IMPORTANT: No additional bug fixes or documentation updates will be released for this version. For the latest information, see the current release documentation.
Elasticsearch Guide [7.7] » Roll up or transform your data » Transforming data » Painless examples for transforms
« Transform examples Troubleshooting transforms »

Painless examples for transformsedit

These examples demonstrate how to use Painless in transforms. You can learn more about the Painless scripting language in the Painless guide.

While the context of the following examples is the transform use case, the Painless scripts in the snippets below can be used in other Elasticsearch search aggregations, too.

Getting top hits by using scripted metric aggregationedit

This snippet shows how to find the latest document, in other words the document with the earliest timestamp. From a technical perspective, it helps to achieve the function of a Top Hits Aggregation by using scripted metric aggregation in a transform, which provides a metric output.

"aggregations": {
  "latest_doc": {
    "scripted_metric": {
      "init_script": "state.timestamp_latest = 0L; state.last_doc = ''", 
      "map_script": """ 
        def current_date = doc['@timestamp'].getValue().toInstant().toEpochMilli();
        if (current_date > state.timestamp_latest)
        {state.timestamp_latest = current_date;
        state.last_doc = new HashMap(params['_source']);}
      """,
      "combine_script": "return state", 
      "reduce_script": """ 
        def last_doc = '';
        def timestamp_latest = 0L;
        for (s in states) {if (s.timestamp_latest > (timestamp_latest))
        {timestamp_latest = s.timestamp_latest; last_doc = s.last_doc;}}
        return last_doc
      """
    }
  }
}

The init_script creates a long type timestamp_latest and a string type last_doc in the state object.

The map_script defines current_date based on the timestamp of the document, then compares current_date with state.timestamp_latest, finally returns state.last_doc from the shard. By using new HashMap(...) you copy the source document, this is important whenever you want to pass the full source object from one phase to the next.

The combine_script returns state from each shard.

The reduce_script iterates through the value of s.timestamp_latest returned by each shard and returns the document with the latest timestamp (last_doc). In the response, the top hit (in other words, the latest_doc) is nested below the latest_doc field.

Check the scope of scripts for detailed explanation on the respective scripts.

You can retrieve the last value in a similar way:

"aggregations": {
  "latest_value": {
    "scripted_metric": {
      "init_script": "state.timestamp_latest = 0L; state.last_value = ''",
      "map_script": """
        def current_date = doc['date'].getValue().toInstant().toEpochMilli();
        if (current_date > state.timestamp_latest)
        {state.timestamp_latest = current_date;
        state.last_value = params['_source']['value'];}
      """,
      "combine_script": "return state",
      "reduce_script": """
        def last_value = '';
        def timestamp_latest = 0L;
        for (s in states) {if (s.timestamp_latest > (timestamp_latest))
        {timestamp_latest = s.timestamp_latest; last_value = s.last_value;}}
        return last_value
      """
    }
  }
}

Getting time features by using aggregationsedit

This snippet shows how to extract time based features by using Painless in a transform. The snippet uses an index where @timestamp is defined as a date type field.

"aggregations": {
  "avg_hour_of_day": { 
    "avg":{
      "script": { 
        "source": """
          ZonedDateTime date =  doc['@timestamp'].value; 
          return date.getHour(); 
        """
      }
    }
  },
  "avg_month_of_year": { 
    "avg":{
      "script": { 
        "source": """
          ZonedDateTime date =  doc['@timestamp'].value; 
          return date.getMonthValue(); 
        """
      }
    }
  },
 ...
}

Name of the aggregation.

Contains the Painless script that returns the hour of the day.

Sets date based on the timestamp of the document.

Returns the hour value from date.

Name of the aggregation.

Contains the Painless script that returns the month of the year.

Sets date based on the timestamp of the document.

Returns the month value from date.

Using Painless in group_byedit

It is possible to base the group_by property of a transform on the output of a script. The following example uses the Kibana sample web logs dataset. The goal here is to make the transform output easier to understand through normalizing the value of the fields that the data is grouped by.

If you use scripts in group_by, the transform cannot deduce the mappings of the field. For this reason, you need to create the destination index and its mappings prior to creating the transform. Please refer to the example below.

Create the index:

PUT /pivot_logs

Create the mapping of the index:

PUT /pivot_logs/_mapping
{
  "properties": {
    "200": {"type": "long"},
    "404":  { "type": "long"},
    "503":  { "type": "long"},
    "agent":  { "type": "keyword"}
  }
}

Then you can create the transform. The example below shows you how to create a preview of the transform.

