Tenzir v4.3

Exciting times, Tenzir v4.3 is out! The headlining feature is Fluent Bit support with the fluent-bit source and sink operators. Imagine you can use all Fluent Bit connectors plus what Tenzir already offers. What a treat!

Fluent Bit on Steroids

Fluent Bit is a remarkable piece of open source software that offers observability pipelines—quite similar to Tenzir actually. That said, our target audience is different: rather than targeting observability, we focus on the intersection of the security and data.

By bringing the two ecosystems together, you, dear user, benefit from the union of features. Before diving into some examples, let's briefly compare the tech. Fluent Bit features inputs and outputs to get data in and out of the ecosystem. These are equivalent to Tenzir's connectors. Fluent Bit also has parsers that map to the equally named concept of Tenzir parsers. Fluent Bit's filters would be implemented as transformations in Tenzir, i.e., operators that have a non-void input and output. The diagram illustrates these relationships:

It's important to note that Tenzir pipelines separate I/O and computation. Connectors do I/O and formats are responsible for (un)structuring data. The data paths are symmetric in that a loader ships bytes to a parser that in turn produces events, and a printer accepts events and turns them in to bytes that a corresponding saver sends away.

We implemented the fluent-bit operator as a fusion of connector and format. We did not integrate Fluent Bit's powerful parser abstraction, as we have an existing framework in place for that. Similarly, we did not integrate Fluent Bit's filters, as we have a variety of transformation operators for that.

How do I use it?

The fluent-bit source and sink operator is where the action happens. They have the following syntax:

fluent-bit <plugin> [<key=value>..]

Both operators are very similar to the fluent-bit command line tool, which has the usage fluent-bit -i <input> -o <output> -p key=value. In Tenzir, the source operator implies the options -i and the sink -o, so you don't have to write them. Similarly, appending properties in the form of key-value pairs is so common that you can omit the -p options.

Let's walk through some examples. Say you want to sample three values with Fluent Bit's random input:

tenzir 'fluent-bit random | head 3 | write json -c'

This prints:

{"timestamp": "2023-09-23T07:56:47.957369", "message": {"rand_value": 8106944690543729752}}
{"timestamp": "2023-09-23T07:56:48.959997", "message": {"rand_value": 2072095294278847853}}
{"timestamp": "2023-09-23T07:56:49.959988", "message": {"rand_value": 5606209024700423100}}

Regarding the framing: Fluent Bit's event format produces events in the form of an array that can have one of two possible shapes:

1. [[TIMESTAMP, METADATA], MESSAGE]
2. [TIMESTAMP, MESSAGE]

We convert this into an event record with 3 fields:

1. timestamp: the event timestamp
2. metadata: object with key-value pairs
3. message: arbitrary object with inferred schema

The field metadata is optional, as shown in the above example.

Many Fluent Bit inputs perform network I/O. Here's a TCP socket example:

# Terminal A
tenzir 'fluent-bit tcp'
# Terminal B
echo '{"foo": {"bar": 42}}' | nc 127.0.0.1 5170

This outputs in terminal A:

{
  "timestamp": "2023-09-23T09:35:10.623745",
  "message": {
    "foo": {
      "bar": 42
    }
  }
}

Let's pick another input, opentelemetry:

tenzir 'fluent-bit opentelemetry'

This opens a socket on port 4318 that you can send now telemetry to. Instead of curl, we're using our own HTTPie-like http connector to issue a POST request:

tenzir 'from http POST 127.0.0.1:4318/v1/logs resourceLogs:=[{"resource":{},"scopeLogs":[{"scope":{},"logRecords":[{"timeUnixNano":"1660296023390371588","body":{"stringValue":"{\"message\":\"dummy\"}"},"traceId":"","spanId":""}]}]}]'

You should then see:

{
  "timestamp": "2022-08-12T09:20:24.698112",
  "message": {
    "log": {
      "message": "dummy"
    }
  }
}

More powerful inputs mimic other applications, like Splunk or ElasticSearch. Want Tenzir to be like Splunk via Fluent Bit? Here you go:

tenzir 'fluent-bit splunk'

You just got a Splunk HEC API waiting for you at port 9880. This is one of the most amazing things about this integration. The entire Fluent Bit connector ecosystem is now at your fingertips!

