Github user gaborgsomogyi commented on a diff in the pull request: https://github.com/apache/spark/pull/22138#discussion_r218709909 --- Diff: external/kafka-0-10-sql/src/main/scala/org/apache/spark/sql/kafka010/InternalKafkaConsumerPool.scala --- @@ -0,0 +1,243 @@ +/* + * 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 java.util.concurrent.ConcurrentHashMap + +import org.apache.commons.pool2.{BaseKeyedPooledObjectFactory, PooledObject, SwallowedExceptionListener} +import org.apache.commons.pool2.impl.{DefaultEvictionPolicy, DefaultPooledObject, GenericKeyedObjectPool, GenericKeyedObjectPoolConfig} + +import org.apache.spark.SparkEnv +import org.apache.spark.internal.Logging +import org.apache.spark.sql.kafka010.InternalKafkaConsumerPool._ +import org.apache.spark.sql.kafka010.KafkaDataConsumer.CacheKey + +/** + * Provides object pool for [[InternalKafkaConsumer]] which is grouped by [[CacheKey]]. + * + * This class leverages [[GenericKeyedObjectPool]] internally, hence providing methods based on + * the class, and same contract applies: after using the borrowed object, you must either call + * returnObject() if the object is healthy to return to pool, or invalidateObject() if the object + * should be destroyed. + * + * The soft capacity of pool is determined by "spark.sql.kafkaConsumerCache.capacity" config value, + * and the pool will have reasonable default value if the value is not provided. + * (The instance will do its best effort to respect soft capacity but it can exceed when there's + * a borrowing request and there's neither free space nor idle object to clear.) + * + * This class guarantees that no caller will get pooled object once the object is borrowed and + * not yet returned, hence provide thread-safety usage of non-thread-safe [[InternalKafkaConsumer]] + * unless caller shares the object to multiple threads. + */ +private[kafka010] class InternalKafkaConsumerPool( + objectFactory: ObjectFactory, + poolConfig: PoolConfig) { + + // the class is intended to have only soft capacity + assert(poolConfig.getMaxTotal < 0) + + private lazy val pool = { + val internalPool = new GenericKeyedObjectPool[CacheKey, InternalKafkaConsumer]( + objectFactory, poolConfig) + internalPool.setSwallowedExceptionListener(CustomSwallowedExceptionListener) + internalPool + } + + /** + * Borrows [[InternalKafkaConsumer]] object from the pool. If there's no idle object for the key, + * the pool will create the [[InternalKafkaConsumer]] object. + * + * If the pool doesn't have idle object for the key and also exceeds the soft capacity, + * pool will try to clear some of idle objects. + * + * Borrowed object must be returned by either calling returnObject or invalidateObject, otherwise + * the object will be kept in pool as active object. + */ + def borrowObject(key: CacheKey, kafkaParams: ju.Map[String, Object]): InternalKafkaConsumer = { + updateKafkaParamForKey(key, kafkaParams) + + if (getTotal == poolConfig.getSoftMaxTotal()) { + pool.clearOldest() + } + + pool.borrowObject(key) + } + + /** Returns borrowed object to the pool. */ + def returnObject(consumer: InternalKafkaConsumer): Unit = { + pool.returnObject(extractCacheKey(consumer), consumer) + } + + /** Invalidates (destroy) borrowed object to the pool. */ + def invalidateObject(consumer: InternalKafkaConsumer): Unit = { + pool.invalidateObject(extractCacheKey(consumer), consumer) + } + + /** Invalidates all idle consumers for the key */ + def invalidateKey(key: CacheKey): Unit = { + pool.clear(key) + } + + /** + * Closes the keyed object pool. Once the pool is closed, + * borrowObject will fail with [[IllegalStateException]], but returnObject and invalidateObject + * will continue to work, with returned objects destroyed on return. + * + * Also destroys idle instances in the pool. + */ + def close(): Unit = { + pool.close() + } + + def getNumIdle: Int = pool.getNumIdle + + def getNumIdle(key: CacheKey): Int = pool.getNumIdle(key) + + def getNumActive: Int = pool.getNumActive + + def getNumActive(key: CacheKey): Int = pool.getNumActive(key) + + def getTotal: Int = getNumIdle + getNumActive + + def getTotal(key: CacheKey): Int = getNumIdle(key) + getNumActive(key) + + private def updateKafkaParamForKey(key: CacheKey, kafkaParams: ju.Map[String, Object]): Unit = { + // We can assume that kafkaParam should not be different for same cache key, + // otherwise we can't reuse the cached object and cache key should contain kafkaParam. + // So it should be safe to put the key/value pair only when the key doesn't exist. + val oldKafkaParams = objectFactory.keyToKafkaParams.putIfAbsent(key, kafkaParams) + require(oldKafkaParams == null || kafkaParams == oldKafkaParams, "Kafka parameters for same " + + s"cache key should be equal. old parameters: $oldKafkaParams new parameters: $kafkaParams") + } + + private def extractCacheKey(consumer: InternalKafkaConsumer): CacheKey = { + new CacheKey(consumer.topicPartition, consumer.kafkaParams) + } +} + +private[kafka010] object InternalKafkaConsumerPool { + + /** + * Builds the pool for [[InternalKafkaConsumer]]. The pool instance is created per each call. + */ + def build: InternalKafkaConsumerPool = { + val objFactory = new ObjectFactory + val poolConfig = new PoolConfig + new InternalKafkaConsumerPool(objFactory, poolConfig) + } + + case class PooledObjectInvalidated(key: CacheKey, lastInvalidatedTimestamp: Long, --- End diff -- I don't see where it's used.
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org