Performance Tuning

For some workloads, it is possible to improve performance by either caching data in memory, or by turning on some experimental options.

Caching Data In Memory

Spark SQL can cache tables using an in-memory columnar format by calling spark.catalog.cacheTable("tableName") or dataFrame.cache(). Then Spark SQL will scan only required columns and will automatically tune compression to minimize memory usage and GC pressure. You can call spark.catalog.uncacheTable("tableName") to remove the table from memory.

Configuration of in-memory caching can be done using the setConf method on SparkSession or by running SET key=value commands using SQL.

Property NameDefaultMeaning
spark.sql.inMemoryColumnarStorage.compressed true When set to true Spark SQL will automatically select a compression codec for each column based on statistics of the data.
spark.sql.inMemoryColumnarStorage.batchSize 10000 Controls the size of batches for columnar caching. Larger batch sizes can improve memory utilization and compression, but risk OOMs when caching data.

Other Configuration Options

The following options can also be used to tune the performance of query execution. It is possible that these options will be deprecated in future release as more optimizations are performed automatically.

Property NameDefaultMeaning
spark.sql.files.maxPartitionBytes 134217728 (128 MB) The maximum number of bytes to pack into a single partition when reading files.
spark.sql.files.openCostInBytes 4194304 (4 MB) The estimated cost to open a file, measured by the number of bytes could be scanned in the same time. This is used when putting multiple files into a partition. It is better to over-estimated, then the partitions with small files will be faster than partitions with bigger files (which is scheduled first).
spark.sql.broadcastTimeout 300

Timeout in seconds for the broadcast wait time in broadcast joins

spark.sql.autoBroadcastJoinThreshold 10485760 (10 MB) Configures the maximum size in bytes for a table that will be broadcast to all worker nodes when performing a join. By setting this value to -1 broadcasting can be disabled. Note that currently statistics are only supported for Hive Metastore tables where the command ANALYZE TABLE <tableName> COMPUTE STATISTICS noscan has been run.
spark.sql.shuffle.partitions 200 Configures the number of partitions to use when shuffling data for joins or aggregations.
spark.sql.sources.parallelPartitionDiscovery.threshold 32 Configures the threshold to enable parallel listing for job input paths. If the number of input paths is larger than this threshold, Spark will list the files by using Spark distributed job. Otherwise, it will fallback to sequential listing. This configuration is only effective when using file-based data sources such as Parquet, ORC and JSON. 1.5.0
spark.sql.sources.parallelPartitionDiscovery.parallelism 10000 Configures the maximum listing parallelism for job input paths. In case the number of input paths is larger than this value, it will be throttled down to use this value. Same as above, this configuration is only effective when using file-based data sources such as Parquet, ORC and JSON. 2.1.1

Broadcast Hint for SQL Queries

The BROADCAST hint guides Spark to broadcast each specified table when joining them with another table or view. When Spark deciding the join methods, the broadcast hash join (i.e., BHJ) is preferred, even if the statistics is above the configuration spark.sql.autoBroadcastJoinThreshold. When both sides of a join are specified, Spark broadcasts the one having the lower statistics. Note Spark does not guarantee BHJ is always chosen, since not all cases (e.g. full outer join) support BHJ. When the broadcast nested loop join is selected, we still respect the hint.

import org.apache.spark.sql.functions.broadcast
broadcast(spark.table("src")).join(spark.table("records"), "key").show()
import static org.apache.spark.sql.functions.broadcast;
broadcast(spark.table("src")).join(spark.table("records"), "key").show();
from pyspark.sql.functions import broadcast
broadcast(spark.table("src")).join(spark.table("records"), "key").show()
src <- sql("SELECT * FROM src")
records <- sql("SELECT * FROM records")
head(join(broadcast(src), records, src$key == records$key))
-- We accept BROADCAST, BROADCASTJOIN and MAPJOIN for broadcast hint
SELECT /*+ BROADCAST(r) */ * FROM records r JOIN src s ON r.key = s.key