/* * SPDX-License-Identifier: Apache-2.0 * * The OpenSearch Contributors require contributions made to * this file be licensed under the Apache-2.0 license or a * compatible open source license. */ /* * Licensed to Elasticsearch under one or more contributor * license agreements. See the NOTICE file distributed with * this work for additional information regarding copyright * ownership. Elasticsearch 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. */ /* * Modifications Copyright OpenSearch Contributors. See * GitHub history for details. */ package org.opensearch.cluster.metadata; import org.apache.logging.log4j.LogManager; import org.apache.logging.log4j.Logger; import org.apache.logging.log4j.message.ParameterizedMessage; import org.opensearch.action.ActionListener; import org.opensearch.action.admin.indices.mapping.put.PutMappingClusterStateUpdateRequest; import org.opensearch.cluster.AckedClusterStateTaskListener; import org.opensearch.cluster.ClusterState; import org.opensearch.cluster.ClusterStateTaskConfig; import org.opensearch.cluster.ClusterStateTaskExecutor; import org.opensearch.cluster.ack.ClusterStateUpdateResponse; import org.opensearch.cluster.node.DiscoveryNode; import org.opensearch.cluster.service.ClusterManagerTaskKeys; import org.opensearch.cluster.service.ClusterManagerTaskThrottler; import org.opensearch.cluster.service.ClusterService; import org.opensearch.common.Nullable; import org.opensearch.common.Priority; import org.opensearch.common.compress.CompressedXContent; import org.opensearch.common.inject.Inject; import org.opensearch.common.unit.TimeValue; import org.opensearch.common.util.io.IOUtils; import org.opensearch.core.common.Strings; import org.opensearch.core.index.Index; import org.opensearch.index.IndexService; import org.opensearch.index.mapper.DocumentMapper; import org.opensearch.index.mapper.MapperService; import org.opensearch.index.mapper.MapperService.MergeReason; import org.opensearch.indices.IndicesService; import java.io.IOException; import java.util.ArrayList; import java.util.Collections; import java.util.HashMap; import java.util.List; import java.util.Map; import static org.opensearch.indices.cluster.IndicesClusterStateService.AllocatedIndices.IndexRemovalReason.NO_LONGER_ASSIGNED; /** * Service responsible for submitting mapping changes * * @opensearch.internal */ public class MetadataMappingService { private static final Logger logger = LogManager.getLogger(MetadataMappingService.class); private final ClusterService clusterService; private final IndicesService indicesService; private final ClusterManagerTaskThrottler.ThrottlingKey putMappingTaskKey; final RefreshTaskExecutor refreshExecutor = new RefreshTaskExecutor(); final PutMappingExecutor putMappingExecutor = new PutMappingExecutor(); @Inject public MetadataMappingService(ClusterService clusterService, IndicesService indicesService) { this.clusterService = clusterService; this.indicesService = indicesService; // Task is onboarded for throttling, it will get retried from associated TransportClusterManagerNodeAction. putMappingTaskKey = clusterService.registerClusterManagerTask(ClusterManagerTaskKeys.PUT_MAPPING_KEY, true); } static class RefreshTask { final String index; final String indexUUID; RefreshTask(String index, final String indexUUID) { this.index = index; this.indexUUID = indexUUID; } @Override public String toString() { return "[" + index + "][" + indexUUID + "]"; } } class RefreshTaskExecutor implements ClusterStateTaskExecutor { @Override public ClusterTasksResult execute(ClusterState currentState, List tasks) throws Exception { ClusterState newClusterState = executeRefresh(currentState, tasks); return ClusterTasksResult.builder().successes(tasks).build(newClusterState); } } /** * Batch method to apply all the queued refresh operations. The idea is to try and batch as much * as possible so we won't create the same index all the time for example for the updates on the same mapping * and generate a single cluster change event out of all of those. */ ClusterState executeRefresh(final ClusterState currentState, final List allTasks) throws Exception { // break down to tasks per index, so we can optimize the on demand index service creation // to only happen for the duration of a single index processing of its respective events Map> tasksPerIndex = new HashMap<>(); for (RefreshTask task : allTasks) { if (task.index == null) { logger.debug("ignoring a mapping task of type [{}] with a null index.", task); } tasksPerIndex.computeIfAbsent(task.index, k -> new ArrayList<>()).add(task); } boolean dirty = false; Metadata.Builder mdBuilder = Metadata.builder(currentState.metadata()); for (Map.Entry> entry : tasksPerIndex.entrySet()) { IndexMetadata indexMetadata = mdBuilder.get(entry.getKey()); if (indexMetadata == null) { // index got deleted on us, ignore... logger.debug("[{}] ignoring tasks - index meta data doesn't exist", entry.getKey()); continue; } final Index index = indexMetadata.getIndex(); // the tasks lists to iterate over, filled with the list of mapping tasks, trying to keep // the latest (based on order) update mapping one per node List allIndexTasks = entry.getValue(); boolean hasTaskWithRightUUID = false; for (RefreshTask task : allIndexTasks) { if (indexMetadata.isSameUUID(task.indexUUID)) { hasTaskWithRightUUID = true; } else { logger.debug("{} ignoring task [{}] - index meta data doesn't match task uuid", index, task); } } if (hasTaskWithRightUUID == false) { continue; } // construct the actual index if needed, and make sure the relevant mappings are there boolean removeIndex = false; IndexService indexService = indicesService.indexService(indexMetadata.getIndex()); if (indexService == null) { // we need to create the index here, and add the current mapping to it, so we can merge indexService = indicesService.createIndex(indexMetadata, Collections.emptyList(), false); removeIndex = true; indexService.mapperService().merge(indexMetadata, MergeReason.MAPPING_RECOVERY); } IndexMetadata.Builder builder = IndexMetadata.builder(indexMetadata); try { boolean indexDirty = refreshIndexMapping(indexService, builder); if (indexDirty) { mdBuilder.put(builder); dirty = true; } } finally { if (removeIndex) { indicesService.removeIndex(index, NO_LONGER_ASSIGNED, "created for mapping processing"); } } } if (!dirty) { return currentState; } return ClusterState.builder(currentState).metadata(mdBuilder).build(); } private boolean refreshIndexMapping(IndexService indexService, IndexMetadata.Builder builder) { boolean dirty = false; String index = indexService.index().getName(); try { MapperService mapperService = indexService.mapperService(); DocumentMapper mapper = mapperService.documentMapper(); if (mapper != null) { if (mapper.mappingSource().equals(builder.mapping().source()) == false) { dirty = true; } } } catch (Exception e) { logger.warn(() -> new ParameterizedMessage("[{}] failed to refresh-mapping in cluster state", index), e); } return dirty; } /** * Refreshes mappings if they are not the same between original and parsed version */ public void refreshMapping(final String index, final String indexUUID) { final RefreshTask refreshTask = new RefreshTask(index, indexUUID); clusterService.submitStateUpdateTask( "refresh-mapping [" + index + "]", refreshTask, ClusterStateTaskConfig.build(Priority.HIGH), refreshExecutor, (source, e) -> logger.warn(() -> new ParameterizedMessage("failure during [{}]", source), e) ); } class PutMappingExecutor implements ClusterStateTaskExecutor { @Override public ClusterTasksResult execute( ClusterState currentState, List tasks ) throws Exception { Map indexMapperServices = new HashMap<>(); ClusterTasksResult.Builder builder = ClusterTasksResult.builder(); try { for (PutMappingClusterStateUpdateRequest request : tasks) { try { for (Index index : request.indices()) { final IndexMetadata indexMetadata = currentState.metadata().getIndexSafe(index); if (indexMapperServices.containsKey(indexMetadata.getIndex()) == false) { MapperService mapperService = indicesService.createIndexMapperService(indexMetadata); indexMapperServices.put(index, mapperService); // add mappings for all types, we need them for cross-type validation mapperService.merge(indexMetadata, MergeReason.MAPPING_RECOVERY); } } currentState = applyRequest(currentState, request, indexMapperServices); builder.success(request); } catch (Exception e) { builder.failure(request, e); } } return builder.build(currentState); } finally { IOUtils.close(indexMapperServices.values()); } } @Override public ClusterManagerTaskThrottler.ThrottlingKey getClusterManagerThrottlingKey() { return putMappingTaskKey; } private ClusterState applyRequest( ClusterState currentState, PutMappingClusterStateUpdateRequest request, Map indexMapperServices ) throws IOException { CompressedXContent mappingUpdateSource = new CompressedXContent(request.source()); final Metadata metadata = currentState.metadata(); final List updateList = new ArrayList<>(); for (Index index : request.indices()) { MapperService mapperService = indexMapperServices.get(index); // IMPORTANT: always get the metadata from the state since it get's batched // and if we pull it from the indexService we might miss an update etc. final IndexMetadata indexMetadata = currentState.getMetadata().getIndexSafe(index); // this is paranoia... just to be sure we use the exact same metadata tuple on the update that // we used for the validation, it makes this mechanism little less scary (a little) updateList.add(indexMetadata); // try and parse it (no need to add it here) so we can bail early in case of parsing exception DocumentMapper existingMapper = mapperService.documentMapper(); DocumentMapper newMapper = mapperService.parse(MapperService.SINGLE_MAPPING_NAME, mappingUpdateSource); if (existingMapper != null) { // first, simulate: just call merge and ignore the result existingMapper.merge(newMapper.mapping(), MergeReason.MAPPING_UPDATE); } } Metadata.Builder builder = Metadata.builder(metadata); boolean updated = false; for (IndexMetadata indexMetadata : updateList) { boolean updatedMapping = false; // do the actual merge here on the master, and update the mapping source // we use the exact same indexService and metadata we used to validate above here to actually apply the update final Index index = indexMetadata.getIndex(); final MapperService mapperService = indexMapperServices.get(index); CompressedXContent existingSource = null; DocumentMapper existingMapper = mapperService.documentMapper(); if (existingMapper != null) { existingSource = existingMapper.mappingSource(); } DocumentMapper mergedMapper = mapperService.merge( MapperService.SINGLE_MAPPING_NAME, mappingUpdateSource, MergeReason.MAPPING_UPDATE ); CompressedXContent updatedSource = mergedMapper.mappingSource(); if (existingSource != null) { if (existingSource.equals(updatedSource)) { // same source, no changes, ignore it } else { updatedMapping = true; // use the merged mapping source if (logger.isDebugEnabled()) { logger.debug("{} update_mapping [{}] with source [{}]", index, mergedMapper.type(), updatedSource); } else if (logger.isInfoEnabled()) { logger.info("{} update_mapping [{}]", index, mergedMapper.type()); } } } else { updatedMapping = true; if (logger.isDebugEnabled()) { logger.debug("{} create_mapping with source [{}]", index, updatedSource); } else if (logger.isInfoEnabled()) { logger.info("{} create_mapping", index); } } IndexMetadata.Builder indexMetadataBuilder = IndexMetadata.builder(indexMetadata); // Mapping updates on a single type may have side-effects on other types so we need to // update mapping metadata on all types DocumentMapper mapper = mapperService.documentMapper(); if (mapper != null) { indexMetadataBuilder.putMapping(new MappingMetadata(mapper.mappingSource())); } if (updatedMapping) { indexMetadataBuilder.mappingVersion(1 + indexMetadataBuilder.mappingVersion()); } /* * This implicitly increments the index metadata version and builds the index metadata. This means that we need to have * already incremented the mapping version if necessary. Therefore, the mapping version increment must remain before this * statement. */ builder.put(indexMetadataBuilder); updated |= updatedMapping; } if (updated) { return ClusterState.builder(currentState).metadata(builder).build(); } else { return currentState; } } } public void putMapping(final PutMappingClusterStateUpdateRequest request, final ActionListener listener) { clusterService.submitStateUpdateTask( "put-mapping " + Strings.arrayToCommaDelimitedString(request.indices()), request, ClusterStateTaskConfig.build(Priority.HIGH, request.masterNodeTimeout()), putMappingExecutor, new AckedClusterStateTaskListener() { @Override public void onFailure(String source, Exception e) { listener.onFailure(e); } @Override public boolean mustAck(DiscoveryNode discoveryNode) { return true; } @Override public void onAllNodesAcked(@Nullable Exception e) { listener.onResponse(new ClusterStateUpdateResponse(e == null)); } @Override public void onAckTimeout() { listener.onResponse(new ClusterStateUpdateResponse(false)); } @Override public TimeValue ackTimeout() { return request.ackTimeout(); } } ); } }