AMashenkov commented on a change in pull request #9118: URL: https://github.com/apache/ignite/pull/9118#discussion_r685860079
########## File path: modules/core/src/main/java/org/apache/ignite/internal/cache/query/index/IndexQueryProcessor.java ########## @@ -0,0 +1,449 @@ +/* + * 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.ignite.internal.cache.query.index; + +import java.util.ArrayList; +import java.util.Arrays; +import java.util.Collection; +import java.util.Comparator; +import java.util.List; +import java.util.NoSuchElementException; +import org.apache.ignite.IgniteCheckedException; +import org.apache.ignite.IgniteException; +import org.apache.ignite.cache.query.IndexQuery; +import org.apache.ignite.cache.query.IndexQueryCriteriaBuilder; +import org.apache.ignite.cache.query.IndexQueryCriterion; +import org.apache.ignite.internal.cache.query.RangeIndexQueryCriterion; +import org.apache.ignite.internal.cache.query.index.sorted.IndexKeyDefinition; +import org.apache.ignite.internal.cache.query.index.sorted.IndexKeyTypeSettings; +import org.apache.ignite.internal.cache.query.index.sorted.IndexRow; +import org.apache.ignite.internal.cache.query.index.sorted.IndexRowComparator; +import org.apache.ignite.internal.cache.query.index.sorted.IndexSearchRowImpl; +import org.apache.ignite.internal.cache.query.index.sorted.InlineIndexRowHandler; +import org.apache.ignite.internal.cache.query.index.sorted.SortedIndexDefinition; +import org.apache.ignite.internal.cache.query.index.sorted.inline.IndexQueryContext; +import org.apache.ignite.internal.cache.query.index.sorted.inline.InlineIndex; +import org.apache.ignite.internal.cache.query.index.sorted.keys.IndexKey; +import org.apache.ignite.internal.cache.query.index.sorted.keys.IndexKeyFactory; +import org.apache.ignite.internal.processors.cache.CacheObjectContext; +import org.apache.ignite.internal.processors.cache.CacheObjectUtils; +import org.apache.ignite.internal.processors.cache.GridCacheContext; +import org.apache.ignite.internal.processors.cache.query.IndexQueryDesc; +import org.apache.ignite.internal.processors.query.QueryUtils; +import org.apache.ignite.internal.util.GridCloseableIteratorAdapter; +import org.apache.ignite.internal.util.lang.GridCloseableIterator; +import org.apache.ignite.internal.util.lang.GridCursor; +import org.apache.ignite.internal.util.typedef.internal.U; +import org.apache.ignite.lang.IgniteBiTuple; + +import static org.apache.ignite.internal.cache.query.index.SortOrder.DESC; + +/** + * Processor of {@link IndexQuery}. + */ +public class IndexQueryProcessor { + /** */ + private final IndexProcessor idxProc; + + /** */ + public IndexQueryProcessor(IndexProcessor idxProc) { + this.idxProc = idxProc; + } + + /** Run query on local node. */ + public <K, V> GridCloseableIterator<IgniteBiTuple<K, V>> queryLocal( + GridCacheContext<K, V> cctx, IndexQueryDesc idxQryDesc, IndexQueryContext qryCtx, boolean keepBinary) + throws IgniteCheckedException { + + Index idx = index(cctx, idxQryDesc); + + if (idx == null) + throw new IgniteCheckedException( + "No index matches index query. Cache=" + cctx.name() + "; Qry=" + idxQryDesc); + + GridCursor<IndexRow> cursor = query(cctx, idx, idxQryDesc.criteria(), qryCtx); + + // Map IndexRow to Cache Key-Value pair. + return new GridCloseableIteratorAdapter<IgniteBiTuple<K, V>>() { + private IndexRow currVal; + + private final CacheObjectContext coctx = cctx.cacheObjectContext(); + + /** {@inheritDoc} */ + @Override protected boolean onHasNext() throws IgniteCheckedException { + if (currVal != null) + return true; + + if (!cursor.next()) + return false; + + currVal = cursor.get(); + + return true; + } + + /** {@inheritDoc} */ + @Override protected IgniteBiTuple<K, V> onNext() { + if (currVal == null) + if (!hasNext()) + throw new NoSuchElementException(); + + IndexRow row = currVal; + + currVal = null; + + K k = (K) CacheObjectUtils.unwrapBinaryIfNeeded(coctx, row.cacheDataRow().key(), keepBinary, false); + V v = (V) CacheObjectUtils.unwrapBinaryIfNeeded(coctx, row.cacheDataRow().value(), keepBinary, false); + + return new IgniteBiTuple<>(k, v); + } + }; + } + + /** Get index to run query by specified description. */ + private Index index(GridCacheContext cctx, IndexQueryDesc idxQryDesc) throws IgniteCheckedException { + Class<?