siddharthteotia commented on code in PR #12223: URL: https://github.com/apache/pinot/pull/12223#discussion_r1462920239
########## pinot-spi/src/main/java/org/apache/pinot/spi/utils/FALFInterner.java: ########## @@ -0,0 +1,148 @@ +/** + * 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.pinot.spi.utils; + + +import com.google.common.collect.Interner; +import java.util.Objects; +import java.util.function.BiPredicate; +import java.util.function.ToIntFunction; + +/** + * Fixed-size Array-based, Lock-Free Interner. + * + * !!!!!!!!!!!!!!! READ THE PARAGRAPH BELOW BEFORE USING THIS CLASS !!!!!!!!!!!!!!!! + * This class is technically not thread-safe. Therefore if it's called from multiple + * threads, it should either be used with proper synchronization (in the same way as + * you would use e.g. a HashMap), or under the following conditions: + * all the objects being interned are not just immutable, but also final (that is, all + * their fields used in equals() and hashCode() methods are explicitly marked final). + * That's to ensure that all threads always see the same contents of these objects. If + * this rule is not followed, using this class from multiple threads may lead to strange + * non-deterministic errors. Note that objects with all private fields that are not + * marked final, or immutable collections created via Collection.unmodifiableMap() etc, + * don't qualify. + * !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! + * + * This interner is intended to be used when either: + * (a) distribution of values among the objects to be interned make them not suitable + * for standard interners + * (b) speed is more important than ultimate memory savings + * + * Problem (a) occurs when both the total number of objects AND the number of unique + * values is large. For example, there are 1M strings that look like "a", "a", "b", "b", + * "c", "c", ... - that is, for each unique value there are only two separate objects. + * Another problematic case is when "a" has 1000 copies, "b" has 900 copies, etc., + * but in the last few hundred thousand objects each one is unique. In both cases, if + * we use a standard interner such as a Guava interner or a ConcurrentHashMap to + * deduplicate such objects, the amount of memory consumed by the interner itself to + * store objects that have few or no duplicates, can be comparable, or even exceed, the + * savings achieved by getting rid of duplicate objects. + * + * This implementation addresses the above problems by interning objects "optimistically". + * It is a fixed-size, open-hashmap-based object cache. When there is a cache miss, + * a cached object in the given slot is always replaced with a new object. There is + * no locking and no synchronization, and thus, no associated overhead. In essence, + * this cache is based on the idea that an object with value X, that has many copies, + * has a higher chance of staying in the cache for long enough to guarantee several + * cache hits for itself before a miss evicts it and replaces it with an object with + * a different value Y. + * + * This interner has a minimum possible memory footprint. You should be careful when Review Comment: So what was the capacity we used in tests to validate the production use case memory footprint reduction from 13Gb to 1GB ? Based on the gain, looks like we were able to optimize away all the duplicates effectively ? -- 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: commits-unsubscr...@pinot.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@pinot.apache.org For additional commands, e-mail: commits-h...@pinot.apache.org
