adelapena commented on code in PR #2673: URL: https://github.com/apache/cassandra/pull/2673#discussion_r1341318727
########## test/distributed/org/apache/cassandra/distributed/test/sai/VectorDistributedTest.java: ########## @@ -0,0 +1,475 @@ +/* + * 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.cassandra.distributed.test.sai; + +import java.util.ArrayList; +import java.util.Arrays; +import java.util.Iterator; +import java.util.List; +import java.util.Map; +import java.util.concurrent.atomic.AtomicInteger; +import java.util.stream.Collectors; +import java.util.stream.IntStream; + +import com.google.common.collect.ArrayListMultimap; +import com.google.common.collect.Multimap; +import org.junit.After; +import org.junit.AfterClass; +import org.junit.Before; +import org.junit.BeforeClass; +import org.junit.Rule; +import org.junit.Test; + +import io.github.jbellis.jvector.vector.VectorSimilarityFunction; +import org.apache.cassandra.cql3.statements.SelectStatement; +import org.apache.cassandra.db.Keyspace; +import org.apache.cassandra.db.marshal.Int32Type; +import org.apache.cassandra.dht.Murmur3Partitioner; +import org.apache.cassandra.distributed.Cluster; +import org.apache.cassandra.distributed.api.ConsistencyLevel; +import org.apache.cassandra.distributed.test.TestBaseImpl; +import org.apache.cassandra.index.sai.SAITester; +import org.apache.cassandra.index.sai.disk.v1.IndexWriterConfig; + +import static org.apache.cassandra.distributed.api.Feature.GOSSIP; +import static org.apache.cassandra.distributed.api.Feature.NETWORK; +import static org.assertj.core.api.Assertions.assertThat; +import static org.assertj.core.api.Assertions.assertThatThrownBy; + +public class VectorDistributedTest extends TestBaseImpl +{ + + @Rule + public SAITester.FailureWatcher failureRule = new SAITester.FailureWatcher(); + + private static final String CREATE_KEYSPACE = "CREATE KEYSPACE %%s WITH replication = {'class': 'SimpleStrategy', 'replication_factor': %d}"; + private static final String CREATE_TABLE = "CREATE TABLE %%s (pk int primary key, val vector<float, %d>)"; + private static final String CREATE_INDEX = "CREATE CUSTOM INDEX ON %%s(%s) USING 'StorageAttachedIndex'"; + + private static final VectorSimilarityFunction function = IndexWriterConfig.DEFAULT_SIMILARITY_FUNCTION; + + private static final String INVALID_LIMIT_MESSAGE = "Use of ANN OF in an ORDER BY clause requires a LIMIT that is not greater than 1000"; + + private static final double MIN_RECALL = 0.8; + + private static final int NUM_REPLICAS = 3; + private static final int RF = 2; + + private static final AtomicInteger seq = new AtomicInteger(); + private static String table; + + private static Cluster cluster; + + private static int dimensionCount; + + @BeforeClass + public static void setupCluster() throws Exception + { + cluster = Cluster.build(NUM_REPLICAS) + .withTokenCount(1) // VSTODO in-jvm-test in CC branch doesn't support multiple tokens + .withDataDirCount(1) // VSTODO vector memtable flush doesn't support multiple directories yet + .withConfig(config -> config.with(GOSSIP).with(NETWORK)) + .start(); + + cluster.schemaChange(withKeyspace(String.format(CREATE_KEYSPACE, RF))); + } + + @AfterClass + public static void closeCluster() + { + if (cluster != null) + cluster.close(); + } + + @Before + public void before() + { + table = "table_" + seq.getAndIncrement(); + dimensionCount = SAITester.getRandom().nextIntBetween(100, 2048); + } + + @After + public void after() + { + cluster.schemaChange(formatQuery("DROP TABLE IF EXISTS %s")); + } + + @Test + public void topKQueryFailsUnlessAtConsistencyLevelOne() + { + cluster.schemaChange(formatQuery(String.format(CREATE_TABLE, dimensionCount))); + cluster.schemaChange(formatQuery(String.format(CREATE_INDEX, "val"))); + SAIUtil.waitForIndexQueryable(cluster, KEYSPACE); + + int vectorCount = SAITester.getRandom().nextIntBetween(500, 1000); + List<float[]> vectors = generateVectors(vectorCount); + + int pk = 0; + for (float[] vector : vectors) + execute("INSERT INTO %s (pk, val) VALUES (" + (pk++) + ", " + vectorString(vector) + " )"); + + // query memtable index + int limit = Math.min(SAITester.getRandom().nextIntBetween(10, 50), vectors.size()); + float[] queryVector = randomVector(); + assertThatThrownBy(() -> execute("SELECT val FROM %s ORDER BY val ann of " + Arrays.toString(queryVector) + " LIMIT " + limit, ConsistencyLevel.QUORUM)) + .hasMessage(SelectStatement.TOPK_CONSISTENCY_LEVEL_ERROR); + + Object[][] result = execute("SELECT val FROM %s ORDER BY val ann of " + Arrays.toString(queryVector) + " LIMIT " + limit, ConsistencyLevel.ONE); + assertThat(result).hasNumberOfRows(limit); + List<float[]> resultVectors = getVectors(result); + assertDescendingScore(queryVector, resultVectors); + double memtableRecall = getRecall(vectors, queryVector, resultVectors); + assertThat(memtableRecall).isGreaterThanOrEqualTo(MIN_RECALL); + } + + @Test + public void testVectorSearch() + { + cluster.schemaChange(formatQuery(String.format(CREATE_TABLE, dimensionCount))); + cluster.schemaChange(formatQuery(String.format(CREATE_INDEX, "val"))); + SAIUtil.waitForIndexQueryable(cluster, KEYSPACE); + + int vectorCount = SAITester.getRandom().nextIntBetween(500, 1000); + List<float[]> vectors = generateVectors(vectorCount); + + int pk = 0; + for (float[] vector : vectors) + execute("INSERT INTO %s (pk, val) VALUES (" + (pk++) + ", " + vectorString(vector) + " )"); + + // query memtable index + int limit = Math.min(SAITester.getRandom().nextIntBetween(10, 50), vectors.size()); + float[] queryVector = randomVector(); + Object[][] result = searchWithLimit(queryVector, limit); + + List<float[]> resultVectors = getVectors(result); + assertDescendingScore(queryVector, resultVectors); + double memtableRecall = getRecall(vectors, queryVector, resultVectors); + assertThat(memtableRecall).isGreaterThanOrEqualTo(MIN_RECALL); + + assertThatThrownBy(() -> searchWithoutLimit(randomVector(), vectorCount)) + .hasMessageContaining(INVALID_LIMIT_MESSAGE); + + int pageSize = SAITester.getRandom().nextIntBetween(40, 70); + limit = SAITester.getRandom().nextIntBetween(20, 50); + result = searchWithPageAndLimit(queryVector, pageSize, limit); + + resultVectors = getVectors(result); + assertDescendingScore(queryVector, resultVectors); + double memtableRecallWithPaging = getRecall(vectors, queryVector, resultVectors); + assertThat(memtableRecallWithPaging).isGreaterThanOrEqualTo(MIN_RECALL); + + assertThatThrownBy(() -> searchWithPageWithoutLimit(randomVector())) + .hasMessageContaining(INVALID_LIMIT_MESSAGE); + + // query on-disk index + cluster.forEach(n -> n.flush(KEYSPACE)); + + limit = Math.min(SAITester.getRandom().nextIntBetween(10, 50), vectors.size()); + queryVector = randomVector(); + result = searchWithLimit(queryVector, limit); + double sstableRecall = getRecall(vectors, queryVector, getVectors(result)); + assertThat(sstableRecall).isGreaterThanOrEqualTo(MIN_RECALL); + } + + @Test + public void testMultiSSTablesVectorSearch() + { + cluster.schemaChange(formatQuery(String.format(CREATE_TABLE, dimensionCount))); + cluster.schemaChange(formatQuery(String.format(CREATE_INDEX, "val"))); + SAIUtil.waitForIndexQueryable(cluster, KEYSPACE); + // disable compaction + String tableName = table; + cluster.forEach(n -> n.runOnInstance(() -> { + Keyspace keyspace = Keyspace.open(KEYSPACE); + keyspace.getColumnFamilyStore(tableName).disableAutoCompaction(); + })); + + int vectorCountPerSSTable = SAITester.getRandom().nextIntBetween(200, 500); + int sstableCount = SAITester.getRandom().nextIntBetween(3, 5); + List<float[]> allVectors = new ArrayList<>(sstableCount * vectorCountPerSSTable); + + int pk = 0; + for (int i = 0; i < sstableCount; i++) + { + List<float[]> vectors = generateVectors(vectorCountPerSSTable); + for (float[] vector : vectors) + execute("INSERT