mccheah commented on a change in pull request #28618: URL: https://github.com/apache/spark/pull/28618#discussion_r488085526
########## File path: core/src/main/java/org/apache/spark/shuffle/api/metadata/ShuffleOutputTracker.java ########## @@ -0,0 +1,56 @@ +/* + * 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.shuffle.api.metadata; + +/** + * :: Private :: + * + * A plugin that can monitor the storage of shuffle data from map tasks, and can provide + * metadata to shuffle readers to aid their reading of shuffle blocks in reduce tasks. + * <p> + * {@link MapOutputMetadata} instances provided from the plugin tree's implementation of + * {@link org.apache.spark.shuffle.api.ShuffleMapOutputWriter} are sent to + * <p> + * Implementations MUST be thread-safe. Spark will invoke methods in this module in parallel. Review comment: The whole point of just saying that implementations must be thread-safe is that implementations cannot make _any_ guarantees about how the caller will invoke this API. Even if we can say that the current implementation of `MapOutputTracker` would guarantee locking on per-shuffle-id, for example, that isn't an API guarantee we can realistically make since `MapOutputTracker` is a module internal to Spark. I think saying that there is no guarantee about concurrent access to all methods suffices. But then that basically means "the implementation should be thread-safe". Is that reasonable, or do we need to be more granular than that? ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
