|
| 1 | +/* |
| 2 | + * AbstractStorageAdapter.java |
| 3 | + * |
| 4 | + * This source file is part of the FoundationDB open source project |
| 5 | + * |
| 6 | + * Copyright 2015-2025 Apple Inc. and the FoundationDB project authors |
| 7 | + * |
| 8 | + * Licensed under the Apache License, Version 2.0 (the "License"); |
| 9 | + * you may not use this file except in compliance with the License. |
| 10 | + * You may obtain a copy of the License at |
| 11 | + * |
| 12 | + * http://www.apache.org/licenses/LICENSE-2.0 |
| 13 | + * |
| 14 | + * Unless required by applicable law or agreed to in writing, software |
| 15 | + * distributed under the License is distributed on an "AS IS" BASIS, |
| 16 | + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 17 | + * See the License for the specific language governing permissions and |
| 18 | + * limitations under the License. |
| 19 | + */ |
| 20 | + |
| 21 | +package com.apple.foundationdb.async.hnsw; |
| 22 | + |
| 23 | +import com.apple.foundationdb.ReadTransaction; |
| 24 | +import com.apple.foundationdb.Transaction; |
| 25 | +import com.apple.foundationdb.linear.AffineOperator; |
| 26 | +import com.apple.foundationdb.linear.Quantizer; |
| 27 | +import com.apple.foundationdb.subspace.Subspace; |
| 28 | +import com.apple.foundationdb.tuple.Tuple; |
| 29 | +import org.slf4j.Logger; |
| 30 | +import org.slf4j.LoggerFactory; |
| 31 | + |
| 32 | +import javax.annotation.Nonnull; |
| 33 | +import javax.annotation.Nullable; |
| 34 | +import java.util.concurrent.CompletableFuture; |
| 35 | + |
| 36 | +/** |
| 37 | + * An abstract base class for {@link StorageAdapter} implementations. |
| 38 | + * <p> |
| 39 | + * This class provides the common infrastructure for managing HNSW graph data within a {@link Subspace}. |
| 40 | + * It handles the configuration, node creation, and listener management, while delegating the actual |
| 41 | + * storage-specific read and write operations to concrete subclasses through the {@code fetchNodeInternal} |
| 42 | + * and {@code writeNodeInternal} abstract methods. |
| 43 | + * |
| 44 | + * @param <N> the type of {@link NodeReference} used to reference nodes in the graph |
| 45 | + */ |
| 46 | +abstract class AbstractStorageAdapter<N extends NodeReference> implements StorageAdapter<N> { |
| 47 | + @Nonnull |
| 48 | + private static final Logger logger = LoggerFactory.getLogger(AbstractStorageAdapter.class); |
| 49 | + |
| 50 | + @Nonnull |
| 51 | + private final Config config; |
| 52 | + @Nonnull |
| 53 | + private final NodeFactory<N> nodeFactory; |
| 54 | + @Nonnull |
| 55 | + private final Subspace subspace; |
| 56 | + @Nonnull |
| 57 | + private final OnWriteListener onWriteListener; |
| 58 | + @Nonnull |
| 59 | + private final OnReadListener onReadListener; |
| 60 | + |
| 61 | + @Nonnull |
| 62 | + private final Subspace dataSubspace; |
| 63 | + |
| 64 | + /** |
| 65 | + * Constructs a new {@code AbstractStorageAdapter}. |
| 66 | + * <p> |
| 67 | + * This constructor initializes the adapter with the necessary configuration, |
| 68 | + * factories, and listeners for managing an HNSW graph. It also sets up a |
| 69 | + * dedicated data subspace within the provided main subspace for storing node data. |
| 70 | + * |
| 71 | + * @param config the HNSW graph configuration |
| 72 | + * @param nodeFactory the factory to create new nodes of type {@code <N>} |
| 73 | + * @param subspace the primary subspace for storing all graph-related data |
| 74 | + * @param onWriteListener the listener to be called on write operations |
| 75 | + * @param onReadListener the listener to be called on read operations |
| 76 | + */ |
| 77 | + protected AbstractStorageAdapter(@Nonnull final Config config, @Nonnull final NodeFactory<N> nodeFactory, |
| 78 | + @Nonnull final Subspace subspace, |
| 79 | + @Nonnull final OnWriteListener onWriteListener, |
