Bloblang, or blobl for short, is a language designed for mapping data of a wide variety of forms. It’s a safe, fast, and powerful way to perform document mapping within Redpanda Connect. It also has a Go API for writing your own functions and methods as plugins.

Bloblang is available as a processor and it’s also possible to use blobl queries in function interpolations.

You can also execute Bloblang mappings on the command-line with the blobl subcommand:

cat data.jsonl | rpk connect blobl 'foo.(bar | baz).buz'

This document outlines the core features of the Bloblang language, but if you’re totally new to Bloblang then it’s worth following the walkthrough first.


A Bloblang mapping expresses how to create a new document by extracting data from an existing input document. Assignments consist of a dot separated path segments on the left-hand side describing a field to be created within the new document, and a right-hand side query describing what the content of the new field should be.

The keyword root on the left-hand side refers to the root of the new document, the keyword this on the right-hand side refers to the current context of the query, which is the read-only input document when querying from the root of a mapping: =
root.type = "yo"

# Both `root` and `this` are optional, and will be inferred in their absence.
content = thing.doc.message

# In:  {"thing":{"id":"wat1","doc":{"title":"wut","message":"hello world"}}}
# Out: {"content":"hello world","id":"wat1","type":"yo"}

Since the document being created starts off empty it is sometimes useful to begin a mapping by copying the entire contents of the input document, which can be expressed by assigning this to root.

root = this = "added value"

# In:  {"id":"wat1","message":"hello world"}
# Out: {"id":"wat1","message":"hello world","foo":"added value"}

If the new document root is never assigned to or otherwise mutated then the original document remains unchanged.

Special characters in paths

Quotes can be used to describe sections of a field path that contain whitespace, dots or other special characters:

# Use quotes around a path segment in order to include whitespace or dots within
# the path
root."".baz = this."buz bev".fub

# In:  {"buz bev":{"fub":"hello world"}}
# Out: {"":{"baz":"hello world"}}

Non-structured data

Bloblang is able to map data that is unstructured, whether it’s a log line or a binary blob, by referencing it with the content function, which returns the raw bytes of the input document:

# Parse a base64 encoded JSON document
root = content().decode("base64").parse_json()

# In:  eyJmb28iOiJiYXIifQ==
# Out: {"foo":"bar"}

And your newly mapped document can also be unstructured, simply assign a value type to the root of your document:

root =

# In:  {"foo":"hello world"}
# Out: hello world

And the resulting message payload will be the raw value you’ve assigned.


It’s possible to selectively delete fields from an object by assigning the function deleted() to the field path:

root = this = deleted()

# In:  {"id":"wat1","message":"hello world","bar":"remove me"}
# Out: {"id":"wat1","message":"hello world"}


Another type of assignment is a let statement, which creates a variable that can be referenced elsewhere within a mapping. Variables are discarded at the end of the mapping and are mostly useful for query reuse. Variables are referenced within queries with $:

# Set a temporary variable
let foo = "yo"

root.new_doc.type = $foo


Redpanda Connect messages contain metadata that is separate from the main payload, in Bloblang you can modify the metadata of the resulting message with the meta assignment keyword. Metadata values of the resulting message are referenced within queries with the @ operator or the metadata() function:

# Reference a metadata value = @kafka_topic # Or `@.kafka_topic` or `metadata("kafka_topic")`

# Delete all metadata
meta = deleted()

# Set metadata values
meta bar = "hello world"
meta baz = {
  "something": "structured"

# Get an object of key/values for all metadata
root.meta_obj = @ # Or `metadata()`


The pipe operator (|) used within brackets allows you to coalesce multiple candidates for a path segment. The first field that exists and has a non-null value will be selected:

root.new_doc.type = this.thing.(article | comment | this).type

# In:  {"thing":{"article":{"type":"foo"}}}
# Out: {"new_doc":{"type":"foo"}}

# In:  {"thing":{"comment":{"type":"bar"}}}
# Out: {"new_doc":{"type":"bar"}}

# In:  {"thing":{"type":"baz"}}
# Out: {"new_doc":{"type":"baz"}}

Opening brackets on a field begins a query where the context of this changes to value of the path it is opened upon, therefore in the above example this within the brackets refers to the contents of this.thing.


