clintropolis commented on code in PR #12277: URL: https://github.com/apache/druid/pull/12277#discussion_r1004319810
########## processing/src/main/java/org/apache/druid/segment/data/FrontCodedIndexed.java: ########## @@ -0,0 +1,491 @@ +/* + * 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.druid.segment.data; + +import com.google.common.base.Preconditions; +import com.google.common.base.Supplier; +import org.apache.druid.common.config.NullHandling; +import org.apache.druid.java.util.common.ISE; +import org.apache.druid.query.monomorphicprocessing.RuntimeShapeInspector; + +import javax.annotation.Nullable; +import java.nio.ByteBuffer; +import java.nio.ByteOrder; +import java.util.Iterator; +import java.util.NoSuchElementException; + +/** + * {@link Indexed} specialized for storing variable-width binary values (such as utf8 encoded strings), which must be + * sorted and unique, using 'front coding'. Front coding is a type of delta encoding for byte arrays, where sorted + * values are grouped into buckets. The first value of the bucket is written entirely, and remaining values are stored + * as a pair of an integer which indicates how much of the first byte array of the bucket to use as a prefix, followed + * by the remaining bytes after the prefix to complete the value. + * + * Getting a value first picks the appropriate bucket, finds its offset in the underlying buffer, then scans the bucket + * values to seek to the correct position of the value within the bucket in order to reconstruct it using the prefix + * length. + * + * Finding the index of a value involves binary searching the first values of each bucket to find the correct bucket, + * then a linear scan within the bucket to find the matching value (or negative insertion point -1 for values that + * are not present). + * + * The value iterator reads an entire bucket at a time, reconstructing the values into an array to iterate within the + * bucket before moving onto the next bucket as the iterator is consumed. + */ +public final class FrontCodedIndexed implements Indexed<ByteBuffer> +{ + public static Supplier<FrontCodedIndexed> read(ByteBuffer buffer, ByteOrder ordering) + { + final ByteBuffer orderedBuffer = buffer.asReadOnlyBuffer().order(ordering); + final byte version = orderedBuffer.get(); + Preconditions.checkArgument(version == 0, "only V0 exists, encountered " + version); + final int bucketSize = orderedBuffer.get(); + final boolean hasNull = NullHandling.IS_NULL_BYTE == orderedBuffer.get(); + final int numValues = VByte.readInt(orderedBuffer); + // size of offsets + values + final int size = VByte.readInt(orderedBuffer); + final int offsetsPosition = orderedBuffer.position(); + // move position to end of buffer + buffer.position(offsetsPosition + size); + + final int numBuckets = (int) Math.ceil((double) numValues / (double) bucketSize); + final int adjustIndex = hasNull ? 1 : 0; + final int div = Integer.numberOfTrailingZeros(bucketSize); + final int rem = bucketSize - 1; + return () -> new FrontCodedIndexed( + orderedBuffer, + bucketSize, + numBuckets, + (numValues & rem) == 0 ? bucketSize : numValues & rem, + hasNull, + numValues + adjustIndex, + adjustIndex, + div, + rem, + offsetsPosition, + offsetsPosition + ((numBuckets - 1) * Integer.BYTES) + ); + } + + private final ByteBuffer buffer; + private final int adjustedNumValues; + private final int adjustIndex; + private final int bucketSize; + private final int numBuckets; + private final int div; + private final int rem; + private final int offsetsPosition; + private final int bucketsPosition; + private final boolean hasNull; + private final int lastBucketNumValues; + + private FrontCodedIndexed( + ByteBuffer buffer, + int bucketSize, + int numBuckets, + int lastBucketNumValues, + boolean hasNull, + int adjustedNumValues, + int adjustIndex, + int div, + int rem, + int offsetsPosition, + int bucketsPosition + ) + { + if (Integer.bitCount(bucketSize) != 1) { + throw new ISE("bucketSize must be a power of two but was[%,d]", bucketSize); + } + this.buffer = buffer.asReadOnlyBuffer().order(buffer.order()); + this.bucketSize = bucketSize; + this.hasNull = hasNull; + this.adjustedNumValues = adjustedNumValues; + this.adjustIndex = adjustIndex; + this.div = div; + this.rem = rem; + this.numBuckets = numBuckets; + this.offsetsPosition = offsetsPosition; + this.bucketsPosition = bucketsPosition; + this.lastBucketNumValues = lastBucketNumValues; + } + + @Override + public int size() + { + return adjustedNumValues; + } + + @Nullable + @Override + public ByteBuffer get(int index) + { + if (hasNull && index == 0) { + return null; + } + + // due to vbyte encoding, the null value is not actually stored in the bucket (no negative values), so we adjust + // the index + final int adjustedIndex = index - adjustIndex; + // find the bucket which contains the value with maths + final int bucket = adjustedIndex >> div; + final int bucketIndex = adjustedIndex & rem; + final int offset = getBucketOffset(bucket); + buffer.position(offset); + return getFromBucket(buffer, bucketIndex); + } + + @Override + public int indexOf(@Nullable ByteBuffer value) Review Comment: I ended up adding `isSorted` which when true strengthens the contract of `indexOf` to be `(-(insertion point) - 1)`. `GenericIndexed` (and `BufferedIndexed` produced by its `singleThreaded`), `FixedIndexed`, and `FrontCodedIndexed` all implement it. -- 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: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
