This application is a NetFlow/IPFIX/sFlow collector in Go.
It gathers network information (IP, interfaces, routers) from different flow protocols, serializes it in a common format.
You will want to use GoFlow if:
- You receive a decent amount of network samples and need horizontal scalability
- Have protocol diversity and need a consistent format
- Require raw samples and build aggregation and custom enrichment
This software is the entry point of a pipeline. The storage, transport, enrichment, graphing, alerting are not provided.
This work is a fork of a previous open-source GoFlow code built and used at Cloudflare. It lives in its own GitHub organization to be maintained more easily.
Among the differences with the original code: The serializer and transport options have been revamped to make this program more user-friendly and target new use-cases like logging providers. Minimal changes in the decoding libraries.
In order to enable load-balancing and optimizations, the GoFlow library has a decoder
which converts
the payload of a flow packet into a Go structure.
The producer
functions (one per protocol) then converts those structures into a protobuf (pb/flow.pb
)
which contains the fields a network engineer is interested in.
The flow packets usually contains multiples samples
This acts as an abstraction of a sample.
The format
directory offers various utilities to process the protobuf. It can convert
The transport
provides different way of processing the protobuf. Either sending it via Kafka or
send it to a file (or stdout).
GoFlow2 is a wrapper of all the functions and chains thems.
You can build your own collector using this base and replace parts:
- Use different transport (e.g: RabbitMQ instead of Kafka)
- Convert to another format (e.g: Cap'n Proto, Avro, instead of protobuf)
- Decode different samples (e.g: not only IP networks, add MPLS)
- Different metrics system (e.g: OpenTelemetry)
The sampling protocols have distinct features:
sFlow is a stateless protocol which sends the full header of a packet with router information (interfaces, destination AS) while NetFlow/IPFIX rely on templates that contain fields (e.g: source IPv6).
The sampling rate in NetFlow/IPFIX is provided by Option Data Sets. This is why it can take a few minutes for the packets to be decoded until all the templates are received (Option Template and Data Template).
Both of these protocols bundle multiple samples (Data Set in NetFlow/IPFIX and Flow Sample in sFlow) in one packet.
The advantages of using an abstract network flow format, such as protobuf, is it enables summing over the protocols (e.g: per ASN or per port, rather than per (ASN, router) and (port, router)).
To read more about the protocols and how they are mapped inside, check out page
Collection:
- NetFlow v5
- IPFIX/NetFlow v9 (sampling rate provided by the Option Data Set)
- sFlow v5
(adding NetFlow v1,7,8 is being evaluated)
Production:
- Convert to protobuf or json
- Prints to the console/file
- Sends to Kafka and partition
Monitoring via Prometheus metrics
To read about agents that samples network traffic, check this page.
To set up the collector, download the latest release corresponding to your OS and run the following command (the binaries have a suffix with the version):
$ ./goflow2
By default, this command will launch an sFlow collector on port :6343
and
a NetFlowV9/IPFIX collector on port :2055
.
By default, the samples received will be printed in JSON format on the stdout.
{
"Type": "SFLOW_5",
"TimeFlowEnd": 1621820000,
"TimeFlowStart": 1621820000,
"TimeReceived": 1621820000,
"Bytes": 70,
"Packets": 1,
"SamplingRate": 100,
"SamplerAddress": "192.168.1.254",
"DstAddr": "10.0.0.1",
"DstMac": "ff:ff:ff:ff:ff:ff",
"SrcAddr": "192.168.1.1",
"SrcMac": "ff:ff:ff:ff:ff:ff",
"InIf": 1,
"OutIf": 2,
"Etype": 2048,
"EtypeName": "IPv4",
"Proto": 6,
"ProtoName": "TCP",
"SrcPort": 443,
"DstPort": 46344,
"FragmentId": 54044,
"FragmentOffset": 16384,
...
"IPTTL": 64,
"IPTos": 0,
"TCPFlags": 16,
}
If you are using a log integration (e.g: Loki with Promtail, Splunk, Fluentd, Google Cloud Logs, etc.), just send the output into a file.
$ ./goflow2 -transport.file /var/logs/goflow2.log
To enable Kafka and send protobuf, use the following arguments:
$ ./goflow2 -transport=kafka -transport.kafka.brokers=localhost:9092 -transport.kafka.topic=flows -format=pb
By default, the distribution will be randomized. To partition the feed (any field of the protobuf is available), the following options can be used:
-transport.kafka.hashing=true \
-format.hash=SamplerAddress,DstAS
By default, compression is disabled when sending data to Kafka. To change the kafka compression type of the producer side configure the following option:
-transport.kafka.compression.type=gzip
The list of codecs is available in the Sarama documentation.
By default, the collector will listen for IPFIX/NetFlow V9 on port 2055
and sFlow on port 6343.
To change the sockets binding, you can set the -listen
argument and a URI
for each protocol (netflow
, sflow
and nfl
as scheme) separated by a comma.
For instance, to create 4 parallel sockets of sFlow and one of NetFlow V5, you can use:
$ ./goflow2 -listen 'sflow://:6343?count=4,nfl://:2055'
You can also run directly with a container:
$ sudo docker run -p 6343:6343/udp -p 2055:2055/udp -ti netsampler/goflow2:latest
In the case of exotic template fields or extra payload not supported by GoFlow2
of out the box, it is possible to pass a mapping file using -mapping mapping.yaml
.
A sample file is available in the cmd/goflow2
directory.
For instance, certain devices producing IPFIX use ingressPhysicalInterface
(id: 252)
and do not use ingressInterface
(id: 10). Using the following you can have the interface mapped
in the InIf protobuf field without changing the code.
ipfix:
mapping:
- field: 252
destination: InIf
- field: 253
destination: OutIf
The JSON format is advised only when consuming a small amount of data directly. For bigger workloads, the protobuf output format provides a binary representation and is preferred. It can also be extended with enrichment as long as the user keep the same IDs.
If you want to develop applications, build pb/flow.proto
into the language you want:
When adding custom fields, picking a field ID โฅ 1000 is suggested.
Check the docs for more information about compiling protobuf.
A basic enrichment tool is available in the cmd/enricher
directory.
You need to load the Maxmind GeoIP ASN and Country databases using -db.asn
and -db.country
.
Running a flow enrichment system is as simple as a pipe. Once you plug the stdin of the enricher to the stdout of GoFlow in protobuf, the source and destination IP addresses will automatically be mapped with a database for Autonomous System Number and Country. Similar output options as GoFlow are provided.
$ ./goflow2 -transport.file.sep= -format=pb -format.protobuf.fixedlen=true | ./enricher -db.asn path-to/GeoLite2-ASN.mmdb -db.country path-to/GeoLite2-Country.mmdb
For a more scalable production setting, Kafka and protobuf are recommended. Stream operations (aggregation and filtering) can be done with stream-processor tools. For instance Flink, or the more recent Kafka Streams and kSQLdb. Direct storage can be done with data-warehouses like Clickhouse.
In some cases, the consumer will require protobuf messages to be prefixed by
length. To do this, use the flag -format.protobuf.fixedlen=true
.
This repository contains examples of pipelines with docker-compose. The available pipelines are:
Are you using GoFlow2 in production at scale? Add yourself here!
This project welcomes pull-requests, whether it's documentation, instrumentation (e.g: docker-compose, metrics), internals (protocol libraries), integration (new CLI feature) or else! Just make sure to check for the use-cases via an issue.
This software would not exist without the testing and commits from its users and contributors.
Licensed under the BSD-3 License.