Support autocomplete indexes on select fields

[Downchuck]

The LinkedIn Cleo project provides type-ahead indexes on set fields, a very handy feature for certain circumstances. This has come up for me in a project where one of the fields is essentially a dynamic enumeration of about five thousand entries; names of people and companies and the like. The Cleo project is available under Apache 2.0, should ES authors see fit to borrow any of its components to fulfill this feature request. ES already has its own Bloom implementation.

An auto-complete search request would allow me to say field X must equal "Z" and field "Y" is the field which I'm running the auto-complete request on.

Example record: { assigned_rep: "Bob", company: "Tractor mowing", company_tokens: ["tractor","mowing","lawn"] }.

My search index would be on assigned_rep and company_tokens. I would pass assigned_rep = Bob and autocomplete_search on company_tokens for "Law", which would match on "lawn" and return the full record (and any other matches).

[Downchuck]

This has been brought up before on mailing lists. The best post I found was Cleo or ElasticSearch with NGram. With the new patch from issue #2674, this feature is very close to supportable. It's likely that a few borrowings from the Cleo codebase would finish the feature.

[Downchuck]

Also see, LinkedIn Blog post about Cleo typeahead search

[Downchuck]

Some of the API for issue #2709 may be relevant.

[Downchuck]

It seems like performance issues partly addressed by issue #988 improvements in Lucene. Simply, bool.must.wildcard on a lowercase filtered + camel case analyzed field works quite well over a medium sized set of documents.

[kimchy]

@Downchuck another option is to simply use edge ngrams analysis on the relevant field you wish to do autocomplete on.

[Downchuck]

ES has this thoroughly handled, closing the bug.