Percolator typeedit

The percolator field type parses a json structure into a native query and stores that query, so that the percolate query can use it to match provided documents.

Any field that contains a json object can be configured to be a percolator field. The percolator field type has no settings. Just configuring the percolator field type is sufficient to instruct Elasticsearch to treat a field as a query.

If the following mapping configures the percolator field type for the query field:

PUT my_index
{
    "mappings": {
        "doc": {
            "properties": {
                "query": {
                    "type": "percolator"
                },
                "field": {
                    "type": "text"
                }
            }
        }
    }
}

Then you can index a query:

PUT my_index/doc/match_value
{
    "query" : {
        "match" : {
            "field" : "value"
        }
    }
}
Important

Fields referred to in a percolator query must already exist in the mapping associated with the index used for percolation. In order to make sure these fields exist, add or update a mapping via the create index or put mapping APIs. Fields referred in a percolator query may exist in any type of the index containing the percolator field type.

Influencing query extractionedit

As part of indexing the percolator query, the percolator field mapper extracts the query terms and numeric ranges from the provided query and indexes that alongside the query in separate internal fields. The percolate query uses these internal fields to build a candidate query from the document being percolated in order to reduce the number of document that need to be verified.

In case a percolator query contains a bool query with must or filter clauses, then the percolator field mapper only has to extract ranges or terms from a single clause. The percolator field mapper will prefer longer terms over shorter terms, because longer terms in general match with less documents. For the same reason it prefers smaller ranges over bigger ranges.

In general this behaviour works well. However sometimes there are fields in a bool query that shouldn’t be taken into account when selecting the best must or filter clause, or fields are known to be more selective than other fields.

For example a status like field may in fact not work well, because each status matches with many percolator queries and then the candidate query the percolate query generates may not be able to filter out that many percolator queries.

The percolator field mapping allows to configure boost_fields in order to indicate to the percolator what fields are important or not important when selecting the best must or filter clause in a bool query:

PUT another_index
{
    "mappings": {
        "doc": {
            "properties": {
                "query": {
                    "type": "percolator",
                    "boost_fields": {
                        "status_field": 0, 
                        "price_field": 2 
                    }
                },
                "status_field": {
                    "type": "keyword"
                },
                "price_field": {
                    "type": "long"
                },
                "field": {
                    "type": "text"
                }
            }
        }
    }
}

A boost of zero hints to the percolator that if there are other clauses in a conjunction query then these should be preferred over this one.

Any boost higher than 1 overrides the default behaviour when it comes to selecting the best clause. The clause that has the field with the highest boost will be selected from a conjunction query for extraction.

The steps the percolator field mapper takes when selecting a clause from a conjunction query:

  • If there are clauses that have boosted fields then the clause with highest boost field is selected.
  • If there are range based clauses and term based clauses then term based clauses are picked over range based clauses
  • From all term based clauses the clause with longest term is picked.
  • In the case when there are only range based clauses then the range clause with smallest range is picked over clauses with wider ranges.

Reindexing your percolator queriesedit

Reindexing percolator queries is sometimes required to benefit from improvements made to the percolator field type in new releases.

Reindexing percolator queries can be reindexed by using the reindex api. Lets take a look at the following index with a percolator field type:

PUT index
{
  "mappings": {
    "doc" : {
      "properties": {
        "query" : {
          "type" : "percolator"
        },
        "body" : {
          "type": "text"
        }
      }
    }
  }
}

POST _aliases
{
  "actions": [
    {
      "add": {
        "index": "index",
        "alias": "queries" 
      }
    }
  ]
}

PUT queries/doc/1?refresh
{
  "query" : {
    "match" : {
      "body" : "quick brown fox"
    }
  }
}

It is always recommended to define an alias for your index, so that in case of a reindex systems / applications don’t need to be changed to know that the percolator queries are now in a different index.

Lets say you’re going to upgrade to a new major version and in order for the new Elasticsearch version to still be able to read your queries you need to reindex your queries into a new index on the current Elasticsearch version:

PUT new_index
{
  "mappings": {
    "doc" : {
      "properties": {
        "query" : {
          "type" : "percolator"
        },
        "body" : {
          "type": "text"
        }
      }
    }
  }
}

POST /_reindex?refresh
{
  "source": {
    "index": "index"
  },
  "dest": {
    "index": "new_index"
  }
}

POST _aliases
{
  "actions": [ 
    {
      "remove": {
        "index" : "index",
        "alias": "queries"
      }
    },
    {
      "add": {
        "index": "new_index",
        "alias": "queries"
      }
    }
  ]
}

If you have an alias don’t forget to point it to the new index.

Executing the percolate query via the queries alias:

GET /queries/_search
{
  "query": {
    "percolate" : {
      "field" : "query",
      "document" : {
        "body" : "fox jumps over the lazy dog"
      }
    }
  }
}

now returns matches from the new index:

{
  "took": 3,
  "timed_out": false,
  "_shards": {
    "total": 5,
    "successful": 5,
    "skipped" : 0,
    "failed": 0
  },
  "hits": {
    "total": 1,
    "max_score": 0.2876821,
    "hits": [
      {
        "_index": "new_index", 
        "_type": "doc",
        "_id": "1",
        "_score": 0.2876821,
        "_source": {
          "query": {
            "match": {
              "body": "quick brown fox"
            }
          }
        },
        "fields" : {
          "_percolator_document_slot" : [0]
        }
      }
    ]
  }
}

Percolator query hit is now being presented from the new index.