POST _transform/_preview
{
  "source": {
    "index": [ 
      "kibana_sample_data_logs"
    ]
  },
  "pivot": {
    "group_by": {
      "agent": {
        "terms": {
          "script": { 
            "source": """String agent = doc['agent.keyword'].value;
            if (agent.contains("MSIE")) {
              return "internet explorer";
            } else if (agent.contains("AppleWebKit")) {
              return "safari";
            } else if (agent.contains('Firefox')) {
              return "firefox";
            } else { return agent }""",
            "lang": "painless"
          }
        }
      }
    },
    "aggregations": { 
      "200": {
        "filter": {
          "term": {
            "response": "200"
          }
        }
      },
      "404": {
        "filter": {
          "term": {
            "response": "404"
          }
        }
      },
      "503": {
        "filter": {
          "term": {
            "response": "503"
          }
        }
      }
    }
  },
  "dest": { 
    "index": "pivot_logs"
  }
}

Specifies the source index or indices.

The script defines an agent string based on the agent field of the documents, then iterates through the values. If an agent field contains "MSIE", than the script returns "Internet Explorer". If it contains AppleWebKit, it returns "safari". It returns "firefox" if the field value contains "Firefox". Finally, in every other case, the value of the field is returned.

The aggregations object contains filters that narrow down the results to documents that contains 200, 404, or 503 values in the response field.

Specifies the destination index of the transform. As the mappings of fields created by scrips cannot be deduced, please define the mappings of the destination index prior to creating the transform.

The API returns the following result:

{
  "preview" : [
    {
      "agent" : "firefox",
      "200" : 4931,
      "404" : 259,
      "503" : 172
    },
    {
      "agent" : "internet explorer",
      "200" : 3674,
      "404" : 210,
      "503" : 126
    },
    {
      "agent" : "safari",
      "200" : 4227,
      "404" : 332,
      "503" : 143
    }
  ],
  "mappings" : {
    "properties" : {
      "200" : {
        "type" : "long"
      },
      "agent" : {
        "type" : "keyword"
      },
      "404" : {
        "type" : "long"
      },
      "503" : {
        "type" : "long"
      }
    }
  }
}

You can see that the agent values are simplified so it is easier to interpret them. The table below shows how normalization modifies the output of the transform in our example compared to the non-normalized values.

Non-normalized agent value Normalized agent value

"Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.1; SV1; .NET CLR 1.1.4322)"

"internet explorer"

"Mozilla/5.0 (X11; Linux i686) AppleWebKit/534.24 (KHTML, like Gecko) Chrome/11.0.696.50 Safari/534.24"

"safari"

"Mozilla/5.0 (X11; Linux x86_64; rv:6.0a1) Gecko/20110421 Firefox/6.0a1"

"firefox"

Getting duration by using bucket scriptedit

This example shows you how to get the duration of a session by client IP from a data log by using bucket script. The example uses the Kibana sample web logs dataset.

PUT _transform/data_log
{
  "source": {
    "index": "kibana_sample_data_logs"
  },
  "dest": {
    "index": "data-logs-by-client"
  },
  "pivot": {
    "group_by": {
      "machine.os": {"terms": {"field": "machine.os.keyword"}},
      "machine.ip": {"terms": {"field": "clientip"}}
    },
    "aggregations": {
      "time_frame.lte": {
        "max": {
          "field": "timestamp"
        }
      },
      "time_frame.gte": {
        "min": {
          "field": "timestamp"
        }
      },
      "time_length": { 
        "bucket_script": {
          "buckets_path": { 
            "min": "time_frame.gte.value",
            "max": "time_frame.lte.value"
          },
          "script": "params.max - params.min" 
        }
      }
    }
  }
}

To define the length of the sessions, we use a bucket script.

The bucket path is a map of script variables and their associated path to the buckets you want to use for the variable. In this particular case, min and max are variables mapped to time_frame.gte.value and time_frame.lte.value.

Finally, the script substracts the start date of the session from the end date which results in the duration of the session.

Counting HTTP responses by using scripted metric aggregationedit

You can count the different HTTP response types in a web log data set by using scripted metric aggregation as part of the transform. The example below assumes that the HTTP response codes are stored as keywords in the response field of the documents.

