Github user zsxwing commented on a diff in the pull request:
https://github.com/apache/spark/pull/15102#discussion_r80154723
--- Diff:
external/kafka-0-10-sql/src/main/scala/org/apache/spark/sql/kafka010/KafkaSource.scala
---
@@ -0,0 +1,474 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements. See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.sql.kafka010
+
+import java.{util => ju}
+
+import scala.collection.JavaConverters._
+
+import org.apache.kafka.clients.consumer.{Consumer, ConsumerConfig,
KafkaConsumer}
+import
org.apache.kafka.clients.consumer.internals.NoOpConsumerRebalanceListener
+import org.apache.kafka.common.TopicPartition
+import org.apache.kafka.common.serialization.ByteArrayDeserializer
+
+import org.apache.spark.internal.Logging
+import org.apache.spark.scheduler.ExecutorCacheTaskLocation
+import org.apache.spark.sql._
+import org.apache.spark.sql.execution.streaming._
+import org.apache.spark.sql.kafka010.KafkaSource._
+import org.apache.spark.sql.sources.{DataSourceRegister,
StreamSourceProvider}
+import org.apache.spark.sql.types._
+import org.apache.spark.SparkContext
+
+/**
+ * A [[Source]] that uses Kafka's own [[KafkaConsumer]] API to reads data
from Kafka. The design
+ * for this source is as follows.
+ *
+ * - The [[KafkaSourceOffset]] is the custom [[Offset]] defined for this
source that contains
+ * a map of TopicPartition -> offset. Note that this offset is 1 +
(available offset). For
+ * example if the last record in a Kafka topic "t", partition 2 is
offset 5, then
+ * KafkaSourceOffset will contain TopicPartition("t", 2) -> 6. This is
done keep it consistent
+ * with the semantics of `KafkaConsumer.position()`.
+ *
+ * - The [[ConsumerStrategy]] class defines which Kafka topics and
partitions should be read
+ * by this source. These strategies directly correspond to the different
consumption options
+ * in . This class is designed to return a configured
+ * [[KafkaConsumer]] that is used by the [[KafkaSource]] to query for
the offsets.
+ * See the docs on
[[org.apache.spark.sql.kafka010.KafkaSource.ConsumerStrategy]] for
+ * more details.
+ *
+ * - The [[KafkaSource]] written to do the following.
+ *
+ * - As soon as the source is created, the pre-configured KafkaConsumer
returned by the
+ * [[ConsumerStrategy]] is used to query the initial offsets that this
source should
+ * start reading from. This used to create the first batch.
+ *
+ * - `getOffset()` uses the KafkaConsumer to query the latest available
offsets, which are
+ * returned as a [[KafkaSourceOffset]].
+ *
+ * - `getBatch()` returns a DF that reads from the 'start offset' until
the 'end offset' in
+ * for each partition. The end offset is excluded to be consistent
with the semantics of
+ * [[KafkaSourceOffset]] and `KafkaConsumer.position()`.
+ *
+ * - The DF returned is based on [[KafkaSourceRDD]] which is constructed
such that the
+ * data from Kafka topic + partition is consistently read by the same
executors across
+ * batches, and cached KafkaConsumers in the executors can be reused
efficiently. See the
+ * docs on [[KafkaSourceRDD]] for more details.
+ */
+private[kafka010] case class KafkaSource(
+ sqlContext: SQLContext,
+ consumerStrategy: ConsumerStrategy,
+ executorKafkaParams: ju.Map[String, Object],
+ sourceOptions: Map[String, String])
+ extends Source with Logging {
+
+ @transient private val consumer = consumerStrategy.createConsumer()
+ @transient private val sc = sqlContext.sparkContext
+ @transient private val initialPartitionOffsets =
fetchPartitionOffsets(seekToLatest = false)
+ logInfo(s"Initial offsets: " + initialPartitionOffsets)
+
+ override def schema: StructType = KafkaSource.kafkaSchema
+
+ /** Returns the maximum available offset for this source. */
+ override def getOffset: Option[Offset] = {
+ val offset = KafkaSourceOffset(fetchPartitionOffsets(seekToLatest =
true))
+ logDebug(s"GetOffset: $offset")
+ Some(offset)
+ }
+
+ /**
+ * Returns the data that is between the offsets [`start`, `end`), i.e.
end is exclusive.
+ */
+ override def getBatch(start: Option[Offset], end: Offset): DataFrame = {
+ logDebug(s"GetBatch called with start = $start, end = $end")
+ val untilPartitionOffsets = KafkaSourceOffset.getPartitionOffsets(end)
+ val fromPartitionOffsets = start match {
+ case Some(prevBatchEndOffset) =>
+ KafkaSourceOffset.getPartitionOffsets(prevBatchEndOffset)
+ case None =>
+ initialPartitionOffsets
+ }
+
+ val newPartitions =
untilPartitionOffsets.keySet.diff(fromPartitionOffsets.keySet)
+ val newPartitionOffsets = if (newPartitions.nonEmpty) {
+ fetchNewPartitionEarliestOffsets(newPartitions.toSeq)
+ } else {
+ Map.empty[TopicPartition, Long]
+ }
+
+ // Sort the partitions and current list of executors to consistently
assign each partition
+ // to the executor. This allows cached KafkaConsumers in the executors
to be re-used to
+ // read the same partition in every batch.
