Skip to main content



Spark Disable BroadCast Join not working in case of BroadcastNestedLoopJoin


We were running an application which was leading to below error - 

Caused by: org.apache.spark.SparkException: Could not execute broadcast in 300 secs. You can increase the timeout for broadcasts via spark.sql.broadcastTimeout or disable broadcast join by setting spark.sql.autoBroadcastJoinThreshold to -1
at org.apache.spark.sql.execution.exchange.BroadcastExchangeExec.doExecuteBroadcast(BroadcastExchangeExec.scala:150)
at org.apache.spark.sql.execution.SparkPlan$$anonfun$executeBroadcast$1.apply(SparkPlan.scala:154)
at org.apache.spark.sql.execution.SparkPlan$$anonfun$executeBroadcast$1.apply(SparkPlan.scala:150)

Even after setting the below property, which is to disable BroadCast Join. We kept getting above error again.

Set spark.sql.autoBroadcastJoinThreshold=-1


On further analysis, we found that it is not a Bug in Spark. You expect the broadcast to stop after you disable the broadcast threshold, by setting spark.sql.autoBroadcastJoinThreshold to -1, but Apache Spark tries to broadcast the bigger table and fails with a broadcast error.

Note that you can see "BroadcastNestedLoopJoin" in Spark WebUI or invoking ".explain(true)" on dataframe to visualize physical plan.


Solution -

To fix above problem, we replaced "not in" clause in our SQL with "not exists" clause. For example - 

select * from Table1 where id not in ( select if from table2)

rewritten as - 

select * from Table 1 where not exists ( select 1 from Table2 where table1.id=table2.id)


Explanation - 

In SQL, not in means that if there is any null value in the not in values, the result is empty. This is why it can only be executed with BroadcastNestedLoopJoin. All not in values must be known in order to ensure there is no null value in the set.


Similarly, 

Below example will result in to BroadcastNestedLoopJoin -

df1.join(df2, $"id1" === $"id2" || $"id2" === $"id3", "left")

rewritten as - 

df1.join(df2, $"id1" === $"id2", "left").join(df2, $"id2" === $"id3", "left")


Reference -

You can read details about same here - https://kb.databricks.com/sql/disable-broadcast-when-broadcastnestedloopjoin.html


Comments

Post a Comment

Popular posts

Spring MongoDB Rename field with derived Value of another field

Input Collection -  [ { 'k' : 'Troubleshooting' , 'hour' : '2024-10-10T16' , 'v' : [ 'WebPage, Login' ] }, { 'k' : 'TroubleshootingMe' , 'hour' : '2024-10-07T01' , 'v' : [ 'Accounts, etc' ] }  ] Expected Output -  [ { 'hour' : '2024-10-10T16' , 'Troubleshooting' : [ 'WebPage, Login' ] }, { 'hour' : '2024-10-07T01' , 'TroubleshootingMe' : [ 'Accounts, etc' ] }  ]   Above Can be achieved by  $replaceRoot / $replaceWith as follows - { $replaceWith : { $mergeObjects : [ { hour : "$hour" }, { "$arrayToObject" : [ [ { k : "$k" , v : "$v" } ] ] } ] } } or { $replaceRoo...
Post a Comment
Read more




What is Leadership

 
Post a Comment
Read more




Spark Error - missing part 0 of the schema, 2 parts are expected

 Exception -  Caused by: org.apache.spark.sql.AnalysisException : Could not read schema from the hive metastore because it is corrupted. (missing part 0 of the schema, 2 parts are expected).; Analysis -  ·          Check for table definition. In TBLProperties, you might find something like this – > spark.sql.sources.schema.numPartCols > 'spark.sql.sources.schema.numParts' 'spark.sql.sources.schema.part.0' > 'spark.sql.sources.schema.part.1' 'spark.sql.sources.schema.part.2' > 'spark.sql.sources.schema.partCol.0' > 'spark.sql.sources.schema.partCol.1' That’s what error seems to say that part1 is defined but part0 is missing.  Solution -  Drop & re-create table. If Table was partitioned  then all partitions  would have been removed. So do either of below -  ·          Msck repair table <db_name>.<table_name> ·    ...
Post a Comment
Read more




Spark MongoDB Connector Not leading to correct count or data while reading

  We are using Scala 2.11 , Spark 2.4 and Spark MongoDB Connector 2.4.4 Use Case 1 - We wanted to read a Shareded Mongo Collection and copy its data to another Mongo Collection. We noticed that after Spark Job successful completion. Output MongoDB did not had many records. Use Case 2 -  We read a MongoDB collection and doing count on dataframe lead to different count on each execution. Analysis,  We realized that MongoDB Spark Connector is missing data on bulk read as a dataframe. We tried various partitioner, listed on page -  https://www.mongodb.com/docs/spark-connector/v2.4/configuration/  But, none of them worked for us. Finally, we tried  MongoShardedPartitioner  this lead to constant count on each execution. But, it was greater than the actual count of records on the collection. This seems to be limitation with MongoDB Spark Connector. But,  MongoShardedPartitioner  seemed closest possible solution to this kind of situation. But, it per...
Post a Comment
Read more




Read from a hive table and write back to it using spark sql

In context to Spark 2.2 - if we read from an hive table and write to same, we get following exception- scala > dy . write . mode ( "overwrite" ). insertInto ( "incremental.test2" ) org . apache . spark . sql . AnalysisException : Cannot insert overwrite into table that is also being read from .; org . apache . spark . sql . AnalysisException : Cannot insert overwrite into table that is also being read from .; 1. This error means that our process is reading from same table and writing to same table. 2. Normally, this should work as process writes to directory .hiveStaging... 3. This error occurs in case of saveAsTable method, as it overwrites entire table instead of individual partitions. 4. This error should not occur with insertInto method, as it overwrites partitions not the table. 5. A reason why this happening is because Hive table has following Spark TBLProperties in its definition. This problem will solve for insertInto met...
Post a Comment
Read more