This application allows you to visualize the highly complex flow Information data exported by network devices, in an easy and understandable graphical way.

It takes a decoded flow data stream from GoFlow2 as input and exports it to Prometheus Server to be stored. Then, data can be queried and graphically represented using tools such as Grafana.

Gobi aims to be a fast, lightweight and easy to use tool for network flows analysis. It favours speed over absolute accuracy. With Gobi, flows are pre-processed and aggregated before getting stored to the DB. The subsequent data visualization operations are therefore greatly improved, allowing short response times even over large analysis intervals.

The examples folder contains two ready to deploy Gobi setups, one for docker-compose and the other for kubernetes environment. Pick one and send to it a NetFlow/IPFIX stream to get a sample of what Gobi can do. Here you can find detailed installation instructions.

To set up a basic flow pipeline, download Gobi and GoFlow2 to your local machine and create a gobi.yml config file like this template:

global:
  metricsaddr: :9310

producer:
  input: stdin

promexporters:
  - metricsname: example
    labelset: ["SamplerAddress"]

Then start the pipeline with the following command:

$ ./goflow2 -transport.file.sep= -format=pb -format.protobuf.fixedlen=true | ./gobi -f gobi.yml

The pipeline is now ready to be scraped by Prometheus server at the IP address of your local machine. A full config keys list supported by gobi.yml is available here. The provided examples are a good starting point for Gobi deployments.

Gobi is modular. The producer receives input flow data, converts it in go structures, enriches it with extended infos and outputs to a set of registered exporter instances. Each exporter can then shape its data as needed and output it in the the desired format. The current Gobi implementation features a GoFlow2 producer and a Prometheus exporter.

It takes its input from stdin or a named pipe. The expected format is GPB, as defined by GoFlow2 proto file (not all fields are currently exported). Each flow is then enriched with data coming from these popular DBs:

• MaxMind GeoLite2 Country.
• MaxMind GeoLite2 ASN.
• TheDiveO netdb.

Finally, a copy of this enriched flow is sent to each registered exporter for further processing.

Flows coming from producer are initially stored in a memory buffer, waiting for a Prometheus server scrape event. When scraping happens, promexporter checks them out and process flows it as follows:

• Aggregates over a user defined field (labels) set.
• Checks if those aggregates are compliant with a minimum user defined data rate over the scrape interval. If not, they are removed from the table and accumulated in separate "untracked" counters.
• Aggregates are then checked for aging. The expired ones are evicted from the table.
• Finally, the remaining aggregates and the untracked counters are sent to the Prometheus server.

There is a MaxScrapeInterval user configurable timer that flushes the buffer if no scrapes occurs.

A key aspect of this architecture is that Prometheus finally gives the clock to the resampling process, with its scrape interval. This allows data accuracy to be kept to acceptable levels while enjoying the simplicity and performance of PromQL.

Licensed under BSD 3-Clause license. See LICENSE.