[ https://issues.apache.org/jira/browse/FLINK-7213?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16126149#comment-16126149 ]
ASF GitHub Bot commented on FLINK-7213: --------------------------------------- Github user StephanEwen commented on a diff in the pull request: https://github.com/apache/flink/pull/4353#discussion_r133009796 --- Diff: flink-runtime/src/main/java/org/apache/flink/runtime/checkpoint/TaskStateSnapshot.java --- @@ -0,0 +1,139 @@ +/* + * 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.flink.runtime.checkpoint; + +import org.apache.flink.runtime.jobgraph.OperatorID; +import org.apache.flink.runtime.state.CompositeStateHandle; +import org.apache.flink.runtime.state.SharedStateRegistry; +import org.apache.flink.runtime.state.StateUtil; +import org.apache.flink.util.Preconditions; + +import java.util.HashMap; +import java.util.Map; +import java.util.Set; + +/** + * This class encapsulates state handles to the snapshots of all operator instances executed within one task. A task + * can run multiple operator instances as a result of operator chaining, and all operator instances from the chain can + * register their state under their operator id. Each operator instance is a physical execution responsible for + * processing a partition of the data that goes through a logical operator. This partitioning happens to parallelize + * execution of logical operators, e.g. distributing a map function. + * + * <p>One instance of this class contains the information that one task will send to acknowledge a checkpoint request by + * the checkpoint coordinator. Tasks run operator instances in parallel, so the union of all + * {@link TaskStateSnapshot} that are collected by the checkpoint coordinator from all tasks represent the whole + * state of a job at the time of the checkpoint. + * + * <p>This class should be called TaskState once the old class with this name that we keep for backwards + * compatibility goes away. + */ +public class TaskStateSnapshot implements CompositeStateHandle { + + private static final long serialVersionUID = 1L; + + /** Mapping from an operator id to the state of one subtask of this operator */ + private final Map<OperatorID, OperatorSubtaskState> subtaskStatesByOperatorID; --- End diff -- A `LinkedHashMap` has a slightly more predictable iteration performance (list traversal) compared to a `HashMap` (search through sparse table array). There are a lot of value iterations done in this class, but we also should have pretty full hash tables (since we never delete), so not sure how much difference it makes... > Introduce state management by OperatorID in TaskManager > ------------------------------------------------------- > > Key: FLINK-7213 > URL: https://issues.apache.org/jira/browse/FLINK-7213 > Project: Flink > Issue Type: Improvement > Components: State Backends, Checkpointing > Affects Versions: 1.4.0 > Reporter: Stefan Richter > Assignee: Stefan Richter > > Flink-5892 introduced the job manager / checkpoint coordinator part of > managing state on the operator level instead of the task level by introducing > explicit operator_id -> state mappings. However, this explicit mapping was > not introduced in the task manager side, so the explicit mapping is still > converted into a mapping that suits the implicit operator chain order. > We should also introduce explicit operator ids to state management on the > task manager. -- This message was sent by Atlassian JIRA (v6.4.14#64029)