转自:http://www.lai18.com/content/7084969.html
Facet说明
我们在浏览网站的时候,经常会遇到按某一类条件查询的情况,这种情况尤以电商网站最多,以天猫商城为例,我们选择某一个品牌,系统会将该品牌对应的商品展示出来,效果图如下:
如上图,我们关注的是品牌,选购热点等方面,对于类似的功能我们用lucene的term查询当然可以,但是在数据量特别大的情况下还用普通查询来实现显然会因为FSDirectory.open等耗时的操作造成查询效率的低下,同时普通查询是全部document都扫描一遍,这样显然造成了查询效率低;
lucene提供了facet查询用于对同一类的document进行聚类化,这样在查询的时候先关注某一个方面,这种显然缩小了查询范围,进而提升了查询效率;
facet模块提供了多个用于处理facet的统计和值处理的方法;
要实现facet的功能,我们需要了解facetField,FacetField定义了dim和此field对应的path,需要特别注意的是我们在做facetField索引的时候,需要事先调用FacetsConfig.build(Document);
FacetField的indexOptions设置为了DOCS_AND_FREQS_AND_POSITIONS的,即既索引又统计出现的频次和出现的位置,这样做主要是为了方便查询和统计;
相应的在存储的时候我们需要利用FacetsConfig和DirectoryTaxonomyWriter;
DirectoryTaxonomyWriter用来利用Directory来存储Taxono信息到硬盘;
DirectoryTaxonomyWriter的构造器如下:
public DirectoryTaxonomyWriter(Directory directory, OpenMode openMode, TaxonomyWriterCache cache) throws IOException { dir = directory; IndexWriterConfig config = createIndexWriterConfig(openMode); indexWriter = openIndexWriter(dir, config); // verify (to some extent) that merge policy in effect would preserve category docids assert !(indexWriter.getConfig().getMergePolicy() instanceof TieredMergePolicy) : "for preserving category docids, merging none-adjacent segments is not allowed"; // after we opened the writer, and the index is locked, it's safe to check // the commit data and read the index epoch openMode = config.getOpenMode(); if (!DirectoryReader.indexExists(directory)) { indexEpoch = 1; } else { String epochStr = null; Map<String, String> commitData = readCommitData(directory); if (commitData != null) { epochStr = commitData.get(INDEX_EPOCH); } // no commit data, or no epoch in it means an old taxonomy, so set its epoch to 1, for lack // of a better value. indexEpoch = epochStr == null ? 1 : Long.parseLong(epochStr, 16); } if (openMode == OpenMode.CREATE) { ++indexEpoch; } FieldType ft = new FieldType(TextField.TYPE_NOT_STORED); ft.setOmitNorms(true); parentStreamField = new Field(Consts.FIELD_PAYLOADS, parentStream, ft); fullPathField = new StringField(Consts.FULL, "", Field.Store.YES); nextID = indexWriter.maxDoc(); if (cache == null) { cache = defaultTaxonomyWriterCache(); } this.cache = cache; if (nextID == 0) { cacheIsComplete = true; // Make sure that the taxonomy always contain the root category // with category id 0. addCategory(new FacetLabel()); } else { // There are some categories on the disk, which we have not yet // read into the cache, and therefore the cache is incomplete. // We choose not to read all the categories into the cache now, // to avoid terrible performance when a taxonomy index is opened // to add just a single category. We will do it later, after we // notice a few cache misses. cacheIsComplete = false; }
}
由上述代码可知,DirectoryTaxonomyWriter先打开一个IndexWriter,在确保indexWriter打开和locked的前提下,读取directory对应的segments中需要提交的内容,如果读取到的内容为空,说明是上次的内容,设置indexEpoch为1,接着对cache进行设置;判断directory中是否还包含有document,如果有设置cacheIsComplete为false,反之为true;
本文转自张昺华-sky博客园博客,原文链接:http://www.cnblogs.com/bonelee/p/6347274.html,如需转载请自行联系原作者