Grok sits on top of regular expressions, so any regular expressions are valid in grok as well. The regular expression library is Oniguruma, and you can see the full supported regexp syntax on the Oniguruma site.

Grok works by leveraging this regular expression language to allow naming existing patterns and combining them into more complex patterns that match your fields.

The syntax for reusing a grok pattern comes in three forms: %{SYNTAX:SEMANTIC}, %{SYNTAX}, %{SYNTAX:SEMANTIC:TYPE}.

The SYNTAX is the name of the pattern that will match your text. For example, 3.44 will be matched by the NUMBER pattern and 55.3.244.1 will be matched by the IP pattern. The syntax is how you match. NUMBER and IP are both patterns that are provided within the default patterns set.

The SEMANTIC is the identifier you give to the piece of text being matched. For example, 3.44 could be the duration of an event, so you could call it simply duration. Further, a string 55.3.244.1 might identify the client making a request.

The TYPE is the type you wish to cast your named field. int, long, double, float and boolean are supported types for coercion.

For example, you might want to match the following text:

3.44 55.3.244.1

You may know that the message in the example is a number followed by an IP address. You can match this text by using the following Grok expression.

%{NUMBER:duration} %{IP:client}

Here is an example of using the provided patterns to extract out and name structured fields from a string field in a document.

POST _ingest/pipeline/_simulate
{
  "pipeline": {
    "description" : "...",
    "processors": [
      {
        "grok": {
          "field": "message",
          "patterns": ["%{IP:client} %{WORD:method} %{URIPATHPARAM:request} %{NUMBER:bytes:int} %{NUMBER:duration:double}"]
        }
      }
    ]
  },
  "docs":[
    {
      "_source": {
        "message": "55.3.244.1 GET /index.html 15824 0.043"
      }
    }
  ]
}

This pipeline will insert these named captures as new fields within the document, like so:

{
  "docs": [
    {
      "doc": {
        "_index": "_index",
        "_type": "_doc",
        "_id": "_id",
        "_source" : {
          "duration" : 0.043,
          "request" : "/index.html",
          "method" : "GET",
          "bytes" : 15824,
          "client" : "55.3.244.1",
          "message" : "55.3.244.1 GET /index.html 15824 0.043"
        },
        "_ingest": {
          "timestamp": "2016-11-08T19:43:03.850+0000"
        }
      }
    }
  ]
}

The Grok processor comes pre-packaged with a base set of patterns. These patterns may not always have what you are looking for. Patterns have a very basic format. Each entry has a name and the pattern itself.

You can add your own patterns to a processor definition under the pattern_definitions option. Here is an example of a pipeline specifying custom pattern definitions:

{
  "description" : "...",
  "processors": [
    {
      "grok": {
        "field": "message",
        "patterns": ["my %{FAVORITE_DOG:dog} is colored %{RGB:color}"],
        "pattern_definitions" : {
          "FAVORITE_DOG" : "beagle",
          "RGB" : "RED|GREEN|BLUE"
        }
      }
    }
  ]
}

Sometimes one pattern is not enough to capture the potential structure of a field. Let’s assume we want to match all messages that contain your favorite pet breeds of either cats or dogs. One way to accomplish this is to provide two distinct patterns that can be matched, instead of one really complicated expression capturing the same or behavior.

Here is an example of such a configuration executed against the simulate API:

POST _ingest/pipeline/_simulate
{
  "pipeline": {
  "description" : "parse multiple patterns",
  "processors": [
    {
      "grok": {
        "field": "message",
        "patterns": ["%{FAVORITE_DOG:pet}", "%{FAVORITE_CAT:pet}"],
        "pattern_definitions" : {
          "FAVORITE_DOG" : "beagle",
          "FAVORITE_CAT" : "burmese"
        }
      }
    }
  ]
},
"docs":[
  {
    "_source": {
      "message": "I love burmese cats!"
    }
  }
  ]
}

response:

{
  "docs": [
    {
      "doc": {
        "_type": "_doc",
        "_index": "_index",
        "_id": "_id",
        "_source": {
          "message": "I love burmese cats!",
          "pet": "burmese"
        },
        "_ingest": {
          "timestamp": "2016-11-08T19:43:03.850+0000"
        }
      }
    }
  ]
}

Both patterns will set the field pet with the appropriate match, but what if we want to trace which of our patterns matched and populated our fields? We can do this with the trace_match parameter. Here is the output of that same pipeline, but with "trace_match": true configured:

{
  "docs": [
    {
      "doc": {
        "_type": "_doc",
        "_index": "_index",
        "_id": "_id",
        "_source": {
          "message": "I love burmese cats!",
          "pet": "burmese"
        },
        "_ingest": {
          "_grok_match_index": "1",
          "timestamp": "2016-11-08T19:43:03.850+0000"
        }
      }
    }
  ]
}

In the above response, you can see that the index of the pattern that matched was "1". This is to say that it was the second (index starts at zero) pattern in patterns to match.

This trace metadata enables debugging which of the patterns matched. This information is stored in the ingest metadata and will not be indexed.

The Grok Processor comes packaged with its own REST endpoint for retrieving which patterns the processor is packaged with.

GET _ingest/processor/grok

The above request will return a response body containing a key-value representation of the built-in patterns dictionary.

{
  "patterns" : {
    "BACULA_CAPACITY" : "%{INT}{1,3}(,%{INT}{3})*",
    "PATH" : "(?:%{UNIXPATH}|%{WINPATH})",
    ...
}

This can be useful to reference as the built-in patterns change across versions.

Grok expressions that take too long to execute are interrupted and the grok processor then fails with an exception. The grok processor has a watchdog thread that determines when evaluation of a grok expression takes too long and is controlled by the following settings:

It is advised to use the Grok Debugger to debug grok patterns. From there you can test one or more patterns in the UI against sample data. Under the covers it uses the same engine as ingest node processor.

Additionally, it is recommended to enable debug logging for Grok so that any additional messages may also be seen in the Elasticsearch server log.

PUT _cluster/settings
{
  "transient": {
    "logger.org.elasticsearch.ingest.common.GrokProcessor": "debug"
  }
}