Github user markgrover commented on a diff in the pull request:
https://github.com/apache/spark/pull/11863#discussion_r59956716
--- Diff:
external/kafka-beta/src/main/scala/org/apache/spark/streaming/kafka/KafkaRDD.scala
---
@@ -0,0 +1,259 @@
+/*
+ * 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.streaming.kafka
+
+import java.{ util => ju }
+
+import scala.collection.mutable.ArrayBuffer
+import scala.reflect.{classTag, ClassTag}
+
+import org.apache.kafka.clients.consumer.{ ConsumerConfig, ConsumerRecord }
+import org.apache.kafka.common.TopicPartition
+
+import org.apache.spark.{Partition, SparkContext, SparkException,
TaskContext}
+import org.apache.spark.internal.Logging
+import org.apache.spark.partial.{BoundedDouble, PartialResult}
+import org.apache.spark.rdd.RDD
+import org.apache.spark.scheduler.ExecutorCacheTaskLocation
+import org.apache.spark.storage.StorageLevel
+
+/**
+ * A batch-oriented interface for consuming from Kafka.
+ * Starting and ending offsets are specified in advance,
+ * so that you can control exactly-once semantics.
+ * @param kafkaParams Kafka
+ * <a
href="http://kafka.apache.org/documentation.htmll#newconsumerconfigs";>
+ * configuration parameters</a>. Requires "bootstrap.servers" to be set
+ * with Kafka broker(s) specified in host1:port1,host2:port2 form.
+ * @param offsetRanges offset ranges that define the Kafka data belonging
to this RDD
+ */
+
+class KafkaRDD[
+ K: ClassTag,
+ V: ClassTag] private[spark] (
+ sc: SparkContext,
+ val kafkaParams: ju.Map[String, Object],
+ val offsetRanges: Array[OffsetRange],
+ val preferredHosts: ju.Map[TopicPartition, String]
+) extends RDD[ConsumerRecord[K, V]](sc, Nil) with Logging with
HasOffsetRanges {
+
+ assert("none" ==
+
kafkaParams.get(ConsumerConfig.AUTO_OFFSET_RESET_CONFIG).asInstanceOf[String],
+ ConsumerConfig.AUTO_OFFSET_RESET_CONFIG +
+ " must be set to none for executor kafka params, else messages may
not match offsetRange")
+
+ assert(false ==
+
kafkaParams.get(ConsumerConfig.ENABLE_AUTO_COMMIT_CONFIG).asInstanceOf[Boolean],
+ ConsumerConfig.AUTO_OFFSET_RESET_CONFIG +
+ " must be set to false for executor kafka params, else offsets may
commit before processing")
+
+ // TODO is it necessary to have separate configs for initial poll time
vs ongoing poll time?
+ private val pollTimeout =
conf.getLong("spark.streaming.kafka.consumer.poll.ms", 256)
+ private val cacheInitialCapacity =
+ conf.getInt("spark.streaming.kafka.consumer.cache.initialCapacity", 16)
+ private val cacheMaxCapacity =
+ conf.getInt("spark.streaming.kafka.consumer.cache.maxCapacity", 64)
+ private val cacheLoadFactor =
+ conf.getDouble("spark.streaming.kafka.consumer.cache.loadFactor",
0.75).toFloat
+
+ override def persist(newLevel: StorageLevel): this.type = {
+ log.error("Kafka ConsumerRecord is not serializable. " +
+ "Use .map to extract fields before calling .persist or .window")
+ super.persist(newLevel)
+ }
+
+ override def getPartitions: Array[Partition] = {
+ offsetRanges.zipWithIndex.map { case (o, i) =>
+ new KafkaRDDPartition(i, o.topic, o.partition, o.fromOffset,
o.untilOffset)
+ }.toArray
+ }
+
+ override def count(): Long = offsetRanges.map(_.count).sum
+
+ override def countApprox(
+ timeout: Long,
+ confidence: Double = 0.95
+ ): PartialResult[BoundedDouble] = {
+ val c = count
+ new PartialResult(new BoundedDouble(c, 1.0, c, c), true)
+ }
+
+ override def isEmpty(): Boolean = count == 0L
+
+ override def take(num: Int): Array[ConsumerRecord[K, V]] = {
+ val nonEmptyPartitions = this.partitions
+ .map(_.asInstanceOf[KafkaRDDPartition])
+ .filter(_.count > 0)
+
+ if (num < 1 || nonEmptyPartitions.size < 1) {
+ return new Array[ConsumerRecord[K, V]](0)
+ }
+
+ // Determine in advance how many messages need to be taken from each
partition
+ val parts = nonEmptyPartitions.foldLeft(Map[Int, Int]()) { (result,
part) =>
+ val remain = num - result.values.sum
+ if (remain > 0) {
+ val taken = Math.min(remain, part.count)
+ result + (part.index -> taken.toInt)
+ } else {
+ result
+ }
+ }
+
+ val buf = new ArrayBuffer[ConsumerRecord[K, V]]
+ val res = context.runJob(
+ this,
+ (tc: TaskContext, it: Iterator[ConsumerRecord[K, V]]) =>
+ it.take(parts(tc.partitionId)).toArray, parts.keys.toArray
+ )
+ res.foreach(buf ++= _)
+ buf.toArray
+ }
+
+ private def executors(): Array[ExecutorCacheTaskLocation] = {
+ val bm = sparkContext.env.blockManager
+ bm.master.getPeers(bm.blockManagerId).toArray
+ .map(x => ExecutorCacheTaskLocation(x.host, x.executorId))
+ .sortWith((a, b) => a.host > b.host || a.executorId > b.executorId)
+ }
+
+ // non-negative modulus, from java 8 math
+ private def floorMod(a: Int, b: Int): Int = ((a % b) + b) % b
+
+ override def getPreferredLocations(thePart: Partition): Seq[String] = {
+ // TODO what about hosts specified by ip vs name
+ val part = thePart.asInstanceOf[KafkaRDDPartition]
+ val allExecs = executors()
+ val tp = part.topicPartition
+ val prefHost = preferredHosts.get(tp)
+ val prefExecs = if (null == prefHost) allExecs else
allExecs.filter(_.host == prefHost)
+ val execs = if (prefExecs.isEmpty) allExecs else prefExecs
+ if (execs.isEmpty) {
+ Seq()
+ } else {
+ val index = this.floorMod(tp.hashCode, execs.length)
--- End diff --
Ok, so this may very well be too late on a Friday but could you help me
understand this part please?
Why set a preferred host when it seems like there isn't any?
---
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 [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
