Skip to main content
Version: Next

read_kv

Read Key-Value pairs from a byte stream.

read_kv [field_split:str, value_split:str, merge=bool, raw=bool, schema=str,
         selector=str, schema_only=bool, unflatten=str]

Description​

The read_kv operator transforms a byte stream into a event stream by parsing the bytes as Key-Value pairs.

Incoming strings are first split into fields according to field_split. This can be a regular expression. For example, the input foo: bar, baz: 42 can be split into foo: bar and baz: 42 with the r",\s*" (a comma, followed by any amount of whitespace) as the field splitter. Note that the matched separators are removed when splitting a string.

Afterwards, the extracted fields are split into their key and value by <value_split>, which can again be a regular expression. In our example, r":\s*" could be used to split foo: bar into the key foo and its value bar, and similarly baz: 42 into baz and 42. The result would thus be {"foo": "bar", "baz": 42}. If the regex matches multiple substrings, only the first match is used.

The supported regular expression syntax is RE2. In particular, this means that lookahead (?=...) and lookbehind (?<=...) are not supported by kv at the moment. However, if the regular expression has a capture group, it is assumed that only the content of the capture group shall be used as the separator. This means that unsupported regular expressions such as (?=foo)bar(?<=baz) can be effectively expressed as foo(bar)baz instead.

Quoted Values​

The parser is aware of double-quotes ("). If the field_split or value_split are found within enclosing quotes, they are not considered matches. This means that both the key and the value may be enclosed in double-quotes.

For example, given \s*,\s* and =, the input

"key"="nested = value",key2="value, and more"

will parse as

{
  "key": "nested = value"
}
{
  "key2": "value, and more"
}

field_split: str (optional)​

The regular expression used to separate individual fields.

Defaults to r"\s".

value_split: str (optional)​

The regular expression used to separate a key from its value.

Defaults to "=".

merge = bool (optional)​

Merges all incoming events into a single schema* that converges over time. This option is usually the fastest for reading highly heterogeneous data, but can lead to huge schemas filled with nulls and imprecise results. Use with caution.

*: In selector mode, only events with the same selector are merged.

This option can not be combined with raw=true, schema=<schema>.

raw = bool (optional)​

Use only the raw types that are native to the parsed format. In the case of KV this means that no parsing of data takes place at all and every value remains a string.

If a known schema is given, fields will still be parsed according to the schema.

Use with caution.

schema = str (optional)​

Provide the name of a schema to be used by the parser.

If a schema with a matching name is installed, the result will always have all fields from that schema.

  • Fields that are specified in the schema, but did not appear in the input will be null.
  • Fields that appear in the input, but not in the schema will also be kept. schema_only=true can be used to reject fields that are not in the schema.

If the given schema does not exist, this option instead assigns the output schema name only.

The schema option is incompatible with the selector option.

selector = str (optional)​

Designates a field value as schema name with an optional dot-separated prefix.

The string is parsed as <fieldname>[:<prefix>]. The prefix is optional and will be prepended to the field value to generate the schema name.

For example, the Suricata EVE JSON format includes a field event_type that contains the event type. Setting the selector to event_type:suricata causes an event with the value flow for the field event_type to map onto the schema suricata.flow.

The selector option is incompatible with the schema option.

schema_only = bool (optional)​

When working with an existing schema, this option will ensure that the output schema has only the fields from that schema. If the schema name is obtained via a selector and it does not exist, this has no effect.

This option requires either schema or selector to be set.

unflatten = str (optional)​

A delimiter that, if present in keys, causes values to be treated as values of nested records.

A popular example of this is the Zeek JSON format. It includes the fields id.orig_h, id.orig_p, id.resp_h, and id.resp_p at the top-level. The data is best modeled as an id record with four nested fields orig_h, orig_p, resp_h, and resp_p.

Without an unflatten separator, the data looks like this:

Without unflattening
{
  "id.orig_h": "1.1.1.1",
  "id.orig_p": 10,
  "id.resp_h": "1.1.1.2",
  "id.resp_p": 5
}

With the unflatten separator set to ., Tenzir reads the events like this:

With 'unflatten'
{
  "id": {
    "orig_h": "1.1.1.1",
    "orig_p": 10,
    "resp_h": "1.1.1.2",
    "resp_p": 5
  }
}

Examples​

Read comma-separated key-value pairs​

// Input: surname:"John Norman", family_name:Smith, date_of_birth: 1995-05-26
read_kv r"\s*,\s*", r"\s*:\s*"
{
  surname: "John Norman",
  family_name: "Smith",
  date_of_birth: 1995-05-26,
}

Extract key-value pairs with more complex rules​

// Input: PATH: C:\foo INPUT_MESSAGE: hello world
read_kv r"(\s+)[A-Z][A-Z_]+:", r":\s*"
{
  PATH: "C:\\foo",
  INPUT_MESSAGE: "hello world",
}

This requires lookahead because not every whitespace acts as a field separator. Instead, we only want to split if the whitespace is followed by [A-Z][A-Z_]+:, i.e., at least two uppercase characters followed by a colon. We can express this as "(\s+)[A-Z][A-Z_]+:", which yields PATH: C:\foo and INPUT_MESSAGE: hello world. We then split the key from its value with ":\s*". Since only the first match is used to split key and value, this leaves the path intact.