Number format Exception For string type

Elastic Stack Elasticsearch
6

Hi,

did you create the new index before running the reindex? Please follow the example below. This worked fine for me on Elasticsearch 2.4.5 (as your talking about "string" I assumed you are on 2.x; in 5.x "string" type has been split into "text" and "keyword").

Let's start fresh and delete all involved indices for this example:

DELETE /old_books
DELETE /new_books

Then create the old index and add a document to it.

PUT /old_books
{
   "mappings": {
      "author": {
         "properties": {
            "name": {
               "type": "string"
            },
            "auid": {
                "type": "string"
            }
         }
      }
   }
}

PUT /old_books/author/1
{
    "name": "Mark Twain",
    "auid": "42"
}

Now we realize we made a mistake and create the new index where we correct the mapping of the auid field to integer:

PUT /new_books
{
   "mappings": {
      "author": {
         "properties": {
            "name": {
               "type": "string"
            },
            "auid": {
                "type": "integer"
            }
         }
      }
   }
}

and let the reindex API do its job:

POST /_reindex
{
  "source": {
    "index": "old_books"
  },
  "dest": {
    "index": "new_books"
  }
}

This worked fine for me. Here is the response:

{
   "took": 56,
   "timed_out": false,
   "total": 1,
   "updated": 0,
   "created": 1,
   "batches": 1,
   "version_conflicts": 0,
   "noops": 0,
   "retries": 0,
   "throttled_millis": 0,
   "requests_per_second": "unlimited",
   "throttled_until_millis": 0,
   "failures": []
}

Note, although Elasticsearch indexes the field now as integer, it is still a string in _source:

GET /new_books/author/_search
{
    "query": {
        "match_all": {}
    }
}

shows:

{
  ...
      "hits": [
         {
            "_index": "new_books",
            "_type": "author",
            "_id": "1",
            "_score": 1,
            "_source": {
               "name": "Mark Twain",
               "auid": "42"
            }
         }
      ]
...

If you do not want that you can use an inline script to convert the source as well during reindexing. Remember that inline scripts are a security risk and that you should rather use stored scripts for this. If you still want to use inline scripts then you need to (temporarily) enable them by setting script.inline: true in config/elasticsearch.yml and restart your cluster:

POST /_reindex
{
  "source": {
    "index": "old_books"
  },
  "dest": {
    "index": "new_books"
  },
"script": {
    "inline": "ctx._source.auid = Integer.valueOf(ctx._source.auid)"
  }
}

This will significantly slow-down the reindexing process though. Also check how you can reindex without downtime.

Daniel