Github user tdas commented on a diff in the pull request:
https://github.com/apache/spark/pull/3798#discussion_r22436219
--- Diff:
external/kafka/src/main/scala/org/apache/spark/streaming/kafka/DeterministicKafkaInputDStream.scala
---
@@ -0,0 +1,123 @@
+/*
+ * 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 scala.annotation.tailrec
+import scala.reflect.{classTag, ClassTag}
+
+import kafka.common.TopicAndPartition
+import kafka.message.MessageAndMetadata
+import kafka.serializer.Decoder
+
+import org.apache.spark.Logging
+import org.apache.spark.rdd.RDD
+import org.apache.spark.rdd.kafka.{KafkaCluster, KafkaRDD}
+import org.apache.spark.streaming.{StreamingContext, Time}
+import org.apache.spark.streaming.dstream._
+
+/** A stream of {@link org.apache.spark.rdd.kafka.KafkaRDD} where
+ * each given Kafka topic/partition corresponds to an RDD partition.
+ * The spark configuration spark.streaming.receiver.maxRate gives the
maximum number of messages
+ * per second that each '''partition''' will accept.
+ * Starting offsets are specified in advance,
+ * and this DStream is not responsible for committing offsets,
+ * so that you can control exactly-once semantics.
+ * For an easy interface to Kafka-managed offsets,
+ * see {@link org.apache.spark.rdd.kafka.KafkaCluster}
+ * @param kafkaParams Kafka <a
href="http://kafka.apache.org/documentation.html#configuration";>
+ * configuration parameters</a>.
+ * Requires "metadata.broker.list" or "bootstrap.servers" to be set
with Kafka broker(s),
+ * NOT zookeeper servers, specified in host1:port1,host2:port2 form.
+ * @param fromOffsets per-topic/partition Kafka offsets defining the
(inclusive)
+ * starting point of the stream
+ * @param messageHandler function for translating each message into the
desired type
+ * @param maxRetries maximum number of times in a row to retry getting
leaders' offsets
+ */
+class DeterministicKafkaInputDStream[
+ K: ClassTag,
+ V: ClassTag,
+ U <: Decoder[_]: ClassTag,
+ T <: Decoder[_]: ClassTag,
+ R: ClassTag](
+ @transient ssc_ : StreamingContext,
+ val kafkaParams: Map[String, String],
+ val fromOffsets: Map[TopicAndPartition, Long],
+ messageHandler: MessageAndMetadata[K, V] => R,
+ maxRetries: Int = 1
+) extends InputDStream[R](ssc_) with Logging {
+
+ private val kc = new KafkaCluster(kafkaParams)
+
+ private val maxMessagesPerPartition: Option[Long] = {
+ val ratePerSec =
ssc.sparkContext.getConf.getInt("spark.streaming.receiver.maxRate", 0)
--- End diff --
Yeah, I get it. However, this still has a different performance
characteristics than receiver based sources. Sometimes it is better (as you are
"always" pulling data in parallel across cluster, instead of the default 1
receiver), sometimes it is worse (window operations that require past data
which needs to be pulled from Kafka every time). For this method to be viable
alternative to the existing, we ideally have to make sure that the performance
characteristics of this method is >= performance of the existing method under
all situations. Then using this method will be justified even if the behavior
is different.
On that note, here is an idea of what we can do. We can store the data
pulled from the Kafka as blocks in the BlockManager, so that subsequent
accesses to the data (due to window or stateful ops) can be faster. One way to
do this is to implement a `KafkaBackedBlockRDD`. This is similar to
`WriteAheadLogBackedBlockRDD` which has the logic to either read from
BlockManager if block is present, or reload the data from the WriteAheadLog
based on file segment info. `KafkaBackedBlockRDD` can be similar, either read
from BlockManager, or load it from Kafka based on offsets.
---
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]
