Painless Scripting Language

edit

Painless Scripting Language

edit

The Painless scripting language is new and is still marked as experimental. The syntax or API may be changed in the future in non-backwards compatible ways if required.

Painless is a simple, secure scripting language available in Elasticsearch by default. It is designed specifically for use with Elasticsearch and can safely be used with inline and stored scripting, which is enabled by default.

The Painless syntax is similar to Groovy.

You can use Painless anywhere a script can be used in Elasticsearch. It is the default if you don’t set the lang parameter but if you want to be explicit you can set the lang parameter to painless.

Painless Features

edit
  • Fast performance: several times faster than the alternatives.
  • Safety: Fine-grained whitelist with method call/field granularity. See Appendix A, Painless API Reference for a complete list of available classes and methods.
  • Optional typing: Variables and parameters can use explicit types or the dynamic def type.
  • Syntax: Extends Java’s syntax with a subset of Groovy for ease of use. See the Syntax Overview.
  • Optimizations: Designed specifically for Elasticsearch scripting.

Painless Examples

edit

To illustrate how Painless works, let’s load some hockey stats into an Elasticsearch index:

PUT hockey/player/_bulk?refresh
{"index":{"_id":1}}
{"first":"johnny","last":"gaudreau","goals":[9,27,1],"assists":[17,46,0],"gp":[26,82,1],"born":"1993/08/13"}
{"index":{"_id":2}}
{"first":"sean","last":"monohan","goals":[7,54,26],"assists":[11,26,13],"gp":[26,82,82],"born":"1994/10/12"}
{"index":{"_id":3}}
{"first":"jiri","last":"hudler","goals":[5,34,36],"assists":[11,62,42],"gp":[24,80,79],"born":"1984/01/04"}
{"index":{"_id":4}}
{"first":"micheal","last":"frolik","goals":[4,6,15],"assists":[8,23,15],"gp":[26,82,82],"born":"1988/02/17"}
{"index":{"_id":5}}
{"first":"sam","last":"bennett","goals":[5,0,0],"assists":[8,1,0],"gp":[26,1,0],"born":"1996/06/20"}
{"index":{"_id":6}}
{"first":"dennis","last":"wideman","goals":[0,26,15],"assists":[11,30,24],"gp":[26,81,82],"born":"1983/03/20"}
{"index":{"_id":7}}
{"first":"david","last":"jones","goals":[7,19,5],"assists":[3,17,4],"gp":[26,45,34],"born":"1984/08/10"}
{"index":{"_id":8}}
{"first":"tj","last":"brodie","goals":[2,14,7],"assists":[8,42,30],"gp":[26,82,82],"born":"1990/06/07"}
{"index":{"_id":39}}
{"first":"mark","last":"giordano","goals":[6,30,15],"assists":[3,30,24],"gp":[26,60,63],"born":"1983/10/03"}
{"index":{"_id":10}}
{"first":"mikael","last":"backlund","goals":[3,15,13],"assists":[6,24,18],"gp":[26,82,82],"born":"1989/03/17"}
{"index":{"_id":11}}
{"first":"joe","last":"colborne","goals":[3,18,13],"assists":[6,20,24],"gp":[26,67,82],"born":"1990/01/30"}

Accessing Doc Values from Painless

edit

Document values can be accessed from a Map named doc.

For example, the following script calculates a player’s total goals. This example uses a strongly typed int and a for loop.

GET hockey/_search
{
  "query": {
    "function_score": {
      "script_score": {
        "script": {
          "lang": "painless",
          "inline": "int total = 0; for (int i = 0; i < doc['goals'].length; ++i) { total += doc['goals'][i]; } return total;"
        }
      }
    }
  }
}

Alternatively, you could do the same thing using a script field instead of a function score:

GET hockey/_search
{
  "query": {
    "match_all": {}
  },
  "script_fields": {
    "total_goals": {
      "script": {
        "lang": "painless",
        "inline": "int total = 0; for (int i = 0; i < doc['goals'].length; ++i) { total += doc['goals'][i]; } return total;"
      }
    }
  }
}

The following example uses a Painless script to sort the players by their combined first and last names. The names are accessed using doc['first'].value and doc['last'].value.

GET hockey/_search
{
  "query": {
    "match_all": {}
  },
  "sort": {
    "_script": {
      "type": "string",
      "order": "asc",
      "script": {
        "lang": "painless",
        "inline": "doc['first.keyword'].value + ' ' + doc['last.keyword'].value"
      }
    }
  }
}

Updating Fields with Painless

edit

You can also easily update fields. You access the original source for a field as ctx._source.<field-name>.

First, let’s look at the source data for a player by submitting the following request:

GET hockey/_search
{
  "stored_fields": [
    "_id",
    "_source"
  ],
  "query": {
    "term": {
      "_id": 1
    }
  }
}

To change player 1’s last name to hockey, simply set ctx._source.last to the new value:

POST hockey/player/1/_update
{
  "script": {
    "lang": "painless",
    "inline": "ctx._source.last = params.last",
    "params": {
      "last": "hockey"
    }
  }
}

You can also add fields to a document. For example, this script adds a new field that contains the player’s nickname, hockey.

