5.6 KiB
title, date, author, draft, weight, tags, description, canonicalURL, cover, showtoc, searchHidden
| title | date | author | draft | weight | tags | description | canonicalURL | cover | showtoc | searchHidden | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Building a Docker Swarm With Pi's | 2021-12-12T17:23:32Z | ahmza | true |
|
Pis are very capable computers, when banded together they're almost unstoppable | https://blog.ahmza.com/Page%28/posts/building-a-docker-swarm-with-pis.md%29 |
|
true | false |
Introduction
The Raspberry Pi from it's first iteration has been a great hit. Adopted by many as the computer of choice for projects, and for very good reason. The Pi is a fully fledged computer that fits in your pocket and the most recent edition has blown fans away with the flagship model boasting an ARM64 CPU and 8GB of RAM.
Now one Raspberry Pi 4 8GB is more than powerful enough for most workloads and in some cases totally unnecessary, so you may even get away with a single 4GB model. However taking into consideration their relatively low price and power consumption why not go all out and build a cluster.
You can never have too many Raspberry Pis, just not enough Ideas
My excuse to build a Pi cluster is, I would like to migrate the services currently running on a bare metal server that sucks a lot of power to a low power solution. What better way to do that than a Pi cluster.
My Setup
- 4 x Raspberry Pi 4 8GB
- 4 x 32GB SD cards
- 4 x PoE hats
- 1 x 5 port PoE Switch
Goal of this post
Hopefully after following this post you will have setup Docker swarm on a cluster of Pis - stay even if you currently only have one - and understand how to bend it to your will 🦾.
Why Docker Swarm
While we can manage each docker node - Raspberry Pi - individually , Docker swarm enables us to manage all the devices from a single main node making managing the cluster a lot easier.
Think of Docker swarm as a set of 🤖 hands which will run tasks on each individual node for us. These tasks can consist of:
- Cluster Management
- Deployment and Scaling across the cluster
- Multi-host networking
- load balancing
- etc...
Well why not Kubernetes
Although Kubernetes is awesome, it adds a bunch of complexity that
is simply not needed in a home lab environment where we only have
a small cluster. As well as that, from my experience creating
docker-compose files is a lot easier than writing manifests.
Docker swarm can do everything I want it to and requires less
effort which is great if you're like me 🥱 😂
Required Stuff
The links below are reccomendations/stuff I used
- Raspberry Pi 4B (Reccomend 4-8 @ 8GB)
- Raspberry Pi PoE Hats (If you can find the PoE+ version get them)
- Cluster Case (Get any case you want)
- PoE switch (Get one that has enough ports for the nuber of pis you want)
- RJ45 Ethernet Cable (Reccomend short cables as it will look neater)
- Micro SD cards (Get enough cards for the number of Pis, Get the biggest cards you can afford)
This is without a doubt an expensive project, however in my eyes it's the cost of another server 😂 and the best part about this project is you can add to the cluster over time spreding the cost which is how I managed to purchase all the hardware. A suggestion would be to buy equipment that allows you room to grow so an 8 port PoE switch instead of a 4 port if you know you will scale to 8 nodes
If you build this then you can safely say another server is not needed. This cluster should be more than capable to serve all your home needs even if you only have 2 nodes. Put it this way, if you have 4 8GB nodes then you have a combined total of:
- 32GB of RAM
- 16 physical CPU cores @ 1.5GHz
Money Saving Tips
- Just get 1-2 nodes to start
- cut PoE and use normal power cables
- Use Raspberry Pis with less memmory eg 4GB
- Grab a really cheap case or make one using these
- Use smaller SD Cards try not to go below 32GB (Reccomended 64GB If money saved elsewhere)
BUILD TIME !!
Image the SD Cards
So interesting fact the CPUs used in the raspberry pi 4 use an ARM64 archetecture meaning they're 64bit. When looking for images of Raspbian you will only find 32bit variants which will work however will not make use of all the hardware which is just a waste. 64bit images exist however are well hidden, Dont fear 😨 you can download the latest images from the raspberrypi.org index
To write the cards I use the Raspberry Pi imager available here. once that is installed and you have the 64bit raspbian image downloaded do the following:
- Click CHOOSE OS
- Click Use custom
- Select the 64bit rasbian image and open
- Click CHOOSE STORAGE
- Select storage device
- Click WRITE
We dont need to unzip the file as the tool is clever enough to handle that for us
repeat steps for number of cards you have