Multireader Support in Searcher Manager

[Shibi-bala]

Description

Copied from #13975

I'd like to use MultiReader inside my searcher manager, but currently there is only support for DirectoryReader. Not sure about the context for. why this was the case initially, seems specific to some desired use-case https://issues.apache.org/jira/browse/LUCENE-6087.

Anyway should be a 1-line change here:

lucene/lucene/core/src/java/org/apache/lucene/search/SearcherManager.java

Line 125 in 9e90fb5

public SearcherManager(DirectoryReader reader, SearcherFactory searcherFactory)

I found a relevant stack overflow question here:
https://stackoverflow.com/questions/49817453/searchermanager-and-multireader-in-lucene

[mikemccand]

Thanks @Shibi-bala -- I agree it's odd it was scoped to just DirectoryReader -- any IndexReader should work as long as it can openIfChanged on itself.

I think English.java (from Lucene's test-framework) was maybe deleted long ago? Maybe simplify the test to not bother with English words... just Integer.toString(i) or "" + i should be fine?

Also, please revert the wildcard import (import org.apache.lucene.index.*) -- I think our style checker (jtidy/spotless) will be unhappy with that.

[vigyasharma]

any IndexReader should work as long as it can openIfChanged on itself.

Does MultiReader implement openIfChanged() ? I see a check in SearcherManager#refreshIfNeeded() that asserts for the reader to be a DirectoryReader instance. This is used by the ReferenceManager base class whenever maybeRefresh() is called.

[jpountz]

To add to @vigyasharma, I have been wondering if we should remove SearcherManager and encourage users to use IndexReaderManager. IndexSearcher is cheap to create and there are som reasons why it may be interesting to create a different IndexSearcher per query already, e.g. to tune parallelism based on current load, or to configure a timeout on the IndexSearcher.

Maybe MultiReader falls in the same bucket. Presumably, the MultiReader is made from multiple DirectoryReaders, so maybe the application code should not create a SearcherManager that works with a MultiReader but instead manage two (or more, one per Directory) ReaderManagers and dynamically create a MultiReader and then an IndexSearcher on top of this MultiReader on every search?