Install Sedona R
apache.sedona (cran.r-project.org/package=apache.sedona) is a sparklyr-based R interface for Apache Sedona. It presents what Apache Sedona has to offer through idiomatic frameworks and constructs in R (e.g., one can build spatial Spark SQL queries using Sedona UDFs in conjunction with a wide range of dplyr expressions), hence making Apache Sedona highly friendly for R users.
Generally speaking, when working with Apache Sedona, one choose between the following two modes:
- Manipulating Sedona Spatial Resilient Distributed Datasets with spatial-RDD-related routines
- Querying geometric columns within Spatial dataframes with Sedona spatial UDFs
While the former option enables more fine-grained control over low-level
implementation details (e.g., which index to build for spatial queries,
which data structure to use for spatial partitioning, etc), the latter
is simpler and leads to a straightforward integration with
sparklyr, and other
sparklyr extensions (e.g., one can build ML
feature extractors with Sedona UDFs and connect them with ML pipelines
ml_*() family of functions in
sparklyr, hence creating ML
workflows capable of understanding spatial data).
Because data from spatial RDDs can be imported into Spark dataframes as geometry columns and vice versa, one can switch between the abovementioned two modes fairly easily.
At the moment
apache.sedona consists of the following components:
- R interface for Spatial-RDD-related functionalities
- Reading/writing spatial data in WKT, WKB, and GeoJSON formats
- Shapefile reader
- Spatial partition, index, join, KNN query, and range query operations
- Visualization routines
dplyr-integration for Sedona spatial UDTs and UDFs
- See SQL APIs for the list of available UDFs
- Functions importing data from spatial RDDs to Spark dataframes and vice versa
Connect to Spark¶
To ensure Sedona serialization routines, UDTs, and UDFs are properly
registered when creating a Spark session, one simply needs to attach
apache.sedona before instantiating a Spark conneciton. apache.sedona
will take care of the rest. For example,
library(sparklyr) library(apache.sedona) spark_home <- "/usr/lib/spark" # NOTE: replace this with your $SPARK_HOME directory sc <- spark_connect(master = "yarn", spark_home = spark_home)
will create a Sedona-capable Spark connection in YARN client mode, and
library(sparklyr) library(apache.sedona) sc <- spark_connect(master = "local")
will create a Sedona-capable Spark connection to an Apache Spark instance running locally.
sparklyr, one can easily inspect the Spark connection object to
sanity-check it has been properly initialized with all Sedona-related
##  "org.apache.sedona:sedona-core-3.0_2.12:1.3.1-incubating" ##  "org.apache.sedona:sedona-sql-3.0_2.12:1.3.1-incubating" ##  "org.apache.sedona:sedona-viz-3.0_2.12:1.3.1-incubating" ##  "org.datasyslab:geotools-wrapper:1.3.0-27.2" ##  "org.datasyslab:sernetcdf:0.1.0" ##  "org.locationtech.jts:jts-core:1.18.0" ##  "org.wololo:jts2geojson:0.14.3"
spark_session(sc) %>% invoke("%>%", list("conf"), list("get", "spark.kryo.registrator")) %>% print()
##  "org.apache.sedona.viz.core.Serde.SedonaVizKryoRegistrator"
For more information about connecting to Spark with
?sparklyr::spark_connect. Also see
Initiate Spark Context and Initiate Spark Session for
minimum and recommended dependencies for Apache Sedona.