A (multi-arch) multitool for container/network testing and troubleshooting. The main docker image is based on Alpine Linux. There is a Fedora variant to be used in environments which require the image to be based only on RedHat Linux, or any of it's derivatives.
The container image contains lots of tools, as well as a nginx
web server, which listens on port 80
and 443
by default. The web server helps to run this container-image in a straight-forward way, so you can simply exec
into the container and use various tools.
Earlier, I created and maintained this tool with Henrik, as praqma/network-multitool
. Praqma was bought by another company, and now the "Praqma" brand is being dismantled. This means the network-multitool's git and docker repositories must go. It was decided by the current representatives of the company to hand it over to me so I can continue maintaining it. So, apart from a small change in the repository name, nothing has changed.
The docker repository to pull this image is now: https://hub.docker.com/r/wbitt/network-multitool
Or:
docker pull wbitt/network-multitool
- linux/386
- linux/amd64
- linux/arm/v7
- linux/arm64
- https://hub.docker.com/r/wbitt/network-multitool (An automated multi-arch build)
- latest, minimal, alpine-minimal ( The main/default 'minimal' image - Alpine based )
- extra, alpine-extra (Alpine based image - with extra tools )
- openshift , openshift-minimal (openshift compatible - minimal) - Ports: 1180, 11443
- openshift-extra (openshift compatible with extra tools) - Ports: 1180, 11443
- fedora, fedora-minimal ( 'Minimal' Fedora based image )
Openshift is very strict about how a container image should run. So, the openshift variant of the multitool has the following limitations / changes:
- Runs as non-root ; which means some tools (e.g.
traceroute
,tcptraceroute
, etc, will not work) - Listens on ports
1180
and11443
- not80
and443
- Some executable files are manually set as
setuid
, so those tools remain usable. Tools set withsetuid
are:- apk
- arping
- busybox
- mii-tool
- tcpdump
- tcptraceroute
- traceroute
- tshark
Remember, this multitool is purely a troubleshooting tool, and should be used as such. It is not designed to abuse openshift (or any system's) security, nor should it be used to do so.
- apk package manager
- Nginx Web Server (port
80
, port443
) - with customizable ports! - awk, cut, diff, find, grep, sed, vi editor, wc
- curl, wget
- dig, nslookup
- ip, ifconfig, route
- traceroute, tracepath, mtr, tcptraceroute (for layer 4 packet tracing)
- ping, arp, arping
- ps, netstat
- gzip, cpio, tar
- telnet client
- tcpdump
- jq
- bash
Size: 16 MB compressed, 38 MB uncompressed
All tools from "minimal", plus:
- iperf3
- ethtool, mii-tool, route
- nmap
- ss
- tshark
- ssh client, lftp client, rsync, scp
- netcat (nc), socat
- ApacheBench (ab)
- mysql & postgresql client
- git
Size: 64 MB compressed, 220 MB uncompressed
- YUM package manager
- Nginx Web Server (port 80, port 443) - customizable ports!
- wget, curl
- dig, nslookup
- ip, ifconfig, route, traceroute, tracepath, mtr
- ping, arp, arping
- ps, netstat
- gzip, cpio, tar
- telnet client
- awk, cut, diff, find, grep, sed, vi editor, wc
- jq
/bin/sh
shell interpreter - not/bin/bash
Size: 72 MB uncompressed
Note: The SSL certificates are generated for "localhost", are self signed, and placed in /certs/
directory. During your testing, ignore the certificate warning/error. While using curl, you can use -k
to ignore SSL certificate warnings/errors.
$ docker run -d wbitt/network-multitool
Then:
$ docker exec -it container-name /bin/bash
Create single pod - without a deployment:
$ kubectl run multitool --image=wbitt/network-multitool
Create a deployment:
$ kubectl create deployment multitool --image=wbitt/network-multitool
Then:
$ kubectl exec -it pod-name /bin/bash
Note: You can pass additional parameter --namespace=<your-desired-namespace>
to the above kubectl commands.
$ oc new-project test-project-1
$ oc new-app wbitt/network-multitool:openshift --name multitool-openshift
$ oc status
$ oc get pods
$ oc logs pod-name
$ oc exec -it pod-name /bin/sh
$ oc port-forward pod-name 1180:1180 11443:11443
Sometimes you want to do testing using the host network. This can be achieved by running the multitool using host networking.
$ docker run --network host -d wbitt/network-multitool
Note: If port 80 and/or 443 are already busy on the host, then use pass the extra arguments to multitool, so it can listen on a different port, as shown below:
$ docker run --network host -e HTTP_PORT=1180 -e HTTPS_PORT=11443 -d wbitt/network-multitool
For Kubernetes, there is YAML/manifest file multitool-daemonset.yaml
in the kubernetes
directory, that will run an instance of the multitool on all hosts in the cluster using host networking.
