viirya commented on code in PR #55420: URL: https://github.com/apache/spark/pull/55420#discussion_r3149298272
########## connector/kafka-0-10-sql/src/test/scala/org/apache/spark/sql/kafka010/RTMKafkaKafkaBenchmarkSuite.scala: ########## @@ -0,0 +1,296 @@ +/* + * 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.nio.file.Files +import java.util.{Properties, Timer, TimerTask} +import java.util.concurrent.{CountDownLatch, TimeUnit} +import java.util.concurrent.atomic.AtomicLong + +import scala.concurrent.duration._ + +import org.apache.kafka.clients.producer.{Callback, KafkaProducer, Producer, ProducerRecord, RecordMetadata} +import org.scalatest.BeforeAndAfterEach +import org.scalatest.matchers.should.Matchers + +import org.apache.spark.{SparkContext, ThreadAudit} +import org.apache.spark.sql.Column +import org.apache.spark.sql.execution.streaming.RealTimeTrigger +import org.apache.spark.sql.functions._ +import org.apache.spark.sql.internal.SQLConf +import org.apache.spark.sql.streaming.StreamingQueryListener +import org.apache.spark.sql.test.TestSparkSession + +/** + * Stateless Kafka-to-Kafka RTM benchmark. Reads from an input Kafka topic, applies a + * stateless transformation, and writes results to an output Kafka topic using + * [[RealTimeTrigger]]. After the run it reports e2e latency percentiles. + * + * This benchmark intentionally runs a real local-cluster and a live Kafka broker, so it + * is slow. Run it explicitly when measuring RTM throughput and latency for the stateless path. + */ +class RTMKafkaKafkaBenchmarkSuite + extends KafkaSourceTest + with ThreadAudit + with BeforeAndAfterEach + with Matchers { + + override protected def createSparkSession = new TestSparkSession( + new SparkContext( + "local-cluster[3, 5, 1024]", + "microbatch-context", + sparkConf + )) + + test("RTM stateless kafka-to-kafka benchmark") { + benchmark(15.seconds.toMillis, 4) + } + + def benchmark(longRunningBatchDurationMs: Long, numBatches: Long): Unit = { + val inputTopic = newTopic() + testUtils.createTopic(inputTopic, partitions = 5) + + val outputTopic = newTopic() + testUtils.createTopic(outputTopic, partitions = 5) + + spark.conf.set(SQLConf.STREAMING_POLLING_DELAY.key, 10) + + val kafkaStream = spark.readStream + .format("kafka") + .option("kafka.bootstrap.servers", testUtils.brokerAddress) + .option("subscribe", inputTopic) + .option("kafka.fetch.max.wait.ms", "10") + .option("kafka.max.partition.fetch.bytes", "10485760") // 10MB + .load() + + val currentTimestampUDF = udf(() => System.currentTimeMillis()) + + val streamWithObserved = kafkaStream + .withColumn("value", base64(col("value"))) + .withColumn( + "headers", + array( + struct( + lit("source-timestamp") as "key", + unix_millis(col("timestamp")).cast("STRING").cast("BINARY") as "value"))) + .withColumn("temp-timestamp", currentTimestampUDF()) + .withColumn( + "latency", + col("temp-timestamp").cast("long") - unix_millis(col("timestamp")).cast("long")) + .observe( + name = "observedLatency", + avg(col("latency")).as("avg"), + max(col("latency")).as("max"), + percentile_approx(col("latency"), lit(0.99), lit(10000)).as("p99"), + percentile_approx(col("latency"), lit(0.5), lit(10000)).as("p50")) + .drop(col("latency")) + .drop(col("temp-timestamp")) + .drop(col("timestamp")) + + val query = streamWithObserved.writeStream + .format("kafka") + .option("kafka.bootstrap.servers", testUtils.brokerAddress) + .option("topic", outputTopic) + .option("checkpointLocation", Files.createTempDirectory("some-prefix").toFile.getName) + .option("kafka.buffer.memory", "67108864") // 64MB + .option("kafka.compression.type", "snappy") + .outputMode("update") + .queryName("rtm-kafka-kafka") + .trigger(RealTimeTrigger.apply(s"${longRunningBatchDurationMs} milliseconds")) + .start() + + val dataGenThread = new Thread(() => { + genData(testUtils.brokerAddress, inputTopic, 1000) + }) + dataGenThread.start() + + val latch = new CountDownLatch(1) + + spark.streams.addListener(new StreamingQueryListener { + override def onQueryStarted( + event: StreamingQueryListener.QueryStartedEvent): Unit = {} + + override def onQueryTerminated( + event: StreamingQueryListener.QueryTerminatedEvent): Unit = {} + + override def onQueryProgress(event: StreamingQueryListener.QueryProgressEvent): Unit = { + if (event.progress.batchId == numBatches - 1) { + latch.countDown() + } + } + }) + + val timeoutMs = numBatches * longRunningBatchDurationMs * 2 + 60 * 1000 + val completed = latch.await(timeoutMs, TimeUnit.MILLISECONDS) + query.stop() + dataGenThread.interrupt() + if (!completed) { + throw new RuntimeException( + s"Benchmark timed out waiting for $numBatches batches to complete after ${timeoutMs}ms.") + } + + getLatencies(longRunningBatchDurationMs, numBatches, outputTopic) + } + + private def genData(url: String, topicName: String, throughput: Long): Unit = { Review Comment: genData now cancels the timer and closes the producer in finally, which is good, but the caller only invokes dataGenThread.interrupt() and then continues without waiting for that cleanup to finish. Since this suite mixes in ThreadAudit, the test may finish while the generator thread is still unwinding, sleeping, or blocked in producer.close(), which can lead to flaky thread-leak failures. Please join the thread with a bounded timeout after interrupting it, and ideally put query shutdown plus generator stop/join in an outer try/finally so cleanup also happens if await is interrupted or another exception is thrown. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