Bloblang supports number, boolean, string, null, array and object literals:

root = [
  7, false, "string", null, {
    "first": 11,
    "second": {"foo":"bar"},
    "third": """multiple
lines on this

# In:  {}
# Out: [7,false,"string",null,{"first":11,"second":{"foo":"bar"},"third":"multiple\nlines on this\nstring"}]

The values within literal arrays and objects can be dynamic query expressions, as well as the keys of object literals.


You might’ve already spotted, comments are started with a hash (#) and end with a line break:

root = this.some.value # And now this is a comment

Boolean logic and arithmetic

Bloblang supports a range of boolean operators !, >, >=, ==, <, <=, &&, || and mathematical operators +, -, *, /, %:

root.is_big = this.number > 100
root.multiplied = this.number * 7

# In:  {"number":50}
# Out: {"is_big":false,"multiplied":350}

# In:  {"number":150}
# Out: {"is_big":true,"multiplied":1050}

For more information about these operators and how they work check out the arithmetic page.

Conditional mapping

Use if as either a statement or an expression in order to perform maps conditionally:

root = this

root.sorted_foo = if == "array" { }

if == "string" {
  root.upper_foo =
  root.lower_foo =

# In:  {"foo":"FooBar"}
# Out: {"foo":"FooBar","lower_foo":"foobar","upper_foo":"FOOBAR"}

# In:  {"foo":["foo","bar"]}
# Out: {"foo":["foo","bar"],"sorted_foo":["bar","foo"]}

And add as many else if queries as you like, followed by an optional final fallback else:

root.sound = if this.type == "cat" {
} else if this.type == "dog" {
} else {
  "sweet sweet silence"

# In:  {"type":"cat","cat":{"meow":"meeeeooooow!"}}
# Out: {"sound":"meeeeooooow!"}

# In:  {"type":"dog","dog":{"woof":"guurrrr woof woof!"}}
# Out: {"sound":"GUURRRR WOOF WOOF!"}

# In:  {"type":"caterpillar","caterpillar":{"name":"oleg"}}
# Out: {"sound":"sweet sweet silence"}

Pattern matching

A match expression allows you to perform conditional mappings on a value, each case should be either a boolean expression, a literal value to compare against the target value, or an underscore (_) which captures values that have not matched a prior case:

root.new_doc = match this.doc {
  this.type == "article" => this.article
  this.type == "comment" => this.comment
  _ => this

# In:  {"doc":{"type":"article","article":{"id":"foo","content":"qux"}}}
# Out: {"new_doc":{"id":"foo","content":"qux"}}

# In:  {"doc":{"type":"comment","comment":{"id":"bar","content":"quz"}}}
# Out: {"new_doc":{"id":"bar","content":"quz"}}

# In:  {"doc":{"type":"neither","content":"some other stuff unchanged"}}
# Out: {"new_doc":{"type":"neither","content":"some other stuff unchanged"}}

Within a match block the context of this changes to the pattern matched expression, therefore this within the match expression above refers to this.doc.

Match cases can specify a literal value for simple comparison:

root = this
root.type = match this.type { "doc" => "document", "art" => "article", _ => this }

# In:  {"type":"doc","foo":"bar"}
# Out: {"type":"document","foo":"bar"}

The match expression can also be left unset which means the context remains unchanged, and the catch-all case can also be omitted:

root.new_doc = match {
  this.doc.type == "article" => this.doc.article
  this.doc.type == "comment" => this.doc.comment

# In:  {"doc":{"type":"neither","content":"some other stuff unchanged"}}
# Out: {"doc":{"type":"neither","content":"some other stuff unchanged"}}

If no case matches then the mapping is skipped entirely, hence we would end up with the original document in this case.


