Skip to main content
Version: Tenzir v4.11

Type System

One design goal of Tenzir is being expressive enough to capture the semantics of the domain. This led us to develop a rich type system for structured security data, inspired by the Zeek network security monitor.


The diagram below illustrates the type system at a glance:

Basic Types

Basic types are stateless types with a static structure. Their representation is known a-priori.

The following tables summarizes the basic types.

noneDenotes an absent or invalid value
boolA boolean value
int64A 64-bit signed integer
uint64A 64-bit unsigned integer
doubleA 64-bit double (IEEE 754)
durationA time span (nanosecond granularity)
timeA time point (nanosecond granularity)
stringA UTF-8 encoded string
blobAn arbitrary sequence of bytes
patternA regular expression
ipAn IPv4 or IPv6 address
subnetAn IPv4 or IPv6 subnet

The blob type is still experimental and not yet fully supported.

Complex Types

Complex types are stateful types that carry additional runtime information.


The enum type is a list of predefined string values. It comes in handy for low-cardinality values from a fixed set of options.

Tenzir implements an enum as an Arrow Dictionary.


The list type is an ordered sequence of values with a fixed element type.

Lists have zero or more elements.


The record type consists of an ordered sequence fields, each of which have a name and type. Records must have at least one field.

The field name is an arbitrary UTF-8 string.

The field type is any Tenzir type.


All types are optional in that there exists an additional null data point in every value domain. Consequently, Tenzir does not have a special type to indicate optionality.


Every type has zero or more attributes, which are free-form key-value pairs to enrich types with custom semantics.

For duration types, the attribute unit can be set to customize how numeric values are interpreted by parsers. For example, #unit=s makes it so that the number 1708430755 is parsed as "1708430755s". The unit attribute can also be set for timestamps, which means that numeric values are interpreted as offsets from the Unix epoch. For example, #unit=s makes it so that 1708430755 is parsed as "2023-02-20T12:05:55".


An alias wraps an existing type under a new name. Aliases are first-class types, meaning you can also attach separate attributes to them.

All alias types have a name. They coexist in a global namespace.

Alias = Specialization

An alias always refines the type it points to, i.e., it is more specific that its parent. For example, let's assume a type U is an alias for an existing type T. Then, the type extractor :U only resolves for types of instance U. However, :T comprises both instances of types U and T.

Type Construction

Tenzir comes with a lean type definition language. Think JSON schema fused with Kaitai, with the goal that humans can jot it down quickly. We use YAML as vehicle to express the structure because that's already Tenzir's language for configuration.

How do you create a type? Let T be an existing type, then you can construct a new type like this:

type: T
- key: value
- value # values are optional

This begs a recursive question: what are valid values for T?

The answer: basic types, such as bool, string, ip, or complex types defined according the rules described below.

Alias Type

Use a dictionary to give an existing type a new name:

url: string

This introduces a type with the name url.

Instead of referencing an existing type, you can also provide an inline definition of a type:

type: string
- unescaped

Aliases nest. For example, you can create three aliased types ABCstring like this:

C: string

Enumeration Type

Use an enum dictionary with a list of allowed values to define it:

- on
- off

List Type

Use the list key to define list with a corresponding element type on the right-hand side:

list: url

Lists nest naturally:

- on
- off

Record Type

Use the record key to define a list of fields:

- src_ip: ip
- dst_ip: ip
- src_port: port # existing alias
- dst_port: port # existing alias
- proto: string

Records nest naturally:

- src:
- ip: ip
- port: uint64
- dst:
- ip: ip
- port: uint64
- proto: string

We can also factor nested records into dedicated types:

- ip: ip
- port: uint64

Thereafter we can reference the endpoint type by its name:

- src: endpoint
- dst: endpoint

Records of lists:

- values:
list: string

Lists of records:

- question: string
- correct: bool

Record Algebra

In addition to nesting records, a frequent use case is embedding fields of one record into another. This occurs often with families of event types that share a large portion of fields. Algebraically, we want to "add" the two records together to combine their field.

For example, consider this common record with a single field shared:

# The base record.
- shared:
type: bool

To embed common in another record, reference it using the base key. This requires moving the fields one layer down under the fields key:

- common
- unique: string

The effect is the same specifying a single record with two fields:

- shared: bool
- unique: string

There are three special keys under the record key that control what to do when field name clashes occur between the two records:

  1. base: raise an error
  2. implant: prefer the base record
  3. extend: prefer the current record

Comparison to Arrow

All Tenzir types have a lossless mapping to Arrow types, however, not all Arrow types have a Tenzir equivalent. As a result, it is currently not yet possible to import arbitrary Arrow data. In the future, we plan to extend our support for Arrow-native types and also offer conversion options for seamless data handover.

Tenzir has a few domain-specific types that map to Arrow extension types. These are currently:

  • enum
  • ip
  • subnet

Note that Tenzir attaches attributes to a top-level type instance, where Arrow only allows type meta data for record fields.

Tenzir treats type meta data differently from Arrow. In Tenzir, the type is the component that contains metadata. In Arrow, it's the record field or the schema. As a result, we can simply define a schema as named record type.

More on Arrow & Tenzir

If you want to learn more about why Tenzir uses Apache Arrow, please read our two blog posts that explain why we build on top of Arrow.