WherobotsRunOperator¶

You can submit your scripts for execution on the Wherobots Cloud. A Job Run tracks the status of that script's execution. When implemented, an Airflow TaskInstance executes a Job Run and waits for it to complete.

For a comprehensive list of the parameters associated with Job Runs, see Runs REST API.

Benefits¶

You can use Airflow to streamline, automate, and manage complex ETL workload tasks that are running on your vector or raster based data. By using WherobotsRunOperator, you can incorporate Wherobots’ geospatial data processing features into your Airflow workflows.

With WherobotsRunOperator, you no longer need to manage and configure your own cluster to complete your geospatial data processing. Wherobots can handle that resource allocation for you, while providing analytics capabilities that are optimized for geospatial concerns.

Before you start¶

Before using WherobotsRunOperator, ensure that you have the following required resources:

An account within a Community, Professional, or Enterprise Edition Organization. For more information, see Create a Wherobots Account.
Wherobots API key. For more information, see API keys in the Wherobots documentation.
The Wherobots Apache Airflow Provider. For installation information, see Wherobots Apache Airflow Provider.
An Airflow Connection. For more information, see Create a new Connection in Airflow Server.
Python version ≥ 3.9
Apache Airflow. For more information, see Installation of Airflow in the Apache Airflow documentation.

DAG files¶

Directed Acyclic Graph (DAG) files are Python scripts that define your Airflow workflows. DAG files need to be accessible to the Airflow scheduler so that your tasks can be executed. DAG files are commonly stored in $AIRFLOW_HOME/dags. For more information, see Setting Configuration Options in the Apache Airflow Documentation.

WherobotsRunOperator constructor¶

The following example details how to integrate WherobotsRunOperator into your DAG file.

In this documentation, we discuss the WherobotsRunOperator-related lines of code which are highlighted below. For more information on the general formatting of Airflow DAG files, see Best Practices in the Airflow Documentation.

Example DAG file
import datetime
import pendulum

from airflow import DAG
from airflow_providers_wherobots.operators.run import WherobotsRunOperator

from wherobots.db.runtime import Runtime

with DAG(
    dag_id="test_run_operator",
    schedule="@once",
    start_date=pendulum.datetime(2021, 1, 1, tz="UTC"),
    catchup=False,
    tags=["example"],
) as test_run_dag:
    operator = WherobotsRunOperator(
        task_id="test_run_smoke",
        name="airflow_operator_test_run_{{ ts_nodash }}",
        runtime=Runtime.TINY,
        run_python={
            "uri": "S3-PATH-TO-YOUR-FILE"
        },
        dag=test_run_dag,
        poll_logs=True,
    )

Can I use any of the runtimes listed in Accepted values?

You might not have access to all of the available runtimes. For example, Community Edition Organizations can only use the Tiny runtime. You can see the runtimes available to your Organization within the Start a Notebook dropdown in Wherobots Cloud.

Parameter	Type	Description	Accepted values
`operator`		`WherobotsRunOperator` instantiates `WherobotsRunOperator`.
`name`	`str`	The name of the run. If not specified, a default name will be generated.
`runtime`	`enum`	Specifies the Wherobots runtime. `runtime` is optional. Defaults to `Runtime.TINY` for Community Edition and `Runtime.SMALL` for Professional Edition. See Runs REST API for more information.	`Runtime.TINY`, `Runtime.SMALL`, `Runtime.MEDIUM`, `Runtime.LARGE`, `Runtime.X_LARGE`, `Runtime.XX_LARGE`, `Runtime.XXXX_LARGE`, `Runtime.MEDIUM_HIMEM`, `Runtime.LARGE_HIMEM`, `Runtime.X_LARGE_HIMEM`, `Runtime.XX_LARGE_HIMEM`, `Runtime.XXXX_LARGE_HIMEM`, `Runtime.TINY_A10_GPU`, `Runtime.SMALL_A10_GPU`, `Runtime.MEDIUM_A10_GPU` You can see the runtimes available to your Organization within the Start a Notebook dropdown in Wherobots Cloud.
`run_python`	`dict`	Provide the Python file's S3 path. Wherobots Managed Storage and Storage Integration S3 paths are supported. You can use either `run_python` or `run_jar`, not both.	Takes the following keys: `uri:`(`str`) and `args:` (`list[str]`).
`poll_logs`	`bool`	Enables Log polling when set to `True`.	`True`, `False`

Additional parameters for Job Runs

There are additional parameters for the Runs API. For an exhaustive list of parameters and their accepted values see Runs REST API.