| 80 | + @Nonnull final OnReadListener onReadListener) { |
| 81 | + this.config = config; |
| 82 | + this.nodeFactory = nodeFactory; |
| 83 | + this.subspace = subspace; |
| 84 | + this.onWriteListener = onWriteListener; |
| 85 | + this.onReadListener = onReadListener; |
| 86 | + this.dataSubspace = subspace.subspace(Tuple.from(SUBSPACE_PREFIX_DATA)); |
| 87 | + } |
| 88 | + |
| 89 | + @Override |
| 90 | + @Nonnull |
| 91 | + public Config getConfig() { |
| 92 | + return config; |
| 93 | + } |
| 94 | + |
| 95 | + @Nonnull |
| 96 | + @Override |
| 97 | + public NodeFactory<N> getNodeFactory() { |
| 98 | + return nodeFactory; |
| 99 | + } |
| 100 | + |
| 101 | + @Override |
| 102 | + @Nonnull |
| 103 | + public Subspace getSubspace() { |
| 104 | + return subspace; |
| 105 | + } |
| 106 | + |
| 107 | + /** |
| 108 | + * Gets the cached subspace for the data associated with this component. |
| 109 | + * <p> |
| 110 | + * The data subspace defines the portion of the directory space where the data |
| 111 | + * for this component is stored. |
| 112 | + * |
| 113 | + * @return the non-null {@link Subspace} for the data |
| 114 | + */ |
| 115 | + @Override |
| 116 | + @Nonnull |
| 117 | + public Subspace getDataSubspace() { |
| 118 | + return dataSubspace; |
| 119 | + } |
| 120 | + |
| 121 | + @Override |
| 122 | + @Nonnull |
| 123 | + public OnWriteListener getOnWriteListener() { |
| 124 | + return onWriteListener; |
| 125 | + } |
| 126 | + |
| 127 | + @Override |
| 128 | + @Nonnull |
| 129 | + public OnReadListener getOnReadListener() { |
| 130 | + return onReadListener; |
| 131 | + } |
| 132 | + |
| 133 | + /** |
| 134 | + * Asynchronously fetches a node from a specific layer of the HNSW. |
| 135 | + * <p> |
| 136 | + * The node is identified by its {@code layer} and {@code primaryKey}. The entire fetch operation is |
| 137 | + * performed within the given {@link ReadTransaction}. After the underlying |
| 138 | + * fetch operation completes, the retrieved node is validated by the |
| 139 | + * {@link #checkNode(Node)} method before the returned future is completed. |
| 140 | + * |
| 141 | + * @param readTransaction the non-null transaction to use for the read operation |
| 142 | + * @param storageTransform an affine vector transformation operator that is used to transform the fetched vector |
| 143 | + * into the storage space that is currently being used |
| 144 | + * @param layer the layer of the tree from which to fetch the node |
| 145 | + * @param primaryKey the non-null primary key that identifies the node to fetch |
| 146 | + * |
| 147 | + * @return a {@link CompletableFuture} that will complete with the fetched {@link AbstractNode} |
| 148 | + * once it has been read from storage and validated |
| 149 | + */ |
| 150 | + @Nonnull |
| 151 | + @Override |
| 152 | + public CompletableFuture<AbstractNode<N>> fetchNode(@Nonnull final ReadTransaction readTransaction, |
| 153 | + @Nonnull final AffineOperator storageTransform, |
| 154 | + int layer, @Nonnull Tuple primaryKey) { |
| 155 | + return fetchNodeInternal(readTransaction, storageTransform, layer, primaryKey).thenApply(this::checkNode); |
| 156 | + } |
| 157 | + |
| 158 | + /** |
| 159 | + * Asynchronously fetches a specific node from the data store for a given layer and primary key. |
| 160 | + * <p> |
| 161 | + * This is an internal, abstract method that concrete subclasses must implement to define |
| 162 | + * the storage-specific logic for retrieving a node. The operation is performed within the |
| 163 | + * context of the provided {@link ReadTransaction}. |
| 164 | + * |
| 165 | + * @param readTransaction the transaction to use for the read operation; must not be {@code null} |
| 166 | + * @param storageTransform an affine vector transformation operator that is used to transform the fetched vector |
