Not load the ids of child documents into memory.

Closes #3028
elastic · May 14, 2013 · eabd876 · eabd876
1 parent abe1707
commit eabd876
Show file tree

Hide file tree

Showing 3 changed files with 21 additions and 4 deletions.
diff --git a/src/main/java/org/elasticsearch/index/cache/id/IdCache.java b/src/main/java/org/elasticsearch/index/cache/id/IdCache.java
@@ -29,7 +29,8 @@
 import java.util.List;
 
 /**
- *
+ * This id cache contains only the ids of parent documents, loaded via the uid or parent field.
+ * This name IdCache is misleading, parentIdCache would be a better name.
  */
 public interface IdCache extends IndexComponent, CloseableComponent, Iterable<IdReaderCache> {
 

diff --git a/src/main/java/org/elasticsearch/index/cache/id/simple/SimpleIdCache.java b/src/main/java/org/elasticsearch/index/cache/id/simple/SimpleIdCache.java
@@ -116,6 +116,15 @@ public void refresh(List<AtomicReaderContext> atomicReaderContexts) throws Excep
                 Map<Object, Map<String, TypeBuilder>> builders = new HashMap<Object, Map<String, TypeBuilder>>();
                 Map<Object, IndexReader> cacheToReader = new HashMap<Object, IndexReader>();
 
+                // We don't want to load uid of child documents, this allows us to not load uids of child types.
+                Set<HashedBytesArray> parentTypes = new HashSet<HashedBytesArray>();
+                for (String type : indexService.mapperService().types()) {
+                    ParentFieldMapper parentFieldMapper = indexService.mapperService().documentMapper(type).parentFieldMapper();
+                    if (parentFieldMapper != null) {
+                        parentTypes.add(new HashedBytesArray(parentFieldMapper.type()));
+                    }
+                }
+
                 // first, go over and load all the id->doc map for all types
                 for (AtomicReaderContext context : atomicReaderContexts) {
                     AtomicReader reader = context.reader();
@@ -138,10 +147,16 @@ public void refresh(List<AtomicReaderContext> atomicReaderContexts) throws Excep
                         DocsEnum docsEnum = null;
                         for (BytesRef term = termsEnum.next(); term != null; term = termsEnum.next()) {
                             HashedBytesArray[] typeAndId = Uid.splitUidIntoTypeAndId(term);
-                            TypeBuilder typeBuilder = readerBuilder.get(typeAndId[0].toUtf8());
+                            // TODO: seek!
+                            if (!parentTypes.contains(typeAndId[0])) {
+                                continue;
+                            }
+
+                            String type = typeAndId[0].toUtf8();
+                            TypeBuilder typeBuilder = readerBuilder.get(type);
                             if (typeBuilder == null) {
                                 typeBuilder = new TypeBuilder(reader);
-                                readerBuilder.put(typeAndId[0].toUtf8(), typeBuilder);
+                                readerBuilder.put(type, typeBuilder);
                             }
 
                             HashedBytesArray idAsBytes = checkIfCanReuse(builders, typeAndId[1]);

diff --git a/src/test/java/org/elasticsearch/benchmark/search/child/ChildSearchBenchmark.java b/src/test/java/org/elasticsearch/benchmark/search/child/ChildSearchBenchmark.java
@@ -69,7 +69,7 @@ public static void main(String[] args) throws Exception {
         int QUERY_COUNT = 50;
         String indexName = "test";
 
-        Thread.sleep(10000);
+        client.admin().cluster().prepareHealth(indexName).setWaitForGreenStatus().setTimeout("10s").execute().actionGet();
         try {
             client.admin().indices().create(createIndexRequest(indexName)).actionGet();
             client.admin().indices().preparePutMapping(indexName).setType("child").setSource(XContentFactory.jsonBuilder().startObject().startObject("type")
@@ -366,6 +366,7 @@ public static void main(String[] args) throws Exception {
         System.out.println("--> has_parent query with match_all Query Avg: " + (totalQueryTime / QUERY_COUNT) + "ms");
 
 
+        System.gc();
         statsResponse = client.admin().cluster().prepareNodesStats()
                 .setJvm(true).setIndices(true).execute().actionGet();