TermsFacet across index returns wrong results (0.90)

[synhershko]

When faceting on more than one index, each with one shard, no replicas, and containing the same data structure, TermsFacet will return the wrong counts. This can be easily verified by running a simple query with the facet value. Additionally, changing the size parameter will cause facet results to change (while we only expected to get a shorter/longer list , we actually see values change and terms disappear).

Our faceting query:

{
  "from" : 0,
  "size" : 0,
  "query" : {
    "filtered" : {
      "query" : {
        "query_string" : {
          "query" : "test",
          "fields" : [ "title", "topic" ]
        }
      },
      "filter" : {
        "numeric_range" : {
          "topic_date" : {
            "from" : "2012-12-17T00:00:00.000+02:00",
            "to" : "2013-03-17T23:59:59.999+02:00",
            "include_lower" : true,
            "include_upper" : true
          }
        }
      }
    }
  },
  "facets" : {
    "termFacet" : {
      "terms" : {
        "fields" : [ "author" ],
        "size" : 10
      },
      "facet_filter" : {
        "numeric_range" : {
          "topic_date" : {
            "from" : "2012-12-17T00:00:00.000+02:00",
            "to" : "2013-03-17T23:59:59.999+02:00",
            "include_lower" : true,
            "include_upper" : true
          }
        }
      }
    }
  }
}

[dadoonet]

IMHO, that's somehow the same issue as #1305, isn't it?

[synhershko]

I'm familiar with that issue, but people are reporting there single shards of multiple indexes works for them, and use that approach as a workaround. If this indeed is the same issue, this becomes a regression issue.

Either way, we would really like to have this fixed...

[dadoonet]

What people are reporting as far as I understand is that when your narrow your search on a single shard, you will will have the exact values. But I can be wrong here.

[clintongormley]

This is indeed a duplicate of #1305 and there is no easy fix, just workarounds:

1. use a single shard, or
2. request more terms than you actually need

It is the nature of distributed search...