[SOLR-2801] IndexNotFoundException thrown when the index directory has no segments file

• Type: Bug
• Status: Closed
• Priority: Trivial
• Resolution: Not A Problem
• Affects Version/s: 3.4
• Component/s: search
• Assignee: Unassigned
• Reporter: Ruben Inoto
• Created: 29/Sep/11 12:02
• Updated: 03/Oct/11 18:10
• Resolved: 03/Oct/11 18:10

Description

If the directory where the index is stored gets somehow corrupted, and it becomes empty or the segments files are deleted, any attempt to access the Solr server (via /select, /update, etc..) will throw the following exception:

org.apache.lucene.index.IndexNotFoundException: no segments* file found in org.apache.lucene.store.NIOFSDirectory@/index_dir

The only workaround we have found is to stop the server, remove the /index directory, and start again.

We have found more useful to create a new implementation of the IndexReaderFactory (that extends StandardIndexReaderFactory), that, in case of a IndexNotFoundException, it tries to "fix" the index directory by opening an IndexWriter on it and doing a commit:

    @Override
    public IndexReader newReader(Directory indexDir, boolean readOnly) throws IOException {
        try {
            return super.newReader(indexDir, readOnly);
        } catch (IndexNotFoundException e) {
            logger.warn(
                    "Warning: Trying to get a new reader threw an exception. Trying to create a writer first, and then get the reader",
                    e);
            IndexWriterConfig config = new IndexWriterConfig(Version.LUCENE_34, new WhitespaceAnalyzer(
                    Version.LUCENE_34));
            config.setOpenMode(OpenMode.CREATE_OR_APPEND);
            IndexWriter writer = new IndexWriter(indexDir, config);
            writer.commit();
            writer.close();
            try {
                return super.newReader(indexDir, readOnly);
            } catch (IndexNotFoundException e2) {
                logger.error(
                        "Trying to commit in the writer didn't work, as the reader is still throwing an exception :(. Re-throwing exception",
                        e2);
                throw e2;
            }
        }
    }

Would it make sense to add something like this to the StandardIndexReaderFactory?