| 167 | + * into the storage space that is currently being used |
| 168 | + * @param layer the layer index from which to fetch the node |
| 169 | + * @param primaryKey the primary key that uniquely identifies the node to be fetched; must not be {@code null} |
| 170 | + * |
| 171 | + * @return a {@link CompletableFuture} that will be completed with the fetched {@link AbstractNode}. |
| 172 | + * The future will complete with {@code null} if no node is found for the given key and layer. |
| 173 | + */ |
| 174 | + @Nonnull |
| 175 | + protected abstract CompletableFuture<AbstractNode<N>> fetchNodeInternal(@Nonnull ReadTransaction readTransaction, |
| 176 | + @Nonnull AffineOperator storageTransform, |
| 177 | + int layer, @Nonnull Tuple primaryKey); |
| 178 | + |
| 179 | + /** |
| 180 | + * Method to perform basic invariant check(s) on a newly-fetched node. |
| 181 | + * |
| 182 | + * @param node the node to check |
| 183 | + * was passed in |
| 184 | + * |
| 185 | + * @return the node that was passed in |
| 186 | + */ |
| 187 | + @Nullable |
| 188 | + private <T extends Node<N>> T checkNode(@Nullable final T node) { |
| 189 | + return node; |
| 190 | + } |
| 191 | + |
| 192 | + /** |
| 193 | + * Writes a given node and its neighbor modifications to the underlying storage. |
| 194 | + * <p> |
| 195 | + * This operation is executed within the context of the provided {@link Transaction}. |
| 196 | + * It handles persisting the node's data at a specific {@code layer} and applies |
| 197 | + * the changes to its neighbors as defined in the {@link NeighborsChangeSet}. |
| 198 | + * This method delegates the core writing logic to an internal method and provides |
| 199 | + * debug logging upon completion. |
| 200 | + * |
| 201 | + * @param transaction the non-null {@link Transaction} context for this write operation |
| 202 | + * @param quantizer the quantizer to use |
| 203 | + * @param node the non-null {@link Node} to be written to storage |
| 204 | + * @param layer the layer index where the node is being written |
| 205 | + * @param changeSet the non-null {@link NeighborsChangeSet} detailing the modifications |
| 206 | + * to the node's neighbors |
| 207 | + */ |
| 208 | + @Override |
| 209 | + public void writeNode(@Nonnull final Transaction transaction, @Nonnull final Quantizer quantizer, |
| 210 | + @Nonnull final AbstractNode<N> node, final int layer, |
| 211 | + @Nonnull final NeighborsChangeSet<N> changeSet) { |
| 212 | + writeNodeInternal(transaction, quantizer, node, layer, changeSet); |
| 213 | + if (logger.isTraceEnabled()) { |
| 214 | + logger.trace("written node with key={} at layer={}", node.getPrimaryKey(), layer); |
| 215 | + } |
| 216 | + } |
| 217 | + |
| 218 | + /** |
| 219 | + * Writes a single node to the data store as part of a larger transaction. |
| 220 | + * <p> |
| 221 | + * This is an abstract method that concrete implementations must provide. |
| 222 | + * It is responsible for the low-level persistence of the given {@code node} at a |
| 223 | + * specific {@code layer}. The implementation should also handle the modifications |
| 224 | + * to the node's neighbors, as detailed in the {@code changeSet}. |
| 225 | + * |
| 226 | + * @param transaction the non-null transaction context for the write operation |
| 227 | + * @param quantizer the quantizer to use |
| 228 | + * @param node the non-null {@link Node} to write |
| 229 | + * @param layer the layer or level of the node in the structure |
| 230 | + * @param changeSet the non-null {@link NeighborsChangeSet} detailing additions or |
| 231 | + * removals of neighbor links |
| 232 | + */ |
| 233 | + protected abstract void writeNodeInternal(@Nonnull Transaction transaction, @Nonnull Quantizer quantizer, |
| 234 | + @Nonnull AbstractNode<N> node, int layer, |
| 235 | + @Nonnull NeighborsChangeSet<N> changeSet); |
| 236 | +} |
0 commit comments