INTO %s (pk, val) VALUES (" + (pk++) + ", " + vectorString(vector) + " )"); + + allVectors.addAll(vectors); + cluster.forEach(n -> n.flush(KEYSPACE)); + } + + // query multiple sstable indexes in multiple node + int limit = Math.min(SAITester.getRandom().nextIntBetween(50, 100), allVectors.size()); + float[] queryVector = randomVector(); + Object[][] result = searchWithLimit(queryVector, limit); + + // expect recall to be at least 0.8 + List<float[]> resultVectors = getVectors(result); + assertDescendingScore(queryVector, resultVectors); + double recall = getRecall(allVectors, queryVector, getVectors(result)); + assertThat(recall).isGreaterThanOrEqualTo(MIN_RECALL); + } + + @Test + public void testPartitionRestrictedVectorSearch() + { + cluster.schemaChange(formatQuery(String.format(CREATE_TABLE, dimensionCount))); + cluster.schemaChange(formatQuery(String.format(CREATE_INDEX, "val"))); + SAIUtil.waitForIndexQueryable(cluster, KEYSPACE); + + int vectorCount = SAITester.getRandom().nextIntBetween(500, 1000); + List<float[]> vectors = generateVectors(vectorCount); + + int pk = 0; + for (float[] vector : vectors) + execute("INSERT INTO %s (pk, val) VALUES (" + (pk++) + ", " + vectorString(vector) + " )"); + + // query memtable index + for (int executionCount = 0; executionCount < 50; executionCount++) + { + int key = SAITester.getRandom().nextIntBetween(0, vectorCount - 1); + float[] queryVector = randomVector(); + searchByKeyWithLimit(key, queryVector, vectors); + } + + cluster.forEach(n -> n.flush(KEYSPACE)); + + // query on-disk index + for (int executionCount = 0; executionCount < 50; executionCount++) + { + int key = SAITester.getRandom().nextIntBetween(0, vectorCount - 1); + float[] queryVector = randomVector(); + searchByKeyWithLimit(key, queryVector, vectors); + } + } + + @Test + public void rangeRestrictedTest() + { + cluster.schemaChange(formatQuery(String.format(CREATE_TABLE, dimensionCount))); + cluster.schemaChange(formatQuery(String.format(CREATE_INDEX, "val"))); + SAIUtil.waitForIndexQueryable(cluster, KEYSPACE); + + int vectorCount = SAITester.getRandom().nextIntBetween(500, 1000); + List<float[]> vectors = IntStream.range(0, vectorCount).mapToObj(s -> randomVector()).collect(Collectors.toList()); + + int pk = 0; + Multimap<Long, float[]> vectorsByToken = ArrayListMultimap.create(); + for (float[] vector : vectors) + { + vectorsByToken.put(Murmur3Partitioner.instance.getToken(Int32Type.instance.decompose(pk)).getLongValue(), vector); + execute("INSERT INTO %s (pk, val) VALUES (" + (pk++) + ',' + vectorString(vector) + " )"); + } + + // query memtable index + for (int executionCount = 0; executionCount < 50; executionCount++) + { + int key1 = SAITester.getRandom().nextIntBetween(1, vectorCount * 2); + long token1 = Murmur3Partitioner.instance.getToken(Int32Type.instance.decompose(key1)).getLongValue(); + int key2 = SAITester.getRandom().nextIntBetween(1, vectorCount * 2); + long token2 = Murmur3Partitioner.instance.getToken(Int32Type.instance.decompose(key2)).getLongValue(); + + long minToken = Math.min(token1, token2); + long maxToken = Math.max(token1, token2); + List<float[]> expected = vectorsByToken.entries().stream() + .filter(e -> e.getKey() >= minToken && e.getKey() <= maxToken) + .map(Map.Entry::getValue) + .collect(Collectors.toList()); + + float[] queryVector = randomVector(); + List<float[]> resultVectors = searchWithRange(queryVector, minToken, maxToken, expected.size()); + if (expected.isEmpty()) + assertThat(resultVectors).isEmpty(); + else + { + double recall = getRecall(resultVectors, queryVector, expected); + assertThat(recall).isGreaterThanOrEqualTo(0.8); + } + } + + cluster.forEach(n -> n.flush(KEYSPACE)); + + // query on-disk index with existing key: + for (int executionCount = 0; executionCount < 50; executionCount++) + { + int key1 = SAITester.getRandom().nextIntBetween(1, vectorCount * 2); + long token1 = Murmur3Partitioner.instance.getToken(Int32Type.instance.decompose(key1)).getLongValue(); + int key2 = SAITester.getRandom().nextIntBetween(1, vectorCount * 2); + long token2 = Murmur3Partitioner.instance.getToken(Int32Type.instance.decompose(key2)).getLongValue(); + + long minToken = Math.min(token1, token2); + long maxToken = Math.max(token1, token2); + List<float[]> expected = vectorsByToken.entries().stream() + .filter(e -> e.getKey() >= minToken && e.getKey() <= maxToken) + .map(Map.Entry::getValue) + .collect(Collectors.toList()); + + float[] queryVector = randomVector(); + + List<float[]> resultVectors = searchWithRange(queryVector, minToken, maxToken, expected.size()); + if (expected.isEmpty()) + assertThat(resultVectors).isEmpty(); + else + { + double recall = getRecall(resultVectors, queryVector, expected); + assertThat(recall).isGreaterThanOrEqualTo(0.8); + } + } + } + + private List<float[]> searchWithRange(float[] queryVector, long minToken, long maxToken, int expectedSize) + { + Object[][] result = execute("SELECT val FROM %s WHERE token(pk) <= " + maxToken + " AND token(pk) >= " + minToken + " ORDER BY val ann of " + Arrays.toString(queryVector) + " LIMIT 1000"); + assertThat(result).hasNumberOfRows(expectedSize); + return getVectors(result); + } + + private Object[][] searchWithLimit(float[] queryVector, int limit) + { + Object[][] result = execute("SELECT val FROM %s ORDER BY val ann of " + Arrays.toString(queryVector) + " LIMIT " + limit); + assertThat(result).hasNumberOfRows(limit); + return result; + } + + private void searchWithoutLimit(float[] queryVector, int results) + { + Object[][] result = execute("SELECT val FROM %s ORDER BY val ann of " + Arrays.toString(queryVector)); + assertThat(result).hasNumberOfRows(results); + } + + + private void searchWithPageWithoutLimit(float[] queryVector) + { + executeWithPaging("SELECT val FROM %s ORDER BY val ann of " + Arrays.toString(queryVector), 10); + } + + private Object[][] searchWithPageAndLimit(float[] queryVector, int pageSize, int limit) + { + // we don't know how many will be returned in case of paging, because coordinator resumes from last-returned-row's partiton + return executeWithPaging("SELECT val FROM %s ORDER BY val ann of " + Arrays.toString(queryVector) + " LIMIT " + limit, pageSize); + } + + private void searchByKeyWithLimit(int key, float[] queryVector, List<float[]> vectors) + { + Object[][] result = execute("SELECT val FROM %s WHERE pk = " + key + " ORDER BY val ann of " + Arrays.toString(queryVector) + " LIMIT 1"); + assertThat(result).hasNumberOfRows(1); + float[] output = getVectors(result).get(0); + assertThat(output).isEqualTo(vectors.get(key)); + } + + private void assertDescendingScore(float[] queryVector, List<float[]> resultVectors) + { + float prevScore = -1; + for (float[] current : resultVectors) + { + float score = function.compare(current, queryVector); + if (prevScore >= 0) + assertThat(score).isLessThanOrEqualTo(prevScore); + + prevScore = score; + } + } + + private double getRecall(List<float[]> vectors, float[] query, List<float[]> result) + { + List<float[]> sortedVectors = new ArrayList<>(vectors); + sortedVectors.sort((a, b) -> Double.compare(function.compare(b, query), function.compare(a, query))); + + assertThat(sortedVectors).containsAll(result); + + List<float[]> nearestNeighbors = sortedVectors.subList(0, result.size()); + + int matches = 0; + for (float[] in : nearestNeighbors) + { + for (float[] out : result) + { + if (Arrays.compare(in, out) ==0) + { + matches++; + break; + } + } + } + + return matches * 1.0 / result.size(); + } + + private List<float[]> generateVectors(int vectorCount) + { + return IntStream.range(0, vectorCount).mapToObj(s -> randomVector()).collect(Collectors.toList()); + } + + private List<float[]> getVectors(Object[][] result) Review Comment: Nit: add `@SuppressWarnings("unchecked")` -- 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]