POST hockey/player/1/_update
{
  "script": {
    "lang": "painless",
    "inline": "ctx._source.last = params.last; ctx._source.nick = params.nick",
    "params": {
      "last": "gaudreau",
      "nick": "hockey"
    }
  }
}

Dates

edit

Dates are a little different to work with than regular values. Here is an example returning the year of every player’s birth:

GET hockey/_search
{
  "script_fields": {
    "birth_year": {
      "script": {
        "inline": "doc.born.date.year"
      }
    }
  }
}

The key here is that instead of indexing directly into doc.born like you would a normal field you have to call doc.born.date to get a ReadableDateTime. From there you can call methods like getYear and getDayOfWeek. In the example above year is a shortcut to getYear().

If the date field is a list then date will always return the first date. To access all the dates use dates instead of date.

Regular expressions

edit

Regexes are disabled by default because they circumvent Painless’s protection against long running and memory hungry scripts. To make matters worse even innocuous looking regexes can have staggering performance and stack depth behavior. They remain an amazing powerful tool but are too scary to enable by default. To enable them yourself set script.painless.regex.enabled: true in elasticsearch.yml. We’d like very much to have a safe alternative implementation that can be enabled by default so check this space for later developments!

Painless’s native support for regular expressions has syntax constructs:

  • /pattern/: Pattern literals create patterns. This is the only way to create a pattern in painless. The pattern inside the `/`s are just Java regular expressions. See Pattern flags for more.
  • =~: The find operator return a boolean, true if a subsequence of the text matches, false otherwise.
  • ==~: The match operator returns a boolean, true if the text matches, false if it doesn’t.

Using the find operator (=~) you can update all hockey players with "b" in their last name:

POST hockey/player/_update_by_query
{
  "script": {
    "lang": "painless",
    "inline": "if (ctx._source.last =~ /b/) {ctx._source.last += \"matched\"} else {ctx.op = 'noop'}"
  }
}

Using the match operator (==~) you can update all the hockey players who’s names start with a consonant and end with a vowel:

POST hockey/player/_update_by_query
{
  "script": {
    "lang": "painless",
    "inline": "if (ctx._source.last ==~ /[^aeiou].*[aeiou]/) {ctx._source.last += \"matched\"} else {ctx.op = 'noop'}"
  }
}

You can use the Pattern.matcher directly to get a Matcher instance and remove all of the vowels in all of their last names:

POST hockey/player/_update_by_query
{
  "script": {
    "lang": "painless",
    "inline": "ctx._source.last = /[aeiou]/.matcher(ctx._source.last).replaceAll('')"
  }
}

Matcher.replaceAll is just a call to Java’s Matcher's replaceAll method so it supports $1 and \1 for replacements:

POST hockey/player/_update_by_query
{
  "script": {
    "lang": "painless",
    "inline": "ctx._source.last = /n([aeiou])/.matcher(ctx._source.last).replaceAll('$1')"
  }
}

If you need more control over replacements you can call replaceAll on a CharSequence with a Function<Matcher, String> that builds the replacement. This does not support $1 or \1 to access replacements because you already have a reference to the matcher and can get them with m.group(1).

Calling Matcher.find inside of the function that builds the replacement is rude and will likely break the replacement process.

This will make all of the vowels in the hockey player’s last names upper case:

POST hockey/player/_update_by_query
{
  "script": {
    "lang": "painless",
    "inline": "ctx._source.last = ctx._source.last.replaceAll(/[aeiou]/, m -> m.group().toUpperCase(Locale.ROOT))"
  }
}

Or you can use the CharSequence.replaceFirst to make the first vowel in their last names upper case:

POST hockey/player/_update_by_query
{
  "script": {
    "lang": "painless",
    "inline": "ctx._source.last = ctx._source.last.replaceFirst(/[aeiou]/, m -> m.group().toUpperCase(Locale.ROOT))"
  }
}

Note: all of the _update_by_query examples above could really do with a query to limit the data that they pull back. While you could use a Script Query it wouldn’t be as efficient as using any other query because script queries aren’t able to use the inverted index to limit the documents that they have to check.

How painless dispatches functions

edit

Painless uses receiver, name, and arity for method dispatch. For example, s.foo(a, b) is resolved by first getting the class of s and then looking up the method foo with two parameters. This is different from Groovy which uses the runtime types of the parameters and Java which uses the compile time types of the parameters.

The consequence of this that Painless doesn’t support overloaded methods like Java, leading to some trouble when it whitelists classes from the Java standard library. For example, in Java and Groovy, Matcher has two methods: group(int) and group(String). Painless can’t whitelist both of them methods because they have the same name and the same number of parameters. So instead it has group(int) and namedGroup(String).

We have a few justifications for this different way of dispatching methods:

  1. It makes operating on def types simpler and, presumably, faster. Using receiver, name, and arity means when Painless sees a call on a def object it can dispatch the appropriate method without having to do expensive comparisons of the types of the parameters. The same is true for invocations with def typed parameters.
  2. It keeps things consistent. It would be genuinely weird for Painless to behave like Groovy if any def typed parameters were involved and Java otherwise. It’d be slow for it to behave like Groovy all the time.
  3. It keeps Painless maintainable. Adding the Java or Groovy like method dispatch feels like it’d add a ton of complexity which’d make maintenance and other improvements much more difficult.