COLLECTOR

Categraf

Nightingale evaluates alerts on data you already have. Categraf is one way to put it there: a single agent covering operating systems, middleware, databases, network gear and Kubernetes, writing over Prometheus Remote Write. MIT licensed, built by the team behind Nightingale.

100+Collection plugins

MITLicensed

1Agent per host

Install and run

Download a release build

Archives are per platform and already contain the conf directory, so the agent runs straight out of the tarball. Linux ships for amd64 and arm64, Windows as a zip. Two variants sit beside the default build: -with-cgo-plugin adds the NVIDIA DCGM collector, and categraf-slim drops everything but metrics collection.

tar xzf categraf-v<version>-linux-amd64.tar.gzcd categraf-v<version>-linux-amd64

The version is part of the filename, so there is no fixed “latest” URL — pick the build from the releases page.

Releases on GitHub →

Docker

The image carries the binary at /usr/bin/categraf and the sample config at /etc/categraf/conf. Set N9E_HOST and the entrypoint rewrites the server address inside config.toml for you; mount your own conf directory when you need more than that.

docker pull flashcatcloud/categraf:latestdocker run -d --name categraf -e N9E_HOST=192.168.1.10:17000 flashcatcloud/categraf:latest

Build from source

Go 1.25 or newer. Package the binary together with conf so the agent keeps its sample configuration alongside it.

go buildtar zcvf categraf.tar.gz categraf conf

On macOS use gnu-tar (gtar) — the bundled bsdtar produces an archive Categraf refuses to load.

See what it collects before writing anywhere

Test mode prints metrics straight to stdout, so you can confirm a plugin works before pointing the agent at a server.

./categraf --test./categraf --test --inputs system:mem

Run it

Configuration lives in the conf directory — config.toml for the agent itself, one input.<plugin> directory per collector. Restrict the run to specific plugins with --inputs.

./categraf./categraf --configs /path/to/conf-directory./categraf --inputs system:mem

Connecting it to Nightingale, tagging hosts, and what to check when no data arrives are covered step by step in the docs.

Categraf docs →

Kubernetes

A DaemonSet collects node and cAdvisor metrics, a Deployment covers cluster-level targets such as apiserver and CoreDNS, and a sidecar covers one service. The manifests ship in the repository.

kubectl apply -n monitoring -f k8s/daemonset.yamlkubectl apply -n monitoring -f k8s/deployment.yamlkubectl apply -n monitoring -f k8s/sidecar.yaml

What it collects

Every plugin is a directory under inputs/ with its own sample config. A selection of what ships today:

Operating system

cpu
mem
disk
diskio
net
netstat
processes
procstat
system
kernel
systemd
ntp

Databases

mysql
postgresql
oracle
sqlserver
mongodb
redis
clickhouse
elasticsearch
greenplum
cassandra

Middleware

kafka
rabbitmq
nats
nsq
activemq
zookeeper
consul
ldap

Web and app servers

nginx
apache
tomcat
jboss
weblogic
phpfpm
haproxy
tengine
jenkins
logstash

Containers and Kubernetes

docker
cadvisor
kubernetes
kubelet
kube_state_metrics
kube_proxy

Network gear and hardware

snmp
snmp_trap
switch_legacy
ipmi
redfish
smart
modbus
gnmi
vsphere

GPU

nvidia_smi
dcgm
amd_rocm_smi
hy_smi

Cloud

aliyun
cloudwatch
googlecloud
appdynamics

Probes

ping
http_response
net_response
dns_query
x509_cert
whois

Generic

exec
prometheus
node_exporter
mtail
tpl

Where the data goes

Categraf pushes over the Prometheus Remote Write protocol, so it writes into Nightingale, Prometheus, VictoriaMetrics or anything else that speaks it. Push means the agent opens the connection — which matters when network policy will not let a central server reach into every subnet.

Not the only option

Nightingale does not require Categraf. Telegraf, the Datadog Agent and standard Prometheus exporters write in the same way, and a Prometheus that already scrapes your targets can stay exactly as it is.

Categraf source and releases →