SparkSession & SparkContext
Apache Spark · Spark Architecture
Theory
SparkSession is the single entry point for modern Spark code. You get one with SparkSession.builder...getOrCreate(), and through it you read data (spark.read), run SQL (spark.sql), create DataFrames, set runtime configuration (spark.conf) and reach the catalog of tables and views. In notebooks and the pyspark shell a session named spark already exists.
SparkContext is the older, lower-level entry point from the RDD era. It represents the connection to the cluster and is still there underneath as spark.sparkContext, used for RDDs, broadcast variables, accumulators and job groups. There is one SparkContext per application, but you can create several SparkSessions on it with spark.newSession(); they share the cluster and cached data but have their own temp views and SQL settings. For example, a notebook can call spark.sparkContext.parallelize([...]) for an RDD demo while all DataFrame work goes through spark. With Spark Connect, only the SparkSession is available on the client.
Why we use it
Before Spark 2.0 you juggled SparkContext, SQLContext and HiveContext. SparkSession unifies them so one object gives you DataFrames, SQL, configuration and the catalog, which keeps code short and consistent.
When to use it
Use SparkSession for everything new: reading, SQL, DataFrames, config. Reach for spark.sparkContext only for RDD APIs, broadcast variables, accumulators or job groups, and remember that Spark Connect clients cannot use it.
How it works
- Program starts
- Is a session already running (notebook, shell)? Yes: getOrCreate() returns it. No: Builder creates SparkContext and SparkSession.
- Read data and run SQL through spark
- Use spark.sparkContext only for RDD-level APIs
- spark.stop() ends the application
Syntax
spark = (SparkSession.builder.appName("app")
.config("key", "value").getOrCreate())
sc = spark.sparkContext
spark.conf.set("spark.sql.shuffle.partitions", "64")
other = spark.newSession()Example
from pyspark.sql import SparkSession, functions as F
spark = (SparkSession.builder.master("local[1]")
.config("spark.ui.enabled", "false")
.getOrCreate())
orders = spark.createDataFrame([
(101, 1, "North", "Laptop", 1, 55000.0, "2026-01-05"),
(102, 2, "West", "Phone", 2, 30000.0, "2026-01-06"),
(103, 1, "North", "Mouse", 3, 1500.0, "2026-01-06"),
(104, 3, "South", "Laptop", 1, 52000.0, "2026-01-07"),
(105, 4, "West", "Tablet", 1, 25000.0, "2026-01-08"),
(106, 2, "East", "Phone", 1, 15000.0, "2026-01-08"),
(107, 5, "North", "Tablet", 2, 48000.0, "2026-01-09"),
(108, 3, "South", "Mouse", 4, 2000.0, "2026-01-10"),
], "order_id INT, customer_id INT, region STRING, product STRING, "
"qty INT, amount DOUBLE, order_date STRING"
).withColumn("order_date", F.to_date("order_date"))
sc = spark.sparkContext # the lower-level context underneath
rdd = sc.parallelize([1, 2, 3, 4])
print("RDD sum:", rdd.sum())
spark.conf.set("spark.sql.shuffle.partitions", "8")
print("shuffle partitions:", spark.conf.get("spark.sql.shuffle.partitions"))
orders.createOrReplaceTempView("orders")
other = spark.newSession() # same SparkContext, separate SQL state
print("same SparkContext:", other.sparkContext is sc)
print("view visible in spark:", spark.catalog.tableExists("orders"))
print("view visible in other:", other.catalog.tableExists("orders"))
print("other shuffle partitions:", other.conf.get("spark.sql.shuffle.partitions"))
# Output:
# RDD sum: 10
# shuffle partitions: 8
# same SparkContext: True
# view visible in spark: True
# view visible in other: False
# other shuffle partitions: 200
Key points
- SparkSession is the entry point for DataFrames, SQL, config and catalog
- getOrCreate() reuses an existing session instead of creating a new one
- SparkContext (spark.sparkContext) is for RDDs, broadcasts, accumulators
- One SparkContext per application; newSession() shares it but isolates temp views and SQL conf
- Spark Connect clients expose SparkSession only, no SparkContext