: python upload_file2cloud.py -b BUCKET_NAME -k FILE_DIR_TO_MIGRATE_TO_CLOUD_INCLUDING_FILENAME

• BUCKET_NAME = land-da
• FILE_DIR_TO_MIGRATE_TO_CLOUD_INCLUDING_FILENAME: For example, the relative directory of "landda-data-2016.tar.gz" should be saved as "current_land_da_release_data/landda-data-2016.tar.gz" within the "main" folder on-premise for its object's key to be set as "current_land_da_release_data/landda-data-2016.tar.gz" in cloud. So, FILE_DIR_TO_MIGRATE_TO_CLOUD_INCLUDING_FILENAME = current_land_da_release_data/landda-data-2016.tar.gz.

• Install miniconda on your machine. Note: Miniconda is a smaller version of Anaconda that only includes conda along with a small set of necessary and useful packages. With Miniconda, you can install only what you need, without all the extra packages that Anaconda comes packaged with: Download latest Miniconda (e.g. 3.9 version):

  • wget https://repo.anaconda.com/miniconda/Miniconda3-py39_4.9.2-Linux-x86_64.sh

• Check integrity downloaded file with SHA-256:

  • sha256sum Miniconda3-py39_4.9.2-Linux-x86_64.sh

Reference SHA256 hash in following link: https://docs.conda.io/en/latest/miniconda.html

• Install Miniconda in Linux:

  • bash Miniconda3-py39_4.9.2-Linux-x86_64.sh

• Next, Miniconda installer will prompt where do you want to install Miniconda. Press ENTER to accept the default install location i.e. your $HOME directory. If you don't want to install in the default location, press CTRL+C to cancel the installation or mention an alternate installation directory. If you've chosen the default location, the installer will display “PREFIX=/var/home//miniconda3” and continue the installation.

• For installation to take into effect, run the following command:

source ~/.bashrc

• Next, you will see the prefix (base) in front of your terminal/shell prompt. Indicating the conda's base environment is activated.

• Once you have conda installed on your machine, perform the following to create a conda environment:

  • To create a new environment (if a YAML file is not provided)

    • conda create -n [Name of your conda environment you wish to create]

(OR)

* To ensure you are running Python 3.9:

    * conda create -n myenv Python=3.9

(OR)

• To create a new environment from an existing YAML file (if a YAML file is provided):

  • conda env create -f environment.yml

*Note: A .yml file is a text file that contains a list of dependencies, which channels a list for installing dependencies for the given conda environment. For the code to utilize the dependencies, you will need to be in the directory where the environment.yml file lives.

conda activate [Name of your conda environment you wish to activate]

• Verify that the new environment was installed correctly via:

  • conda info --env

*Note:

• From this point on, must activate conda environment prior to .py script(s) or jupyter notebooks execution using the following command: conda activate
  • To deactivate a conda environment:
    • conda deactivate

• Unfortunately, there is no way to navigate to the "/work/" filesystem from within the Jupyter interface when working on the remote server, Orion. The best way to workaround is to create a symbolic link in your home folder that will take you to the /work/ filesystem. Run the following command from a linux terminal on Orion to create the link:

  • ln -s /work /home/[Your user account name]/work

• Now, when you navigate to the /home/[Your user account name]/work directory in Jupyter, it will take you to the /work folder. Allowing you to obtain any data residing within the /work filesystem that you have permission to access from Jupyter. This same procedure will work for any filesystem available from the root directory.

*Note: On Orion, user must sym link from their home directory to the main directory containing the datasets of interest.

• Open OnDemand has a built-in file explorer and file transfer application available directly from its dashboard via:

  • Login to https://orion-ood.hpc.msstate.edu/

• In the Open OnDemand Interface, select Interactive Apps > Jupyter Notbook

To create a .yml file, execute the following commands:

• Activate the environment to export:

  • conda activate myenv

• Export your active environment to a new file:

  • conda env export > [ENVIRONMENT FILENAME].yml

Within the download, you will find the following directories and files:

• Scripts:

  • Modules:
    • get_land_da_data.py
      • Extracts the data directories of a tar & partitions data by external model used in the creation of model analysis files.
    • upload_data.py
      • Uploads the UFS Land DA Application via AWS SDK
    • progress_bar.py
      • Monitors uploading progress of datasets to cloud
  • Main:
    • upload_file2cloud.py
      • Main executable script for extracting & uploading the full Land DA datasets residing on-prem to cloud. Sets unique key for the tar.gz data supporting the Land DA. Note: Must save data of interest to migrate to cloud within the repository's "main" folder & structure it on-premise as you would like it structured in cloud. For example, the relative directory of "landda-data-2016.tar.gz" should be saved as "current_land_da_release_data/landda-data-2016.tar.gz" within the "main" folder on-premise for the object's key to be set as "current_land_da_release_data/landda-data-2016.tar.gz" in cloud.
    • get_objects_list.py
      • Obtains list of unique data files within the Land DA cloud storage
    • delete_cloud_object.py
      • Main executable script for deleting the objects of the Land DA in datasets in cloud based on their unique key.

• List of Dependencies:

  • cloud_xfer_env.yml

• Draft as of 02/16/23