elasticsearch index 之 put mapping

標籤：idt   詳細   cut   last   沒有   pes   span   建立   arc   

mapping機制使得elasticsearch索引資料變的更加靈活，近乎於no schema。mapping可以在建立索引時設定，也可以在後期設定。後期設定可以是修改mapping（無法對已有的field屬性進行修改，一般來說只是增加新的field）或者對沒有mapping的索引設定mapping。put mapping操作必須是master節點來完成，因為它涉及到叢集matedata的修改，同時它跟index和type密切相關。修改只是針對特定index的特定type。

在Action support分析中我們分析過幾種Action的抽象類別型，put mapping Action屬於TransportMasterNodeOperationAction的子類。它實現了masterOperation方法，每個繼承自TransportMasterNodeOperationAction的子類都會根據自己的具體功能來實現這個方法。這裡的實現如下所示：

    protected void masterOperation(final PutMappingRequest request, final ClusterState state, final ActionListener<PutMappingResponse> listener) throws ElasticsearchException {        final String[] concreteIndices = clusterService.state().metaData().concreteIndices(request.indicesOptions(), request.indices());
　　　　　　//構造request        PutMappingClusterStateUpdateRequest updateRequest = new PutMappingClusterStateUpdateRequest()                .ackTimeout(request.timeout()).masterNodeTimeout(request.masterNodeTimeout())                .indices(concreteIndices).type(request.type())                .source(request.source()).ignoreConflicts(request.ignoreConflicts());　　　　　　//調用putMapping方法，同時傳入一個Listener        metaDataMappingService.putMapping(updateRequest, new ActionListener<ClusterStateUpdateResponse>() {            @Override            public void onResponse(ClusterStateUpdateResponse response) {                listener.onResponse(new PutMappingResponse(response.isAcknowledged()));            }            @Override            public void onFailure(Throwable t) {                logger.debug("failed to put mappings on indices [{}], type [{}]", t, concreteIndices, request.type());                listener.onFailure(t);            }        });    }

以上是TransportPutMappingAction對masterOperation方法的實現，這裡並沒有多少複雜的邏輯和操作。具體操作在matedataMappingService中。跟之前的CreateIndex一樣，put Mapping也是向master提交一個updateTask。所有邏輯也都在execute方法中。這個task的基本跟CreateIndex一樣，也需要在給定的時間內響應。它的代碼如下所示：