This extends to the outputs as well. Most mundanely, you can use the stdout output from Fluent Bit as follows:

tenzir 'show operators | head 3 | fluent-bit stdout'

This yields:

[0] lib.0: [[1695494117.866096973, {}], {"name"=>"batch", "source"=>false, "transformation"=>true, "sink"=>false}]
[1] lib.0: [[1695494117.866101980, {}], {"name"=>"compress", "source"=>false, "transformation"=>true, "sink"=>false}]
[2] lib.0: [[1695494117.866103887, {}], {"name"=>"decapsulate", "source"=>false, "transformation"=>true, "sink"=>false}]

Want to send Suricata alerts to Slack? Here is your pipeline:

from file --follow eve.json
| where #schema == "suricata.alert"
| fluent-bit slack webhook=https://hooks.slack.com/services/T00000000/B00000000/XXXXXXXXXXXXXXXXXXXXXXXX

Or send 'em to Splunk by changing the sink to:

fluent-bit splunk host=127.0.0.1 port=8088 tls=on tls.verify=off

Oh wait, Elastic? Here you go:

fluent-bit es host=192.168.2.3 port=9200 index=my_index type=my_type

Hopefully the general pattern is clear now.

Finally, there's a cool property of Fluent Bit: it's symmetric like Tenzir. Remember how you can use ZeroMQ sockets to bridge pipelines?

# Terminal A
tenzir 'from zmq'
# Terminal B
tenzir 'show version | to zmq'

You can do the same with Fluent Bit's Forward protocol:

# Terminal A
tenzir 'fluent-bit forward'
# Terminal B
tenzir 'show version | fluent-bit forward'

(We'll leave it to you to do the same with Kafka.)

Implementation: MsgPack vs. Arrow

Fluent Bit uses MsgPack internally, a binary version of JSON with a slightly richer set of types. Once a Fluent Bit input onboards data into the internal format, all operations compute on MsgPack. And before data exits Fluent Bit, it gets converted from MsgPack to the native format of the output.

Incidentally, Tenzir also had an optional MsgPack implementation of its data plane. However, we dropped the MsgPack encoding and switched to Apache Arrow exclusively. The reason is that most of our workloads are analytical, where a columnar representation (especially with large batching) outperforms due data locality and the ability to tap into vectorized computations. Moreover, our objective is to soon integrate natively with several data tools, such as DuckDB, pandas, polars, etc.—all of which speak Arrow.

Want to try it yourself? Head over to app.tenzir.com where you start for free and manage Tenzir nodes and run Tenzir and Fluent Bit pipelines.

Tidbits

Besides Fluent Bit, the team at Tenzir has been working on some other noteworthy improvements and features that we would like to share:

JSON Parser Improvements

We've revamped our JSON parser to be a lot faster and more accurate in type inference.

Schema inference now supports empty records and empty lists. Previously both were indistinguishable from null values. This is best illustrated on an example:

{"foo": [], "bar": {}, "baz": "127.0.0.1"}
{"foo": [null], "bar": null, "baz": "::1"}
{"foo": null, "bar": {}, "baz": "localhost"}

With Tenzir v4.2, The fields foo and bar would've been dropped from the input, and baz had the type string for all three events.

With Tenzir v4.3, foo is of type list<null>, bar of type record {}, and baz of type ip for the first two events, and of type string for the third.

YAML Format

The new yaml format supports reading and writing YAML documents and document streams.

For example, you can now render the configuration of the current node as valid YAML:

show config | write yaml

This yields:

---
tenzir:
  allow-unsafe-pipelines: true
  operators:
    suricata: "shell 'suricata -r /dev/stdin --set outputs.1.eve-log.filename=/dev/stdout --set logging.outputs.0.console.enabled=no' | read suricata\n"
    zeek: "shell 'eval \"$(zkg env)\" && zeek -r - LogAscii::output_to_stdout=T JSONStreaming::disable_default_logs=T JSONStreaming::enable_log_rotation=F json-streaming-logs' | read zeek-json --no-infer\n"
...

Another example, perhaps just a party tricks, is converting YAML to JSON:

tenzir 'read yaml | write json' < input.json

Pipeline Labels

Nodes now support setting labels for pipelines. This feature isn't yet enabled in the app, but will be available soon for all nodes updated to v4.3 or newer.