AODN Cloud Optimised Conversion

A tool to convert IMOS NetCDF files and CSV into Cloud Optimised format (Zarr/Parquet)

Installation

mamba env create --file=environment.yml
mamba activate CloudOptimisedParquet

pip install -e .  # development (edit mode)
pip install . # prod

Requirements

AWS SSO to push files to S3

Features List

Parquet Features

Feature	Status	Comment
Process IMOS tabular NetCDF to Parquet with GenericHandler	Done	Converts NetCDF files to Parquet format using a generic handler.
Process CSV to Parquet with GenericHandler	Done	Converts CSV files to Parquet format using a generic handler.
Specific Handlers inherit all methods from GenericHandler with super()	Done	Simplifies the creation of new handlers by inheriting methods.
Unittests implemented	Done	Tests to ensure functionality and reliability.
Reprocessing of files already converted to Parquet	Done	Reprocessing of NetCDF files; original method can be slow for large datasets.
Metadata variable attributes in sidecar parquet file	Done	Metadata attributes available in dataset sidecars.
Add new variables to dataset	Done	Addition of new variables such as site_code, deployment_code, filename attributes.
Add timestamp variable for partition key	Done	Enhances query performance by adding a timestamp variable.
Remove NaN timestamp when NetCDF not CF compliant	Done	Eliminates NaN timestamps, particularly for non CF compliant data like Argo.
Create dataset Schema	Done	Creation of a schema for the dataset.
Create missing variables available in Schema	Done	Ensures dataset consistency by adding missing variables from the schema.
Warning when new variable from NetCDF is missing from Schema	Done	Alerts when a new variable from NetCDF is absent in the schema.
Creating metadata parquet sidecar	Done
Create AWS OpenData Registry Yaml	Done
Config file JSON validation against schema	Done
Create polygon variable to facilite geometry queries	Done

Zarr Features

Feature	Status	Comment
Process IMOS Gridded NetCDF to Zarr with GenericHandler	Done	Converts NetCDF files to Parquet format using a generic handler.
Specific Handlers inherit all methods from GenericHandler with super()	Done	Simplifies the creation of new handlers by inheriting methods.

Usage

Parquet

The GenericHandler for parquet dataset creation is designed to be used either as a standalone class or as a base class for more specialised handler implementations. Here's a basic usage example:

# Read the content of the dataset template JSON file (with comments)
#import commentjson
#with open('aodn_cloud_optimised/config/dataset/dataset_template.json', 'r') as file:
#   json_with_comments = file.read()
#dataset_config = commentjson.loads(json_with_comments)

import importlib.resources
from aodn_cloud_optimised.lib.config import load_dataset_config
from aodn_cloud_optimised.lib.CommonHandler import cloud_optimised_creation

dataset_config = load_dataset_config(str(importlib.resources.path("aodn_cloud_optimised.config.dataset", "anfog_slocum_glider.json")))

cloud_optimised_creation('object/path/netcdf_file.nc',
                          dataset_config=dataset_config
                         )

Parquet GenericHandler - handler steps

The conversion process in GenericHandler is broken down into a series of ordered steps, each responsible for a specific task. These steps include:

1. delete_existing_matching_parquet: Deletes existing Parquet files that match the current processing criteria.

2. preprocess_data: Generates a DataFrame and Dataset from the input NetCDF file.

3. publish_cloud_optimised: Creates Parquet files containing the processed data.

   • _add_timestamp_df: Adds timestamp information to the DataFrame. Useful for partitioning.
   • _add_columns_df: Adds generic columns such as site_code and filename to the DataFrame.
   • _add_columns_df_custom Adds custom columns (useful for specific handlers)
   • _rm_bad_timestamp_df: Removes rows with bad timestamps from the DataFrame.
   • _add_metadata_sidecar: Adds metadata from the PyArrow table to the xarray dataset as sidecar attributes.

4. postprocess: Cleans up resources used during data processing.

Dataset configuration

Every dataset should be configured with a config JSON file. A template exists at aodn_cloud_optimised/config/dataset/dataset_template.json

See documentation to learn how to add a new dataset

Notebooks

Notebooks exist under https://github.com/aodn/architecturereview/blob/main/cloud-optimised/cloud-optimised-team/parquet/notebooks/

For each new dataset, it is a good practice to use the provided template cloud-optimised/cloud-optimised-team/parquet/notebooks/template.ipynb and create a new notebook.

These notebooks use a common library of python functions to help with creating the geo-spatial filters: cloud-optimised/cloud-optimised-team/parquet/notebooks/parquet_queries.py