Optimizing query time text analysisedit

When the percolator verifies a percolator candidate match it is going to parse, perform query time text analysis and actually run the percolator query on the document being percolated. This is done for each candidate match and every time the percolate query executes. If your query time text analysis is relatively expensive part of query parsing then text analysis can become the dominating factor time is being spent on when percolating. This query parsing overhead can become noticeable when the percolator ends up verifying many candidate percolator query matches.

To avoid the most expensive part of text analysis at percolate time. One can choose to do the expensive part of text analysis when indexing the percolator query. This requires using two different analyzers. The first analyzer actually performs text analysis that needs be performed (expensive part). The second analyzer (usually whitespace) just splits the generated tokens that the first analyzer has produced. Then before indexing a percolator query, the analyze api should be used to analyze the query text with the more expensive analyzer. The result of the analyze api, the tokens, should be used to substitute the original query text in the percolator query. It is important that the query should now be configured to override the analyzer from the mapping and just the second analyzer. Most text based queries support an analyzer option (match, query_string, simple_query_string). Using this approach the expensive text analysis is performed once instead of many times.

Lets demonstrate this workflow via a simplified example.

Lets say we want to index the following percolator query:

{
  "query" : {
    "match" : {
      "body" : {
        "query" : "missing bicycles"
      }
    }
  }
}

with these settings and mapping:

PUT /test_index
{
  "settings": {
    "analysis": {
      "analyzer": {
        "my_analyzer" : {
          "tokenizer": "standard",
          "filter" : ["lowercase", "porter_stem"]
        }
      }
    }
  },
  "mappings": {
    "doc" : {
      "properties": {
        "query" : {
          "type": "percolator"
        },
        "body" : {
          "type": "text",
          "analyzer": "my_analyzer" 
        }
      }
    }
  }
}

For the purpose of this example, this analyzer is considered expensive.

First we need to use the analyze api to perform the text analysis prior to indexing:

POST /test_index/_analyze
{
  "analyzer" : "my_analyzer",
  "text" : "missing bicycles"
}

This results the following response:

{
  "tokens": [
    {
      "token": "miss",
      "start_offset": 0,
      "end_offset": 7,
      "type": "<ALPHANUM>",
      "position": 0
    },
    {
      "token": "bicycl",
      "start_offset": 8,
      "end_offset": 16,
      "type": "<ALPHANUM>",
      "position": 1
    }
  ]
}

All the tokens in the returned order need to replace the query text in the percolator query:

PUT /test_index/doc/1?refresh
{
  "query" : {
    "match" : {
      "body" : {
        "query" : "miss bicycl",
        "analyzer" : "whitespace" 
      }
    }
  }
}

It is important to select a whitespace analyzer here, otherwise the analyzer defined in the mapping will be used, which defeats the point of using this workflow. Note that whitespace is a built-in analyzer, if a different analyzer needs to be used, it needs to be configured first in the index’s settings.

The analyze api prior to the indexing the percolator flow should be done for each percolator query.

At percolate time nothing changes and the percolate query can be defined normally:

GET /test_index/_search
{
  "query": {
    "percolate" : {
      "field" : "query",
      "document" : {
        "body" : "Bycicles are missing"
      }
    }
  }
}

This results in a response like this:

{
  "took": 6,
  "timed_out": false,
  "_shards": {
    "total": 5,
    "successful": 5,
    "skipped" : 0,
    "failed": 0
  },
  "hits": {
    "total": 1,
    "max_score": 0.2876821,
    "hits": [
      {
        "_index": "test_index",
        "_type": "doc",
        "_id": "1",
        "_score": 0.2876821,
        "_source": {
          "query": {
            "match": {
              "body": {
                "query": "miss bicycl",
                "analyzer": "whitespace"
              }
            }
          }
        },
        "fields" : {
          "_percolator_document_slot" : [0]
        }
      }
    ]
  }
}

Dedicated Percolator Indexedit

Percolate queries can be added to any index. Instead of adding percolate queries to the index the data resides in, these queries can also be added to a dedicated index. The advantage of this is that this dedicated percolator index can have its own index settings (For example the number of primary and replica shards). If you choose to have a dedicated percolate index, you need to make sure that the mappings from the normal index are also available on the percolate index. Otherwise percolate queries can be parsed incorrectly.

Forcing Unmapped Fields to be Handled as Stringsedit

In certain cases it is unknown what kind of percolator queries do get registered, and if no field mapping exists for fields that are referred by percolator queries then adding a percolator query fails. This means the mapping needs to be updated to have the field with the appropriate settings, and then the percolator query can be added. But sometimes it is sufficient if all unmapped fields are handled as if these were default text fields. In those cases one can configure the index.percolator.map_unmapped_fields_as_text setting to true (default to false) and then if a field referred in a percolator query does not exist, it will be handled as a default text field so that adding the percolator query doesn’t fail.

Limitationsedit

Parent/childedit

Because the percolate query is processing one document at a time, it doesn’t support queries and filters that run against child documents such as has_child and has_parent.

Fetching queriesedit

There are a number of queries that fetch data via a get call during query parsing. For example the terms query when using terms lookup, template query when using indexed scripts and geo_shape when using pre-indexed shapes. When these queries are indexed by the percolator field type then the get call is executed once. So each time the percolator query evaluates these queries, the fetches terms, shapes etc. as the were upon index time will be used. Important to note is that fetching of terms that these queries do, happens both each time the percolator query gets indexed on both primary and replica shards, so the terms that are actually indexed can be different between shard copies, if the source index changed while indexing.

Script queryedit

The script inside a script query can only access doc values fields. The percolate query indexes the provided document into an in-memory index. This in-memory index doesn’t support stored fields and because of that the _source field and other stored fields are not stored. This is the reason why in the script query the _source and other stored fields aren’t available.