Elasticsearch: field "title" was indexed without position data; cannot run PhraseQuery

8
votes

I have an index in ElasticSearch with the following mapping:

mappings: {
    feed: {
        properties: {
            html_url: {
               index: not_analyzed
               omit_norms: true
               index_options: docs
               type: string
            }
            title: {
                index_options: offsets
                type: string
            }
            created: {
                store: true
                format: yyyy-MM-dd HH:mm:ss
                type: date
            }
            description: {
                type: string
            }
       }
}

getting the following error when performing phrase search ("video games"):

IllegalStateException[field \"title\" was indexed without position data; cannot run PhraseQuery (term=video)];

Single word searches work fine. Tried "index_options: positions" as well but with no luck. Title field contains text in multiple languages, sometimes empty. Interesting that it seems to fail randomly, for example it would fail with 200K documents or 800K using the same dataset. Is there a reason some titles wouldn't get indexed with positions?

Elastic search version 0.90.5

2
elasticsearch

2 Answers

4
votes

Just in case someone else has the same issue. There was another type/table (feed2) in the same index with the same "title" field that was set to "not_analyzed".

For some reason even if you specify the type: http://elasticsearchhost.com:9200/index_name/feed/_search the other type is still being searched as well. Changing the mapping for feed2 type fixed the problem.

2
votes

You probably have another field named 'title' with a different mapping in another type but in the same index.

Basically if you have 2 fields with the same name in the same index - even if they are in different types - they cannot have different mappings: to be more precise, even if they have the same type (eg: "string") but one of them is "analyzed" and the other is "not analyzed", problems will arise.

I mean, yeah, you can try to setup 2 different mappings, and ElasticSearch will not complain, but when searching you get strange result and everything will go bananas.

You can read more about this issue here where they say:

[...] In the end, we opted to enforce the rule that all fields with the same name in the same index must have the same mapping [...]

And yeah, considering how the promise of ElasticSearch has always been "it just works" this little detail took a lot of people by surprise.