+ val topicPartitionOrdering = new Ordering[TopicPartition] {
+ override def compare(l: TopicPartition, r: TopicPartition): Int = {
+ implicitly[Ordering[(String, Long)]].compare(
+ (l.topic, l.partition),
+ (r.topic, r.partition))
+ }
+ }
+ val sortedTopicPartitions =
untilPartitionOffsets.keySet.toSeq.sorted(topicPartitionOrdering)
+ val sortedExecutors = getSortedExecutorList(sc)
+ val numExecutors = sortedExecutors.size
+ logDebug("Sorted executors: " + sortedExecutors.mkString(", "))
+ val offsetRanges = sortedTopicPartitions.map { tp =>
+ val fromOffset = fromPartitionOffsets.get(tp).getOrElse {
+ newPartitionOffsets.getOrElse(tp, {
+ // This should not happen since newPartitionOffsets contains all
paritions not in
+ // fromPartitionOffsets
+ throw new IllegalStateException(s"$tp doesn't have a offset")
+ })
+ }
+ val untilOffset = untilPartitionOffsets(tp)
+ val preferredLoc = if (numExecutors > 0) {
+ Some(sortedExecutors(positiveMod(tp.hashCode, numExecutors)))
+ } else None
+ KafkaSourceRDD.OffsetRange(tp, fromOffset, untilOffset, preferredLoc)
+ }.toArray
+
+ // Create a RDD that reads from Kafka and get the (key, value) pair as
byte arrays.
+ val rdd = new KafkaSourceRDD(
+ sc, executorKafkaParams, offsetRanges, sourceOptions).map { cr =>
+ Row(cr.checksum, cr.key, cr.offset, cr.partition,
cr.serializedKeySize,
+ cr.serializedValueSize, cr.timestamp, cr.timestampType.id,
cr.topic, cr.value)
+ }
+
+ logInfo("GetBatch: " +
offsetRanges.sortBy(_.topicPartition.toString).mkString(", "))
+ sqlContext.createDataFrame(rdd, schema)
+ }
+
+ /** Stop this source and free any resources it has allocated. */
+ override def stop(): Unit = synchronized {
+ consumer.close()
+ }
+
+ override def toString(): String = s"KafkaSource[$consumerStrategy]"
+
+ private def fetchPartitionOffsets(seekToLatest: Boolean):
Map[TopicPartition, Long] = {
+ synchronized {
+ logTrace("\tPolling")
+ consumer.poll(0)
+ val partitions = consumer.assignment()
+ consumer.pause(partitions)
+ logDebug(s"\tPartitioned assigned to consumer: $partitions")
+ if (seekToLatest) {
+ consumer.seekToEnd(partitions)
+ logDebug("\tSeeked to the end")
+ }
+ logTrace("Getting positions")
+ val partitionToOffsets = partitions.asScala.map(p => p ->
consumer.position(p))
--- End diff --
note to myself: this line will throw NPE when the topic is deleted just
before it.
```
[info] java.lang.NullPointerException:
[info] at java.util.ArrayList.addAll(ArrayList.java:577)
[info] at
org.apache.kafka.clients.Metadata.getClusterForCurrentTopics(Metadata.java:257)
[info] at org.apache.kafka.clients.Metadata.update(Metadata.java:177)
[info] at
org.apache.kafka.clients.NetworkClient$DefaultMetadataUpdater.handleResponse(NetworkClient.java:605)
[info] at
org.apache.kafka.clients.NetworkClient$DefaultMetadataUpdater.maybeHandleCompletedReceive(NetworkClient.java:582)
[info] at
org.apache.kafka.clients.NetworkClient.handleCompletedReceives(NetworkClient.java:450)
[info] at
org.apache.kafka.clients.NetworkClient.poll(NetworkClient.java:269)
[info] at
org.apache.kafka.clients.consumer.internals.ConsumerNetworkClient.clientPoll(ConsumerNetworkClient.java:360)
[info] at
org.apache.kafka.clients.consumer.internals.ConsumerNetworkClient.poll(ConsumerNetworkClient.java:224)
[info] at
org.apache.kafka.clients.consumer.internals.ConsumerNetworkClient.poll(ConsumerNetworkClient.java:192)
[info] at
org.apache.kafka.clients.consumer.internals.ConsumerNetworkClient.poll(ConsumerNetworkClient.java:163)
[info] at
org.apache.kafka.clients.consumer.internals.Fetcher.listOffset(Fetcher.java:315)
[info] at
org.apache.kafka.clients.consumer.internals.Fetcher.resetOffset(Fetcher.java:298)
[info] at
org.apache.kafka.clients.consumer.internals.Fetcher.updateFetchPositions(Fetcher.java:170)
[info] at
org.apache.kafka.clients.consumer.KafkaConsumer.updateFetchPositions(KafkaConsumer.java:1409)
[info] at
org.apache.kafka.clients.consumer.KafkaConsumer.position(KafkaConsumer.java:1197)
```
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
