IndexSearcher instance efficiency

[vladak]

When working on PR #4008, I noticed that the MultiReader wrapping basically ignores the IndexSearcher objects, even for single project searches. Basically RuntimeEnvironment keeps the IndexSearcher objects in a map that is then used solely to get IndexReader objects (out of IndexSearcher objects) in order to create new IndexSearcher instance, outside of ThreadpoolSearcherFactory.

Also, project-less setup might want to leverage the SearcherManager approach too - to avoid opening the index directory (supposedly expensive operation) each time a search is performed:

opengrok/opengrok-indexer/src/main/java/org/opengrok/indexer/web/SearchHelper.java

Lines 383 to 389 in c3b0c92

	if (projects.isEmpty()) {
	// no project setup
	FSDirectory dir = FSDirectory.open(indexDir.toPath());
	reader = DirectoryReader.open(dir);
	searcher = new IndexSearcher(reader);
	closeOnDestroy = true;
	} else {

[vladak]

Also, DirectoryHistoryReader should get the IndexReader objects from the SearcherManagers in RuntimeEnvironment and avoid IndexDatabase#getIndexReader(). Actually, it is IndexDatabase#getIndexReader() that should be replaced. This is an area (IndexDatabase#getDocument()) where IndexSearcher from the SearcherManager should be used (via RuntimeEnvironment.getInstance().getIndexSearcher(String projectName)), i.e. avoid creating new one. IndexDatabase#getDocument() is actually used pretty heavily in the webapp (to get the most recent revision of given document in order to construct URL for the redirect) so this might provide some speedup.