"aggregations": { 
  "responses.counts": { 
    "scripted_metric": { 
      "init_script": "state.responses = ['error':0L,'success':0L,'other':0L]", 
      "map_script": """ 
        def code = doc['response.keyword'].value;
        if (code.startsWith('5') || code.startsWith('4')) {
          state.responses.error += 1 ;
        } else if(code.startsWith('2')) {
          state.responses.success += 1;
        } else {
          state.responses.other += 1;
        }
        """,
      "combine_script": "state.responses", 
      "reduce_script": """ 
        def counts = ['error': 0L, 'success': 0L, 'other': 0L];
        for (responses in states) {
          counts.error += responses['error'];
          counts.success += responses['success'];
          counts.other += responses['other'];
        }
        return counts;
        """
      }
    },
  ...
}

The aggregations object of the transform that contains all aggregations.

Object of the scripted_metric aggregation.

This scripted_metric performs a distributed operation on the web log data to count specific types of HTTP responses (error, success, and other).

The init_script creates a responses array in the state object with three properties (error, success, other) with long data type.

The map_script defines code based on the response.keyword value of the document, then it counts the errors, successes, and other responses based on the first digit of the responses.

The combine_script returns state.responses from each shard.

The reduce_script creates a counts array with the error, success, and other properties, then iterates through the value of responses returned by each shard and assigns the different response types to the appropriate properties of the counts object; error responses to the error counts, success responses to the success counts, and other responses to the other counts. Finally, returns the counts array with the response counts.

Comparing indices by using scripted metric aggregationsedit

This example shows how to compare the content of two indices by a transform that uses a scripted metric aggregation.

POST _transform/_preview
{
  "id" : "index_compare",
  "source" : { 
    "index" : [
      "index1",
      "index2"
    ],
    "query" : {
      "match_all" : { }
    }
  },
  "dest" : { 
    "index" : "compare"
  },
  "pivot" : {
    "group_by" : {
      "unique-id" : {
        "terms" : {
          "field" : "<unique-id-field>" 
        }
      }
    },
    "aggregations" : {
      "compare" : { 
        "scripted_metric" : {
          "init_script" : "",
          "map_script" : "state.doc = new HashMap(params['_source'])", 
          "combine_script" : "return state", 
          "reduce_script" : """ 
            if (states.size() != 2) {
              return "count_mismatch"
            }
            if (states.get(0).equals(states.get(1))) {
              return "match"
            } else {
              return "mismatch"
            }
            """
        }
      }
    }
  }
}

The indices referenced in the source object are compared to each other.

The dest index contains the results of the comparison.

The group_by field needs to be a unique identifier for each document.

Object of the scripted_metric aggregation.

The map_script defines doc in the state object. By using new HashMap(...) you copy the source document, this is important whenever you want to pass the full source object from one phase to the next.

The combine_script returns state from each shard.

The reduce_script checks if the size of the indices are equal. If they are not equal, than it reports back a count_mismatch. Then it iterates through all the values of the two indices and compare them. If the values are equal, then it returns a match, otherwise returns a mismatch.

Getting web session details by using scripted metric aggregationedit

This example shows how to derive multiple features from a single transaction. Let’s take a look on the example source document from the data:

Source document 
{
  "_index":"apache-sessions",
  "_type":"_doc",
  "_id":"KvzSeGoB4bgw0KGbE3wP",
  "_score":1.0,
  "_source":{
    "@timestamp":1484053499256,
    "apache":{
      "access":{
        "sessionid":"571604f2b2b0c7b346dc685eeb0e2306774a63c2",
        "url":"http://www.leroymerlin.fr/v3/search/search.do?keyword=Carrelage%20salle%20de%20bain",
        "path":"/v3/search/search.do",
        "query":"keyword=Carrelage%20salle%20de%20bain",
        "referrer":"http://www.leroymerlin.fr/v3/p/produits/carrelage-parquet-sol-souple/carrelage-sol-et-mur/decor-listel-et-accessoires-carrelage-mural-l1308217717?resultOffset=0&resultLimit=51&resultListShape=MOSAIC&priceStyle=SALEUNIT_PRICE",
        "user_agent":{
          "original":"Mobile Safari 10.0 Mac OS X (iPad) Apple Inc.",
          "os_name":"Mac OS X (iPad)"
        },
        "remote_ip":"0337b1fa-5ed4-af81-9ef4-0ec53be0f45d",
        "geoip":{
          "country_iso_code":"FR",
          "location":{
            "lat":48.86,
            "lon":2.35
          }
        },
        "response_code":200,
        "method":"GET"
      }
    }
  }
}
...

