[
https://issues.apache.org/jira/browse/SPARK-54753?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
xihuan updated SPARK-54753:
---------------------------
Description:
For Apache Spark *4.0.1* local mode, the memory is not released after long time
running, while downgrade the spark version {*}v3.5.6{*}, there is no issue.
The issue can be reproduced with simple test cases:
```
package spark;
import org.apache.spark.sql.Dataset;
import org.apache.spark.sql.Row;
import org.apache.spark.sql.SparkSession;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import java.util.concurrent.Executors;
import java.util.concurrent.ScheduledExecutorService;
public class SparkApp {
private static final Logger log = LoggerFactory.getLogger(SparkApp.class);
private static final String SPARK_MASTER_URL = "local[4]";
private static final String SPARK_MEMORY = "500m";
public static void main(String[] args)
{ log.debug("Starting application..."); SparkSession sparkSession =
SparkSession.builder() .appName("Test Application") .master(SPARK_MASTER_URL)
.config("spark.driver.memory", SPARK_MEMORY)
//.config("spark.sql.sources.bucketing.autoBucketedScan.enabled", false) // To
get rid of the memory leak .getOrCreate(); processData(sparkSession); }
private static void processData(SparkSession sparkSession) {
while (true)
{ //load from a local csv file Dataset<Row> dataset =
sparkSession.read().csv("c:/City.csv"); dataset.show(5); log.debug("Persist
dataset..."); // This is the place where the memory leak occurs
dataset.persist(); // Do something... log.debug("Do something with the
persisted dataset"); // ...and unpersist the dataset log.debug("Unpersist
dataset..."); dataset.unpersist(); log.debug("Processing data completed"); }
}
}
```
with *Apache Spark 4.0.0 or 4.0.1* as JVM heap size to 1GB, around 9 minutes,
it will encounter OOM.
!screenshot-1.png|width=1070,height=504!
While, {*}with Apache Spark 3.5.6, with max heap size of 500MB{*}, no such
issue occurs.
!image-2025-12-23-05-38-53-324.png|width=1075,height=415!
was:
For Apache Spark *4.0.1* local mode, the memory is not released after long time
running, while downgrade the spark version {*}v3.5.6{*}, there is no issue.
The issue can be reproduced with simple test cases:
```
package spark;
import org.apache.spark.sql.Dataset;
import org.apache.spark.sql.Row;
import org.apache.spark.sql.SparkSession;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import java.util.concurrent.Executors;
import java.util.concurrent.ScheduledExecutorService;
public class SparkApp {
private static final Logger log = LoggerFactory.getLogger(SparkApp.class);
private static final String SPARK_MASTER_URL = "local[4]";
private static final String SPARK_MEMORY = "500m";
public static void main(String[] args)
{ log.debug("Starting application..."); SparkSession sparkSession =
SparkSession.builder() .appName("Test Application") .master(SPARK_MASTER_URL)
.config("spark.driver.memory", SPARK_MEMORY)
//.config("spark.sql.sources.bucketing.autoBucketedScan.enabled", false) // To
get rid of the memory leak .getOrCreate(); processData(sparkSession); }
private static void processData(SparkSession sparkSession) {
while (true)
{ //load from a local csv file Dataset<Row> dataset =
sparkSession.read().csv("c:/City.csv"); dataset.show(5); log.debug("Persist
dataset..."); // This is the place where the memory leak occurs
dataset.persist(); // Do something... log.debug("Do something with the
persisted dataset"); // ...and unpersist the dataset log.debug("Unpersist
dataset..."); dataset.unpersist(); log.debug("Processing data completed"); }
}
}
```
with Apache Spark 4.0.0 or 4.0.1, around 9 minutes, will encounter oom.
!screenshot-1.png!
While, with Apache Spark 3.5.6, no such issue occurs.
!image-2025-12-23-05-38-53-324.png|width=637,height=246!
> memory leak in Apache Spark 4.0.1
> ---------------------------------
>
> Key: SPARK-54753
> URL: https://issues.apache.org/jira/browse/SPARK-54753
> Project: Spark
> Issue Type: Bug
> Components: Spark Core
> Affects Versions: 4.0.1
> Reporter: xihuan
> Priority: Critical
> Attachments: image-2025-12-23-05-38-53-324.png, pom.xml,
> screenshot-1.png
>
>
> For Apache Spark *4.0.1* local mode, the memory is not released after long
> time running, while downgrade the spark version {*}v3.5.6{*}, there is no
> issue.
> The issue can be reproduced with simple test cases:
> ```
> package spark;
> import org.apache.spark.sql.Dataset;
> import org.apache.spark.sql.Row;
> import org.apache.spark.sql.SparkSession;
> import org.slf4j.Logger;
> import org.slf4j.LoggerFactory;
> import java.util.concurrent.Executors;
> import java.util.concurrent.ScheduledExecutorService;
> public class SparkApp {
> private static final Logger log = LoggerFactory.getLogger(SparkApp.class);
> private static final String SPARK_MASTER_URL = "local[4]";
> private static final String SPARK_MEMORY = "500m";
> public static void main(String[] args)
> { log.debug("Starting application..."); SparkSession sparkSession =
> SparkSession.builder() .appName("Test Application") .master(SPARK_MASTER_URL)
> .config("spark.driver.memory", SPARK_MEMORY)
> //.config("spark.sql.sources.bucketing.autoBucketedScan.enabled", false) //
> To get rid of the memory leak .getOrCreate(); processData(sparkSession); }
> private static void processData(SparkSession sparkSession) {
> while (true)
> { //load from a local csv file Dataset<Row> dataset =
> sparkSession.read().csv("c:/City.csv"); dataset.show(5); log.debug("Persist
> dataset..."); // This is the place where the memory leak occurs
> dataset.persist(); // Do something... log.debug("Do something with the
> persisted dataset"); // ...and unpersist the dataset log.debug("Unpersist
> dataset..."); dataset.unpersist(); log.debug("Processing data completed"); }
> }
> }
> ```
>
> with *Apache Spark 4.0.0 or 4.0.1* as JVM heap size to 1GB, around 9 minutes,
> it will encounter OOM.
> !screenshot-1.png|width=1070,height=504!
>
> While, {*}with Apache Spark 3.5.6, with max heap size of 500MB{*}, no such
> issue occurs.
> !image-2025-12-23-05-38-53-324.png|width=1075,height=415!
>
>
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
