diff --git a/python/pyspark/sql/connect/client/core.py b/python/pyspark/sql/connect/client/core.py
index 1bd9725a9a7f0..461b68bb48d59 100644
--- a/python/pyspark/sql/connect/client/core.py
+++ b/python/pyspark/sql/connect/client/core.py
@@ -313,7 +313,7 @@ class DefaultChannelBuilder(ChannelBuilder):
 
     @staticmethod
     def default_port() -> int:
-        if "SPARK_TESTING" in os.environ and not is_remote_only():
+        if "SPARK_LOCAL_REMOTE" in os.environ and not is_remote_only():
             from pyspark.sql.session import SparkSession as PySparkSession
 
             # In the case when Spark Connect uses the local mode, it starts the regular Spark
diff --git a/python/pyspark/sql/connect/session.py b/python/pyspark/sql/connect/session.py
index c01c1e42a3185..fee9f9fdc88f9 100644
--- a/python/pyspark/sql/connect/session.py
+++ b/python/pyspark/sql/connect/session.py
@@ -1045,6 +1045,8 @@ def _start_connect_server(master: str, opts: Dict[str, Any]) -> None:
             overwrite_conf = opts
             overwrite_conf["spark.master"] = master
             overwrite_conf["spark.local.connect"] = "1"
+            # Always use an ephemeral port for local connect server
+            overwrite_conf["spark.connect.grpc.binding.port"] = "0"
             os.environ["SPARK_LOCAL_CONNECT"] = "1"
 
             # Configurations to be set if unset.
@@ -1054,11 +1056,6 @@ def _start_connect_server(master: str, opts: Dict[str, Any]) -> None:
                 "spark.sql.artifact.isolation.alwaysApplyClassloader": "true",
             }
 
-            if "SPARK_TESTING" in os.environ:
-                # For testing, we use 0 to use an ephemeral port to allow parallel testing.
-                # See also SPARK-42272.
-                overwrite_conf["spark.connect.grpc.binding.port"] = "0"
-
             origin_remote = os.environ.get("SPARK_REMOTE", None)
             try:
                 if origin_remote is not None: