mumrah commented on code in PR #16969: URL: https://github.com/apache/kafka/pull/16969#discussion_r1729433979
########## core/src/main/java/kafka/server/share/DelayedShareFetchKey.java: ########## @@ -0,0 +1,65 @@ +/* + * 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 kafka.server.share; + +import kafka.server.DelayedOperationKey; + +import org.apache.kafka.common.TopicIdPartition; + +import java.util.Objects; +import java.util.Set; + +/** + * A key for delayed operations that fetch data for share consumers. + */ +public class DelayedShareFetchKey implements DelayedOperationKey { + private final Set<TopicIdPartition> topicIdPartitions; Review Comment: Each entry in the new purgatory should correspond to a share fetch, but the keys should be more granular. This is really the power of request purgatories. The keys represent dimensions of the entry which can cause it to be completed. For fetch requests, any partition in the request can cause the request to be completed via MinBytes. So, we create a DelayedFetchRequest with keys for each partition. https://github.com/apache/kafka/blob/trunk/core/src/main/scala/kafka/server/ReplicaManager.scala#L1587-L1601 For share fetch, we'll do something similar. For example: * SFR1 arrives (SP1, SP2, SP3), no locks available so we create DSF1 * Purgatory now contains one entry DSF1 with three keys SPK1, SPK2, SPK3 -- one key per share partition * SFR2 arrives with acks for SP1 from another member, we need to call tryComplete for SPK1 * The completion of SPK1 may complete SFR1 if MinBytes is hit ########## core/src/main/java/kafka/server/share/DelayedShareFetch.java: ########## @@ -0,0 +1,271 @@ +/* + * 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 kafka.server.share; + +import kafka.server.DelayedOperation; +import kafka.server.QuotaFactory; +import kafka.server.ReplicaManager; + +import org.apache.kafka.common.TopicIdPartition; +import org.apache.kafka.common.message.ShareFetchResponseData; +import org.apache.kafka.common.protocol.Errors; +import org.apache.kafka.common.record.FileRecords; +import org.apache.kafka.common.requests.FetchRequest; +import org.apache.kafka.common.requests.ListOffsetsRequest; +import org.apache.kafka.storage.internals.log.FetchPartitionData; + +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +import java.util.Collections; +import java.util.HashMap; +import java.util.LinkedHashMap; +import java.util.List; +import java.util.Map; +import java.util.Optional; +import java.util.Set; +import java.util.concurrent.CompletableFuture; +import java.util.stream.Collectors; + +import scala.Option; +import scala.Tuple2; +import scala.jdk.javaapi.CollectionConverters; +import scala.runtime.BoxedUnit; + +/** + * A delayed share fetch operation has been introduced in case there is no share partition for which we can acquire records. We will try to wait + * for MaxWaitMs for records to be released else complete the share fetch request. + */ +public class DelayedShareFetch extends DelayedOperation { + private final SharePartitionManager.ShareFetchPartitionData shareFetchPartitionData; + private final ReplicaManager replicaManager; + private final Map<SharePartitionManager.SharePartitionKey, SharePartition> partitionCacheMap; + + private static final Logger log = LoggerFactory.getLogger(DelayedShareFetch.class); + + DelayedShareFetch( + SharePartitionManager.ShareFetchPartitionData shareFetchPartitionData, + ReplicaManager replicaManager, + Map<SharePartitionManager.SharePartitionKey, SharePartition> partitionCacheMap) { + super(shareFetchPartitionData.fetchParams().maxWaitMs, Option.empty()); + this.shareFetchPartitionData = shareFetchPartitionData; + this.replicaManager = replicaManager; + this.partitionCacheMap = partitionCacheMap; + } + + @Override + public void onExpiration() { + } + + /** + * Complete the share fetch operation by fetching records for all partitions in the share fetch request irrespective + * of whether they have any acquired records. This is called when the fetch operation is forced to complete either + * because records can be acquired for some partitions or due to MaxWaitMs timeout. + */ + @Override + public void onComplete() { + log.trace("onCompletion of delayed share fetch request for group {}, member {}, " + + "topic partitions {}", shareFetchPartitionData.groupId(), + shareFetchPartitionData.memberId(), shareFetchPartitionData.partitionMaxBytes().keySet()); + + Map<TopicIdPartition, FetchRequest.PartitionData> topicPartitionData = topicPartitionDataForAcquirablePartitions(); + try { + if (topicPartitionData.isEmpty()) { + // No locks for share partitions could be acquired, so we complete the request with an empty response. + shareFetchPartitionData.future().complete(Collections.emptyMap()); + return; + } + log.trace("Fetchable share partitions data: {} with groupId: {} fetch params: {}", + topicPartitionData, shareFetchPartitionData.groupId(), shareFetchPartitionData.fetchParams()); + + replicaManager.fetchMessages( Review Comment: When `onComplete` is called, we _must_ complete the purgatory entry. What happens if ReplicaManager#fetchMessages does not return synchronously here? I guess we would end up creating a fetch purgatory entry that is competed later on. If we are completing the purgatory entry due to a timeout, we shouldn't allow fetchMessages to create a new fetch purgatory entry as that would delay the response beyond the deadline. We should probably check the time remaining from MaxWaitMs when we call fetchMessages in `onComplete ` -- 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: jira-unsubscr...@kafka.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
