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") or dataFrame.unpersist() 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.

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.

Join Strategy Hints for SQL Queries

The join strategy hints, namely BROADCAST, MERGE, SHUFFLE_HASH and SHUFFLE_REPLICATE_NL, instruct Spark to use the hinted strategy on each specified relation when joining them with another relation. For example, when the BROADCAST hint is used on table ‘t1’, broadcast join (either broadcast hash join or broadcast nested loop join depending on whether there is any equi-join key) with ‘t1’ as the build side will be prioritized by Spark even if the size of table ‘t1’ suggested by the statistics is above the configuration spark.sql.autoBroadcastJoinThreshold.

When different join strategy hints are specified on both sides of a join, Spark prioritizes the BROADCAST hint over the MERGE hint over the SHUFFLE_HASH hint over the SHUFFLE_REPLICATE_NL hint. When both sides are specified with the BROADCAST hint or the SHUFFLE_HASH hint, Spark will pick the build side based on the join type and the sizes of the relations.

Note that there is no guarantee that Spark will choose the join strategy specified in the hint since a specific strategy may not support all join types.

spark.table("src").join(spark.table("records").hint("broadcast"), "key").show()

spark.table("src").join(spark.table("records").hint("broadcast"), "key").show();

spark.table("src").join(spark.table("records").hint("broadcast"), "key").show()

src <- sql("SELECT * FROM src")
records <- sql("SELECT * FROM records")
head(join(src, hint(records, "broadcast"), 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

For more details please refer to the documentation of Join Hints.

Coalesce Hints for SQL Queries

Coalesce hints allow Spark SQL users to control the number of output files just like coalesce, repartition and repartitionByRange in the Dataset API, they can be used for performance tuning and reducing the number of output files. The “COALESCE” hint only has a partition number as a parameter. The “REPARTITION” hint has a partition number, columns, or both/neither of them as parameters. The “REPARTITION_BY_RANGE” hint must have column names and a partition number is optional. The “REBALANCE” hint has an initial partition number, columns, or both/neither of them as parameters.

SELECT /*+ COALESCE(3) */ * FROM t;
SELECT /*+ REPARTITION(3) */ * FROM t;
SELECT /*+ REPARTITION(c) */ * FROM t;
SELECT /*+ REPARTITION(3, c) */ * FROM t;
SELECT /*+ REPARTITION */ * FROM t;
SELECT /*+ REPARTITION_BY_RANGE(c) */ * FROM t;
SELECT /*+ REPARTITION_BY_RANGE(3, c) */ * FROM t;
SELECT /*+ REBALANCE */ * FROM t;
SELECT /*+ REBALANCE(3) */ * FROM t;
SELECT /*+ REBALANCE(c) */ * FROM t;
SELECT /*+ REBALANCE(3, c) */ * FROM t;

For more details please refer to the documentation of Partitioning Hints.

Adaptive Query Execution

Adaptive Query Execution (AQE) is an optimization technique in Spark SQL that makes use of the runtime statistics to choose the most efficient query execution plan, which is enabled by default since Apache Spark 3.2.0. Spark SQL can turn on and off AQE by spark.sql.adaptive.enabled as an umbrella configuration. As of Spark 3.0, there are three major features in AQE: including coalescing post-shuffle partitions, converting sort-merge join to broadcast join, and skew join optimization.

Coalescing Post Shuffle Partitions

This feature coalesces the post shuffle partitions based on the map output statistics when both spark.sql.adaptive.enabled and spark.sql.adaptive.coalescePartitions.enabled configurations are true. This feature simplifies the tuning of shuffle partition number when running queries. You do not need to set a proper shuffle partition number to fit your dataset. Spark can pick the proper shuffle partition number at runtime once you set a large enough initial number of shuffle partitions via spark.sql.adaptive.coalescePartitions.initialPartitionNum configuration.

Spliting skewed shuffle partitions

Converting sort-merge join to broadcast join

AQE converts sort-merge join to broadcast hash join when the runtime statistics of any join side is smaller than the adaptive broadcast hash join threshold. This is not as efficient as planning a broadcast hash join in the first place, but it’s better than keep doing the sort-merge join, as we can save the sorting of both the join sides, and read shuffle files locally to save network traffic(if spark.sql.adaptive.localShuffleReader.enabled is true)

Converting sort-merge join to shuffled hash join

AQE converts sort-merge join to shuffled hash join when all post shuffle partitions are smaller than a threshold, the max threshold can see the config spark.sql.adaptive.maxShuffledHashJoinLocalMapThreshold.

Optimizing Skew Join

Data skew can severely downgrade the performance of join queries. This feature dynamically handles skew in sort-merge join by splitting (and replicating if needed) skewed tasks into roughly evenly sized tasks. It takes effect when both spark.sql.adaptive.enabled and spark.sql.adaptive.skewJoin.enabled configurations are enabled.

Misc