Other parameters¶

The following parameters aren't included in the example but are also compatible with the WherobotsRunsOperator.

Parameter	Type	Description	Accepted values
`run_jar`	`dict`	Wherobots Managed Storage and Storage Integration S3 paths are supported. You can use either `run_python` or `run_jar`, not both.	Takes the following keys: `uri:`(`str`), `args:` (`list[str]`), `mainClass:` (`str`).
`polling_interval`	`int`	The interval in seconds to poll the status of the run. The default value is `30`.
`environment`	`dict`	The model for runtime environment configs, including Spark cluster configs and dependencies. Defaults to `{}`. For more information see Environment keys.

Environment keys¶

Environment parameter	Type	Required or Optional	Parameter description
`sparkDriverDiskGB`	`int`	Optional	The driver disk size of the Spark cluster.
`sparkExecutorDiskGB:`	`int`	Optional	The executor disk size of the Spark cluster.
`sparkConfigs`	`dict{str:str}`	Optional	The user specified Spark configs.
`dependencies`	`list[dict]`	Optional	Indicates the 3^rd party dependencies need to add to the runtime. Required if adding `sparkConfigs`. Must be an array (list) of objects (dictionaries). Each object must represent either a Python Package Index or a File Dependency.

Dependencies¶

The following details information on the third-party dependencies that can be added to the runtime environment.

Parameter	Type	Details	Rules for accepted values
`sourceType`	string	Enum for the source type of dependency	Valid values: `PYPI` or `FILE`.
`filePath`	string	The file path to the dependency file. Must be a valid file path accessible within the runtime environment. The file extension must be one of: `.jar`, `.whl`, or `.zip`. Can only be used with the `FILE` `sourceType`.
`libraryName`	string	The python package name. Can only be used with the `PYPI` `sourceType`.
`libraryVersion`	string	The Python package version. Can only be used with the `PYPI` `sourceType`.

JAR and Python Scripts¶

You can use Python or JAR files to further leverage Wherobots' geospatial features and optimizations in conjunction with an Airflow DAG.

Tile generation¶

This example uses Sedona to process and generate vector tiles. The example reads buildings and roads data from Wherobots open datasets, filters the data based on a specified region, and then generates vector tiles using the wherobots.vtiles library. It writes the tiles to a PMTiles file in the user's S3 storage and displays a sample of the tiles.

Tile generation example Python script
from sedona.spark import *
from wherobots import vtiles
from sedona.sql.st_constructors import ST_GeomFromText
from sedona.sql.st_predicates import ST_Intersects
import pyspark.sql.functions as f
import os
import pyspark.sql.functions as f

config = SedonaContext.builder().getOrCreate()
sedona = SedonaContext.create(config)

# Set to False to generate tiles for the entire dataset, True to generate only for region_wkt area
filter = True
region_wkt = "POLYGON ((-122.097931 47.538528, -122.048836 47.566566, -121.981888 47.510012, -122.057076 47.506302, -122.097931 47.538528))"
filter_expression = ST_Intersects(f.col("geometry"), ST_GeomFromText(f.lit(region_wkt)))

buildings_df = (
    sedona.table("wherobots_open_data.overture_2024_02_15.buildings_building")
    .select(
        f.col("geometry"),
        f.lit("buildings").alias("layer"),
        f.element_at(f.col("sources"), 1).dataset.alias("source")
    )
)

buildings_df.show()

roads_df = (
    sedona.table("wherobots_open_data.overture_2024_02_15.transportation_segment")
    .select(
        f.col("geometry"),
        f.lit("roads").alias("layer"),
        f.element_at(f.col("sources"), 1).dataset.alias("source")
    )
)

roads_df.show()

features_df = roads_df.union(buildings_df)

if filter:
    features_df = features_df.filter(ST_Intersects(f.col("geometry"), ST_GeomFromText(f.lit(region_wkt))))

features_df.count()

tiles_df = vtiles.generate(features_df)

tiles_df.show(3, 150, True)

full_tiles_path = os.getenv("USER_S3_PATH") + "tiles.pmtiles"
vtiles.write_pmtiles(tiles_df, full_tiles_path, features_df=features_df)

vtiles.show_pmtiles(full_tiles_path)

sample_tiles_path = os.getenv("USER_S3_PATH") + "sampleTiles.pmtiles"
vtiles.generate_quick_pmtiles(features_df, sample_tiles_path)

Raster inference¶

This example performs batch image classification using Sedona and Wherobots. It reads raster data from an S3 bucket, applies a classification model to each image, extracts the most confident prediction for each image, and displays a few sample images along with their predicted labels.