$ kubectl apply -f kubernetes/multitool-daemonset.yaml
Notes:
- You can pass additional parameter
--namespace=<your-desired-namespace>
to the above kubectl command. - Due to a possibility of something (some service) already listening on port 80 and 443 on the worker nodes, the
daemonset
is configured to run multitool on port1180
and11443
. You can change this in the YAML file if you want.
There are times when one may want to join this (multitool) container to another container's IP namespace for troubleshooting, or on the host network. This is true for both Docker and Kubernetes platforms. During that time if the container in question is a web server (nginx, apache, etc), or a reverse-proxy (traefik, nginx, haproxy, etc), then network-multitool cannot join it in the same IP namespace on Docker, and similarly it cannot join the same pod on Kubernetes. This happens because network multitool also runs a web server on port 80 (and 443), and this results in port conflict on the same IP address. To help in this sort of troubleshooting, there are two environment variables HTTP_PORT and HTTPS_PORT , which you can use to provide the values of your choice instead of 80 and 443. When the container starts, it uses the values provided by you/user to listen for incoming connections. Below is an example:
$ docker run -e HTTP_PORT=1180 -e HTTPS_PORT=11443 \
-p 1180:1180 -p 11443:11443 -d local/network-multitool
4636efd4660c2436b3089ab1a979e5ce3ae23055f9ca5dc9ffbab508f28dfa2a
$ docker ps
CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
4636efd4660c local/network-multitool "/docker-entrypoint.…" 4 seconds ago Up 3 seconds 80/tcp, 0.0.0.0:1180->1180/tcp, 443/tcp, 0.0.0.0:11443->11443/tcp recursing_nobel
6e8b6ed8bfa6 nginx "nginx -g 'daemon of…" 56 minutes ago Up 56 minutes 80/tcp nginx
$ curl http://localhost:1180
Praqma Network MultiTool (with NGINX) - 4636efd4660c - 172.17.0.3/16 - HTTP: 1180 , HTTPS: 11443
$ curl -k https://localhost:11443
Praqma Network MultiTool (with NGINX) - 4636efd4660c - 172.17.0.3/16 - HTTP: 1180 , HTTPS: 11443
If these environment variables are absent/not-provided, the container will listen on normal/default ports 80 and 443.
Well, normally, if a container does not run a daemon/service, then running it (the container) involves using creative ways / hacks to keep it alive. If you don't want to suddenly start browsing the internet for "those creative ways", then it is best to run a small web server in the container - as the default process.
This helps you when you are using Docker. You simply execute:
$ docker run -d wbitt/network-multitool
This also helps when you are using kubernetes. You simply execute:
$ kubectl run multitool --image=wbitt/network-multitool
The multitool container starts as web server - so it remains UP
. Then, you simply connect to it using:
$ docker exec -it some-silly-container-name /bin/sh
Or, on Kubernetes:
$ kubectl exec -it multitool-3822887632-pwlr1 -- /bin/sh
This is why it is good to have a web-server in this tool. Hope this answers the question! Besides, I believe that having a web server in a multitool is like having yet another tool! Personally, I think this is cool! Henrik thinks the same!
We have tried to put in all the most commonly used tools, while keeping it small and practical. We can't have all the tools under the sun, otherwise it will end up as something like this.
However, if you have a special need, for a special tool, for your special use-case, then I would recommend to simply build your own docker image using this one as base image, and expanding it with the tools you need.
There is absolutely no need to use LetsEncrypt. This is a testing tool, and validity of SSL certificates does not matter.
One could argue that it is possible to simply install the tools on the hosts and get over with it. However, we should keep the infrastructure immutable and not install anything on the hosts. Ideally we should never ssh
to our cluster worker nodes. Some of the reasons are:
- It is generally cumbersome to install the tools since they might be needed on several hosts.
- New packages may conflict with existing packages, and may break some functionality of the host.
- Removing the tools and dependencies after use could be difficult, as it may break some functionality of the host.
- By using a
daemonset
, it makes it easier to integrate with other resources. e.g. Use volumes for packet capture files, etc. - Using the
daemonset
provides a 'cloud native' approach to provision debugging/testing tools. - You can
exec
into thedaemonset
, without needing to SSH into the node.
Contributions are welcome for packages/tools considered "absolutely necessary", of "core" nature, are "minimal" in size, and "have large number of use-cases". Remember, the goal is not to create yet another Linux distribution! :)