The exporter measures the number of TCP connections established to the ports of the host machine. To be measured, a connection has to be established for a defined duration. For this reason the exported metric is called netstat_tcp_longterm_connections_total.

The exporter uses the package github.com/namsral/flag and can be configured via:

• Flags
• Environment variables
• Configuration file

simple: [bool] If set to true the exported metric will be a gauge counting all longterm established connections. If false (default) it will create a gauge with labels for each observed port and TCP version.

tcpv6: [bool] If set to true (default) IPv6 connections will be measured, too.

port: [int] Sets a specific port to observe. The default is -1 which measures all ports (except the listening port of the exporter).

listen: [int] Sets the listening port of the exporter. The default value is 9690.

duration: [int] The minimal duration in seconds after a connection is concerned as a longterm connection. The default value is 6.

The exporter is written in Go and can be build and used with

#Get & build the exporter
go get github.com/ipbhalle/tcp_established_exporter
#Run the exporter
go run github.com/ipbhalle/tcp_established_exporter

You can export the settings as environmental variables:

export SIMPLE=true
export TCPV6=true

flag:

tcp_established_exporter --simple=true --tcpv6=true

or config file:

tcp_established_exporter --conf sample.conf

with sample.conf

simple true
tcpv6 true

The metrics will be exposed at /metrics

The exporter can be used to measure/observe the number of connections to one or multiple services. It can be useful for applications that use WebSockets, because each user establishes a longterm connection. It was designed to count the number of users for Shiny web applications. But it could be also used for other use cases, e.g. if you want to guarantee that there is a connection to a service (and trigger an alert otherwise).

The docker image of this exporter can be used as a sidecar in Kubernetes to measure the longterm connections of a pod. Different containers running on the same pod are sharing the resources of the pod. In that way it possible to detect the TCP connections of applications running in other containers. A complete example will be shown in the future.