Spark Custom Kafka Partitioner

Custom Partitoner can be implemented by extending org.apache.kafka.clients.producer.Partitioner.

This can be used with Spark-SQL Kafka Data Source by setting property "kafka.partitioner.class"

For example

df.write.format("kafka").option("kafka.partitioner.class", "com.mycustom.ipartitioner")

We implemented one such custom partitioner extending org.apache.kafka.clients.producer.RoundRobinPartitioner.

Complete Source code is available @ https://github.com/dinesh028/engineering/blob/master/Kafka/com/aquaifer/producer/KeyPartitioner.scala

This paritioner -

Reads a configuration file which has Kafka Key and PrimaryKey Name mapping. Value in Kafka is a JSON Message which has a Primary Key with unique Value.
Idea is to partition messages based on this unique value, such that messages with same value for primarykey go into same partition.
Once, configurations are loaded. For each byte array message-

convert it to String JSON
Parse JSON
Get unique value for PrimaryKey.
Do Modulus of Postive Hash Code of Value by available partition in Kafka Topic.
The output of Modulus determines the partition for given message.

Also, we noticed that custom properties are not passed to Partitioner configurations, w.r.t same - we did raise a defect with Spark. Refer - https://issues.apache.org/jira/browse/SPARK-45666

Comments

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...

Scala Spark building Jar leads java.lang.StackOverflowError

Exception - [Thread-3] ERROR scala_maven.ScalaCompileMojo - error: java.lang.StackOverflowError [Thread-3] INFO scala_maven.ScalaCompileMojo - at scala.collection.generic.TraversableForwarder$class.isEmpty(TraversableForwarder.scala:36) [Thread-3] INFO scala_maven.ScalaCompileMojo - at scala.collection.mutable.ListBuffer.isEmpty(ListBuffer.scala:45) [Thread-3] INFO scala_maven.ScalaCompileMojo - at scala.collection.mutable.ListBuffer.toList(ListBuffer.scala:306) [Thread-3] INFO scala_maven.ScalaCompileMojo - at scala.collection.mutable.ListBuffer.result(ListBuffer.scala:300) [Thread-3] INFO scala_maven.ScalaCompileMojo - at scala.collection.mutable.Stack$StackBuilder.result(Stack.scala:31) [Thread-3] INFO scala_maven.ScalaCompileMojo - at scala.collection.mutable.Stack$StackBuilder.result(Stack.scala:27) [Thread-3] INFO scala_maven.ScalaCompileMojo - at scala.collection.generic.GenericCompanion.apply(GenericCompanion.scala:50) [Thread-3] INFO scala_maven.ScalaCompile...

MongoDB Chunk size many times bigger than configure chunksize (128 MB)

Shard Shard_0 at Shard_0/xyz.com:27018 { data: '202.04GiB', docs: 117037098, chunks: 5, 'estimated data per chunk': '40.4GiB', 'estimated docs per chunk': 23407419 } --- Shard Shard_1 at Shard_1/abc.com:27018 { data: '201.86GiB', docs: 116913342, chunks: 4, 'estimated data per chunk': '50.46GiB', 'estimated docs per chunk': 29228335 } Per MongoDB- Starting in 6.0.3, we balance by data size instead of the number of chunks. So the 128MB is now only the size of data we migrate at-a-time. So large data size per chunk is good now, as long as the data size per shard is even for the collection. refer - https://www.mongodb.com/community/forums/t/chunk-size-many-times-bigger-than-configure-chunksize-128-mb/212616 https://www.mongodb.com/docs/v6.0/release-notes/6.0/#std-label-release-notes-6.0-balancing-policy-changes

AWS EMR Spark – Much Larger Executors are Created than Requested

Starting EMR 5.32 and EMR 6.2 you can notice that Spark can launch much larger executors that you request in your job settings. For example - We started a Spark Job with spark.executor.cores = 4 But, one can see that the executors with 20 cores (instead of 4 as defined by spark.executor.cores) were launched. The reason for allocating larger executors is that there is a AWS specific Spark option spark.yarn.heterogeneousExecutors.enabled (exists in EMR only, does not exist in Open Source Spark) that is set to true by default that combines multiple executor creation requests on the same node into a larger executor container. So as the result you have fewer executor containers than you expected, each of them has more memory and cores that you specified. If you disable this option (--conf "spark.yarn.heterogeneousExecutors.enabled=false"), EMR will create containers with the specified spark.executor.memory and spark.executor.cores settings and will not co...

Hive Count Query not working

Hive with Tez execution engine - count(*) not working , returning 0 results. Solution - set hive.compute.query.using.stats=false Refer - https://cwiki.apache.org/confluence/display/Hive/Configuration+Properties hive.compute.query.using.stats Default Value: false Added In: Hive 0.13.0 with HIVE-5483 When set to true Hive will answer a few queries like min, max, and count(1) purely using statistics stored in the metastore. For basic statistics collection, set the configuration property hive.stats.autogather to true. For more advanced statistics collection, run ANALYZE TABLE queries.

QueryDB

Search This Blog