Not available to Community Edition Organizations

Raster inference is not available to Community Edition Organizations. To perform raster inference, you need access to Wherobots' GPU-Optimized runtimes. For more information on GPU-Optimized runtimes, see Runtimes. For more information on the different Organization plans available, see Wherobots Pricing.

Raster inference example script
from pyspark.sql.functions import col
from sedona.raster_utils.SedonaUtils import SedonaUtils
from pyspark.sql.functions import rand
from wherobots.inference.engine.register import create_single_label_classification_udfs
from wherobots.inference.data.io import read_raster_table
from sedona.spark import SedonaContext
from pyspark.sql.functions import expr

import warnings

warnings.filterwarnings('ignore')

config = SedonaContext.builder().appName('classification-batch-inference')\
    .getOrCreate()

sedona = SedonaContext.create(config)

tif_folder_path = 's3a://wherobots-examples/data/eurosat_small'
files_df = read_raster_table(tif_folder_path, sedona)
df_raster_input = files_df.withColumn(
        "outdb_raster", expr("RS_FromPath(path)")
    )
df_raster_input.cache().count()
df_raster_input.show(truncate=False)

# %%time
df_raster_input.createOrReplaceTempView("df_raster_input")
model_id = 'landcover-eurosat-sentinel2'
predictions_df = sedona.sql(f"SELECT name, outdb_raster, RS_CLASSIFY('{model_id}', outdb_raster) AS preds FROM df_raster_input")
predictions_df.cache().count()
predictions_df.show(truncate=False)
predictions_df.createOrReplaceTempView("predictions_df")

max_predictions_df = sedona.sql(f"SELECT name, outdb_raster, RS_MAX_CONFIDENCE(preds).max_confidence_label, RS_MAX_CONFIDENCE(preds).max_confidence_score FROM predictions_df")
max_predictions_df.show(20, truncate=False)

rs_classify, rs_max_confidence = create_single_label_classification_udfs(batch_size = 10, sedona=sedona)
df_predictions = df_raster_input.withColumn("preds", rs_classify(model_id, 'outdb_raster'))
df_predictions.show(1)

df_max_predictions = df_predictions.withColumn("max_confidence_temp", rs_max_confidence(col("preds"))) \
                            .withColumn("max_confidence_label", col("max_confidence_temp.max_confidence_label")) \
                            .withColumn("max_confidence_score", col("max_confidence_temp.max_confidence_score")) \
                            .drop("max_confidence_temp", "preds")
df_max_predictions.cache().count()
df_max_predictions.show(2, truncate=False)

df_rast = sedona.read.format("binaryFile").option("pathGlobFilter", "*.tif").option("recursiveFileLookup", "true").load(tif_folder_path).selectExpr("RS_FromGeoTiff(content) as raster")

htmlDF = df_max_predictions.selectExpr("RS_Band(outdb_raster, Array(4, 3, 2)) as image_raster", "name", "max_confidence_label")\
    .selectExpr("RS_NormalizeAll(image_raster, 1, 65535, True) as image_raster", "name", "max_confidence_label")\
    .selectExpr("RS_AsImage(image_raster, 500) as image_raster", "name", "max_confidence_label")

SedonaUtils.display_image(htmlDF.orderBy(rand()).limit(3))

Reviewing logs¶

WherobotsRunOperator removes the need to write your own logic to poll the Wherobots Job Run logs. Airflow users can access the logs by specifying poll_logs=true in a DAG. Wherobots polls the logs and streams them in the Airflow Task logs.

Airflow UI¶

When you start your Airflow DAG from the Airflow server, logs can be seen in the Airflow Server's Logs tab. Relevant logs begin with the line INFO - === Logs for Run <run_id> Start:. The run_id can be found in the Airflow Server's XCom tab.

Limitations¶

WherobotsRunOperator has the following limitations:

Wherobots only stores logs for 30 days.
The amount of concurrent Job Runs that can occur are determined by your Organization’s maximum concurrent Spatial Unit consumption quota. For more information on Spatial Units and quota, see Runtimes.