 public void putMapping(final PutMappingClusterStateUpdateRequest request, final ActionListener<ClusterStateUpdateResponse> listener) {　　　　//提交一個高基本的updateTask        clusterService.submitStateUpdateTask("put-mapping [" + request.type() + "]", Priority.HIGH, new AckedClusterStateUpdateTask<ClusterStateUpdateResponse>(request, listener) {            @Override            protected ClusterStateUpdateResponse newResponse(boolean acknowledged) {                return new ClusterStateUpdateResponse(acknowledged);            }            @Override            public ClusterState execute(final ClusterState currentState) throws Exception {                List<String> indicesToClose = Lists.newArrayList();                try {
　　　　　　　　　　　　//必須針對已經在matadata中存在的index，否則拋出異常                    for (String index : request.indices()) {                        if (!currentState.metaData().hasIndex(index)) {                            throw new IndexMissingException(new Index(index));                        }                    }                    //還需要存在於indices中，否則無法進行操作。所以這裡要進行預建                    for (String index : request.indices()) {                        if (indicesService.hasIndex(index)) {                            continue;                        }                        final IndexMetaData indexMetaData = currentState.metaData().index(index);
　　　　　　　　　　　　　　//不存在就進行建立                        IndexService indexService = indicesService.createIndex(indexMetaData.index(), indexMetaData.settings(), clusterService.localNode().id());                        indicesToClose.add(indexMetaData.index());                        // make sure to add custom default mapping if exists                        if (indexMetaData.mappings().containsKey(MapperService.DEFAULT_MAPPING)) {                            indexService.mapperService().merge(MapperService.DEFAULT_MAPPING, indexMetaData.mappings().get(MapperService.DEFAULT_MAPPING).source(), false);                        }                        // only add the current relevant mapping (if exists)                        if (indexMetaData.mappings().containsKey(request.type())) {                            indexService.mapperService().merge(request.type(), indexMetaData.mappings().get(request.type()).source(), false);                        }                    }　　　　　　　　　　　　//合并更新Mapping                    Map<String, DocumentMapper> newMappers = newHashMap();                    Map<String, DocumentMapper> existingMappers = newHashMap();
　　　　　　　　　　　　//針對每個index進行Mapping合并                    for (String index : request.indices()) {                        IndexService indexService = indicesService.indexServiceSafe(index);                        // try and parse it (no need to add it here) so we can bail early in case of parsing exception                        DocumentMapper newMapper;                        DocumentMapper existingMapper = indexService.mapperService().documentMapper(request.type());                        if (MapperService.DEFAULT_MAPPING.equals(request.type())) {//存在defaultmapping則合并default mapping                            // _default_ types do not go through merging, but we do test the new settings. Also don‘t apply the old default                            newMapper = indexService.mapperService().parse(request.type(), new CompressedString(request.source()), false);                        } else {                            newMapper = indexService.mapperService().parse(request.type(), new CompressedString(request.source()), existingMapper == null);                            if (existingMapper != null) {                                // first, simulate                                DocumentMapper.MergeResult mergeResult = existingMapper.merge(newMapper, mergeFlags().simulate(true));                                // if we have conflicts, and we are not supposed to ignore them, throw an exception                                if (!request.ignoreConflicts() && mergeResult.hasConflicts()) {                                    throw new MergeMappingException(mergeResult.conflicts());                                }                            }                        }                        newMappers.put(index, newMapper);                        if (existingMapper != null) {                            existingMappers.put(index, existingMapper);                        }                    }                    String mappingType = request.type();                    if (mappingType == null) {                        mappingType = newMappers.values().iterator().next().type();                    } else if (!mappingType.equals(newMappers.values().iterator().next().type())) {                        throw new InvalidTypeNameException("Type name provided does not match type name within mapping definition");                    }                    if (!MapperService.DEFAULT_MAPPING.equals(mappingType) && !PercolatorService.TYPE_NAME.equals(mappingType) && mappingType.charAt(0) == ‘_‘) {                        throw new InvalidTypeNameException("Document mapping type name can‘t start with ‘_‘");                    }                    final Map<String, MappingMetaData> mappings = newHashMap();                    for (Map.Entry<String, DocumentMapper> entry : newMappers.entrySet()) {                        String index = entry.getKey();                        // do the actual merge here on the master, and update the mapping source                        DocumentMapper newMapper = entry.getValue();                        IndexService indexService = indicesService.indexService(index);                        if (indexService == null) {                            continue;                        }                        CompressedString existingSource = null;                        if (existingMappers.containsKey(entry.getKey())) {                            existingSource = existingMappers.get(entry.getKey()).mappingSource();                        }                        DocumentMapper mergedMapper = indexService.mapperService().merge(newMapper.type(), newMapper.mappingSource(), false);                        CompressedString updatedSource = mergedMapper.mappingSource();                        if (existingSource != null) {                            if (existingSource.equals(updatedSource)) {                                // same source, no changes, ignore it                            } else {                                // use the merged mapping source                                mappings.put(index, new MappingMetaData(mergedMapper));                                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 {                            mappings.put(index, new MappingMetaData(mergedMapper));                            if (logger.isDebugEnabled()) {                                logger.debug("[{}] create_mapping [{}] with source [{}]", index, newMapper.type(), updatedSource);                            } else if (logger.isInfoEnabled()) {                                logger.info("[{}] create_mapping [{}]", index, newMapper.type());                            }                        }                    }                    if (mappings.isEmpty()) {                        // no changes, return                        return currentState;                    }　　　　　　　　　　　　//根據mapping的更新情況重建matadata                    MetaData.Builder builder = MetaData.builder(currentState.metaData());                    for (String indexName : request.indices()) {                        IndexMetaData indexMetaData = currentState.metaData().index(indexName);                        if (indexMetaData == null) {                            throw new IndexMissingException(new Index(indexName));                        }                        MappingMetaData mappingMd = mappings.get(indexName);                        if (mappingMd != null) {                            builder.put(IndexMetaData.builder(indexMetaData).putMapping(mappingMd));                        }                    }                    return ClusterState.builder(currentState).metaData(builder).build();                } finally {                    for (String index : indicesToClose) {                        indicesService.removeIndex(index, "created for mapping processing");                    }                }            }        });    }

 以上就是mapping的設定過程，首先它跟Create index一樣，只有master節點才能操作，而且是以task的形式提交給master；其次它的本質是將request中的mapping和index現存的或者是default mapping合并，並最終產生新的matadata更新到叢集的各個節點。

總結：叢集中的master操作無論是index方面還是叢集方面，最終都是叢集matadata的更新過程。而這些操作只能在master上進行，並且都是會逾時的任務。put mapping當然也不例外。上面的兩段代碼基本概況了mapping的設定過程。這裡就不再重複了。這裡還有一個問題沒有涉及到就是mapping的合并。mapping合并會在很多地方用到。在下一節中會它進行詳細分析。

elasticsearch index 之 put mapping