By using the sessionid as a group-by field, you are able to enumerate events through the session and get more details of the session by using scripted metric aggregation.

POST _transform/_preview
{
  "source": {
    "index": "apache-sessions"
  },
  "pivot": {
    "group_by": {
      "sessionid": { 
        "terms": {
          "field": "apache.access.sessionid"
        }
      }
    },
    "aggregations": { 
      "distinct_paths": {
        "cardinality": {
          "field": "apache.access.path"
        }
      },
      "num_pages_viewed": {
        "value_count": {
          "field": "apache.access.url"
        }
      },
      "session_details": {
        "scripted_metric": {
          "init_script": "state.docs = []", 
          "map_script": """ 
            Map span = [
              '@timestamp':doc['@timestamp'].value,
              'url':doc['apache.access.url'].value,
              'referrer':doc['apache.access.referrer'].value
            ];
            state.docs.add(span)
          """,
          "combine_script": "return state.docs;", 
          "reduce_script": """ 
            def all_docs = [];
            for (s in states) {
              for (span in s) {
                all_docs.add(span);
              }
            }
            all_docs.sort((HashMap o1, HashMap o2)->o1['@timestamp'].millis.compareTo(o2['@timestamp'].millis));
            def size = all_docs.size();
            def min_time = all_docs[0]['@timestamp'];
            def max_time = all_docs[size-1]['@timestamp'];
            def duration = max_time.millis - min_time.millis;
            def entry_page = all_docs[0]['url'];
            def exit_path = all_docs[size-1]['url'];
            def first_referrer = all_docs[0]['referrer'];
            def ret = new HashMap();
            ret['first_time'] = min_time;
            ret['last_time'] = max_time;
            ret['duration'] = duration;
            ret['entry_page'] = entry_page;
            ret['exit_path'] = exit_path;
            ret['first_referrer'] = first_referrer;
            return ret;
          """
        }
      }
    }
  }
}

The data is grouped by sessionid.

The aggregations counts the number of paths and enumerate the viewed pages during the session.

The init_script creates an array type doc in the state object.

The map_script defines a span array with a timestamp, a URL, and a referrer value which are based on the corresponding values of the document, then adds the value of the span array to the doc object.

The combine_script returns state.docs from each shard.

The reduce_script defines various objects like min_time, max_time, and duration based on the document fields, then declares a ret object, and copies the source document by using new HashMap (). Next, the script defines first_time, last_time, duration and other fields inside the ret object based on the corresponding object defined earlier, finally returns ret.

The API call results in a similar response:

{
  "num_pages_viewed" : 2.0,
  "session_details" : {
    "duration" : 131374,
    "first_referrer" : "https://www.bing.com/",
    "entry_page" : "http://www.leroymerlin.fr/v3/p/produits/materiaux-menuiserie/porte-coulissante-porte-interieure-escalier-et-rambarde/barriere-de-securite-l1308218463",
    "first_time" : "2017-01-10T21:22:52.982Z",
    "last_time" : "2017-01-10T21:25:04.356Z",
    "exit_path" : "http://www.leroymerlin.fr/v3/p/produits/materiaux-menuiserie/porte-coulissante-porte-interieure-escalier-et-rambarde/barriere-de-securite-l1308218463?__result-wrapper?pageTemplate=Famille%2FMat%C3%A9riaux+et+menuiserie&resultOffset=0&resultLimit=50&resultListShape=PLAIN&nomenclatureId=17942&priceStyle=SALEUNIT_PRICE&fcr=1&*4294718806=4294718806&*14072=14072&*4294718593=4294718593&*17942=17942"
  },
  "distinct_paths" : 1.0,
  "sessionid" : "000046f8154a80fd89849369c984b8cc9d795814"
},
{
  "num_pages_viewed" : 10.0,
  "session_details" : {
    "duration" : 343112,
    "first_referrer" : "https://www.google.fr/",
    "entry_page" : "http://www.leroymerlin.fr/",
    "first_time" : "2017-01-10T16:57:39.937Z",
    "last_time" : "2017-01-10T17:03:23.049Z",
    "exit_path" : "http://www.leroymerlin.fr/v3/p/produits/porte-de-douche-coulissante-adena-e168578"
  },
  "distinct_paths" : 8.0,
  "sessionid" : "000087e825da1d87a332b8f15fa76116c7467da6"
}
...
« Transform examples Troubleshooting transforms »