Functions can be placed anywhere and allow you to extract information from your environment, generate values, or access data from the underlying message being mapped: = uuid_v4()
root.doc.received_at = now() = hostname()

Functions support both named and nameless style arguments:

root.values_one = range(start: 0, stop: this.max, step: 2)
root.values_two = range(0, this.max, 2)

You can find a full list of functions and their parameters in the functions page.


Methods are similar to functions but enact upon a target value, these provide most of the power in Bloblang as they allow you to augment query values and can be added to any expression (including other methods): =
root.doc.reduced_nums = this.thing.nums.map_each(num -> if num < 10 {
} else {
  num - 10
root.has_good_taste = ["pikachu","mewtwo","magmar"].contains(this.user.fav_pokemon)

Methods also support both named and nameless style arguments:

root.foo_one = this.(bar | baz).trim().replace_all(old: "dog", new: "cat")
root.foo_two = this.(bar | baz).trim().replace_all("dog", "cat")

You can find a full list of methods and their parameters in the methods page.


Defining named maps allows you to reuse common mappings on values with the apply method:

map things {
  root.first  = this.thing_one
  root.second = this.thing_two
} = this.value_one.apply("things") = this.value_two.apply("things")

# In:  {"value_one":{"thing_one":"hey","thing_two":"yo"},"value_two":{"thing_one":"sup","thing_two":"waddup"}}
# Out: {"foo":{"first":"hey","second":"yo"},"bar":{"first":"sup","second":"waddup"}}

Within a map the keyword root refers to a newly created document that will replace the target of the map, and this refers to the original value of the target. The argument of apply is a string, which allows you to dynamically resolve the mapping to apply.

Import maps

It’s possible to import maps defined in a file with an import statement:

import "./common_maps.blobl" = this.value_one.apply("things") = this.value_two.apply("things")

Imports from a Bloblang mapping within a Redpanda Connect config are relative to the process running the config. Imports from an imported file are relative to the file that is importing it.


By assigning the root of a mapped document to the deleted() function you can delete a message entirely:

# Filter all messages that have fewer than 10 URLs.
root = if this.doc.urls.length() < 10 { deleted() }

Error handling

Functions and methods can fail under certain circumstances, such as when they receive types they aren’t able to act upon. These failures, when not caught, will cause the entire mapping to fail. However, the method catch can be used in order to return a value when a failure occurs instead:

# Map an empty array to `foo` if the field `bar` is not a string. =",").catch([])

Since catch is a method it can also be attached to bracketed map expressions:

# Map `false` if any of the operations in this boolean query fail.
root.thing = ( > && this.baz.contains("wut") ).catch(false)

And one of the more powerful features of Bloblang is that a single catch method at the end of a chain of methods can recover errors from any method in the chain:

# Catch errors caused by:
# - foo not existing
# - foo not being a string
# - an element from split foo not being a valid JSON string
root.things =",").map_each( ele -> ele.parse_json() ).catch([])

# Specifically catch a JSON parse error
root.things =",").map_each( ele -> ele.parse_json().catch({}) )

However, the catch method only acts on errors, sometimes it’s also useful to set a fall back value when a query returns null in which case the method or can be used the same way:

# Map "default" if either the element index 5 does not exist, or the underlying
# element is `null`. ="default")

Unit testing

It’s possible to execute unit tests for your Bloblang mappings using the standard Redpanda Connect unit test capabilities outlined in this document.


  1. I’m seeing unable to reference message as structured (with 'this') when I try to run mappings with rpk connect blobl.

That particular error message means the mapping is failing to parse what’s being fed in as a JSON document. Make sure that the data you are feeding in is valid JSON, and also that the documents do not contain line breaks as rpk connect blobl will parse each line individually.

Why? That’s a good question. Bloblang supports non-JSON formats too, so it can’t delimit documents with a streaming JSON parser like tools such as jq, so instead it uses line breaks to determine the boundaries of each message.