> valCls = idxQryDesc.valCls() != null ? loadValClass(cctx, idxQryDesc.valCls()) : null; + + String tableName = cctx.kernalContext().query().tableName(cctx.name(), valCls); + + if (tableName == null) + return null; + + // Find index by specified name. + if (idxQryDesc.idxName() != null) { + String name = idxQryDesc.idxName(); + + if (!QueryUtils.PRIMARY_KEY_INDEX.equals(name)) + name = name.toUpperCase(); + + String schema = cctx.kernalContext().query().schemaName(cctx); + + IndexName idxName = new IndexName(cctx.name(), schema, tableName, name); + + Index idx = idxProc.index(idxName); + + if (idx == null) + return null; + + return checkIndex(idxProc.indexDefinition(idx.id()), idxQryDesc.criteria()) ? idx : null; + } + + // Try get index by list of fields to query. + // Check all indexes by cache to find best index match: count of index fields equals to count of index criteria fields. + Collection<Index> idxs = idxProc.indexes(cctx); + + for (Index i: idxs) { + IndexDefinition idxDef = idxProc.indexDefinition(i.id()); + + if (!tableName.equals(idxDef.idxName().tableName())) + continue; + + if (checkIndex(idxDef, idxQryDesc.criteria())) + return i; + } + + return null; + } + + /** Checks that specified index matches index query criteria. */ + private boolean checkIndex(IndexDefinition idxDef, List<IndexQueryCriterion> criteria) { + if (criteria.size() > idxDef.indexKeyDefinitions().size()) + return false; + + for (int i = 0; i < criteria.size(); i++) { + if (!idxDef.indexKeyDefinitions().get(i).name().equalsIgnoreCase(criteria.get(i).field())) + return false; + } + + return true; + } + + /** */ + private Class<?> loadValClass(GridCacheContext cctx, String valClsName) throws IgniteCheckedException { + try { + ClassLoader clsLdr = U.resolveClassLoader(cctx.kernalContext().config()); + + return clsLdr.loadClass(valClsName); + } + catch (ClassNotFoundException e) { + throw new IgniteCheckedException("No cache serves class: " + valClsName); + } + } + + /** Runs a query and return single cursor or cursor over multiple index segments. */ + private GridCursor<IndexRow> query(GridCacheContext cctx, Index idx, List<IndexQueryCriterion> criteria, IndexQueryContext qryCtx) + throws IgniteCheckedException { + + int segmentsCnt = cctx.isPartitioned() ? cctx.config().getQueryParallelism() : 1; + + if (segmentsCnt == 1) + return query(0, idx, criteria, qryCtx); + + final GridCursor<IndexRow>[] segments = new GridCursor[segmentsCnt]; + + // Actually it just traverse BPlusTree to find boundaries. It's too fast to parallelize this. + for (int i = 0; i < segmentsCnt; i++) + segments[i] = query(i, idx, criteria, qryCtx); + + return new SegmentedIndexCursor(segments, ((SortedIndexDefinition) idxProc.indexDefinition(idx.id())).rowComparator()); + } + + /** Coordinate query criteria. */ + private GridCursor<IndexRow> query(int segment, Index idx, List<IndexQueryCriterion> criteria, IndexQueryContext qryCtx) + throws IgniteCheckedException { + + assert !criteria.isEmpty() : "Index query criteria list has not to be empty."; + + if (criteria.get(0) instanceof RangeIndexQueryCriterion) + return treeIndexRange((InlineIndex) idx, criteria, segment, qryCtx); + + throw new IllegalStateException("Doesn't support index query criteria: " + criteria.getClass().getName()); + } + + /** + * Runs range query over specified segment. There are 2 steps to run query: + * 1. Traverse index by specified boundaries; + * 2. Scan over cursor and filter rows that doesn't match user criteria. + * + * Filtering is required in 2 cases: + * 1. Exclusion of one of boundaries, as idx.find() includes both of them; + * 2. To apply criteria on non-first index fields. Tree apply boundaries field by field, if first field match + * a boundary, then second field isn't checked within traversing. + */ + private GridCursor<IndexRow> treeIndexRange(InlineIndex idx, List<IndexQueryCriterion> criteria, int segment, + IndexQueryContext qryCtx) throws IgniteCheckedException { + + InlineIndexRowHandler hnd = idx.segment(0).rowHandler(); + CacheObjectContext coctx = idx.segment(0).cacheContext().cacheObjectContext(); + + IndexKey[] lowerBounds = new IndexKey[hnd.indexKeyDefinitions().size()]; + IndexKey[] upperBounds = new IndexKey[hnd.indexKeyDefinitions().size()]; + + boolean lowerAllNulls = true; + boolean upperAllNulls = true; + + List<RangeIndexQueryCriterion> treeCriteria = new ArrayList<>(); + + for (int i = 0; i < criteria.size(); i++) { + String f = criteria.get(i).field(); + + IndexKeyDefinition def = hnd.indexKeyDefinitions().get(i); + + if (!def.name().equalsIgnoreCase(f)) + throw new IgniteCheckedException("Range query doesn't match index '" + idx.name() + "'"); + + RangeIndexQueryCriterion c = (RangeIndexQueryCriterion) criteria.get(i); + + // If index is desc, then we need to swap boundaries as user declare criteria in straight manner. + // For example, there is an idx (int Val desc). It means that index stores data in reverse order (1 < 0). + // But user won't expect for criterion gt(1) to get 0 as result, instead user will use lt(1) for getting + // 0. Then we need to swap user criterion. + if (def.order().sortOrder() == DESC) + c = c.swap(); + + treeCriteria.add(c); + + IndexKey l = key(c.lower(), def, hnd.indexKeyTypeSettings(), coctx); + IndexKey u = key(c.upper(), def, hnd.indexKeyTypeSettings(), coctx); + + if (l != null) + lowerAllNulls = false; + + if (u != null) + upperAllNulls = false; + + lowerBounds[i] = l; + upperBounds[i] = u; + } + + IndexRow lower = lowerAllNulls ? null : new IndexSearchRowImpl(lowerBounds, hnd); + IndexRow upper = upperAllNulls ? null : new IndexSearchRowImpl(upperBounds, hnd); + + // Step 1. Traverse index. + GridCursor<IndexRow> findRes = idx.find(lower, upper, segment, qryCtx); + + // Step 2. Scan and filter. + return new GridCursor<IndexRow>() { + /** */ + private final IndexRowComparator rowCmp = ((SortedIndexDefinition) idxProc.indexDefinition(idx.id())).rowComparator(); + + /** {@inheritDoc} */ + @Override public boolean next() throws IgniteCheckedException { + if (!findRes.next()) + return false; + + while (exclude(get(), lower, 1) || exclude(get(), upper, -1)) { + if (!findRes.next()) + return false; + } Review comment: Got it. Assume, there is a composite index on (A,B). So, we need to scan contiguous range on A (top level) and filter out B (lower levels). As I understand, idx.find(...) already returns values that satisfy condition on A, and (as you wrote) don't filter out conditions on B. Why we need additional checks on A? Is this because we may need to filter out rows in case of exclusive bounds? If so, SortedIndex interface must specify whether bounds are inclusive or exclusive. Also, the rows that returned from the Cursor are materialized, even if they will be filtered out. It looks ok for now, but let's create a ticket for improvement. At first glance, IndexKey might have inclusive/exclusive flag, to avoid Row materialization if range has exclusive bounds. And we can think about passing a filter. -- 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]
