Will it Cinode...
Contents
My recent experiment around Cinode was about checking how it works in different execution environments. Cinode is written in the Go language so is should be trivial to run it on many CPU architectures and operating systems. Let’s see how it works in practice though. So today I’d like to share the results of my recent test where I tried to run Cinode on Raspberry Pi Zero W and La Frite.
Will it Cinode? Let’s find out.
Raspberry Pi Zero W
My first test subject will be Raspberry Pi Zero W. I assume that this brand doesn’t needs any introduction. It’s a great fit for small DIY projects and small servers so why not to run Cinode on it?
I got this test board as a gift few years ago. It was already used in some of my older projects (you can even see some evidence of poor quality soldering when you take a closer look 😉), went through some serious damage (the board has experienced pretty high voltage thus no I2C working anymore 😆) but finally retired as a member of my Cinode test cluster 😎.
What I wanted to test here is how Cinode deals with relatively low cpu specs. That Raspberry has a single-core 32-bit ARMv6 processor - a perfect target for testing 32-bit architectures which can barely be found on the x86 family these days. It is important because of the difference between the int type in golang which is either 32-bit or 64-bit depending on the target CPU architecture. This may lead to some subtle errors on 32-bit CPUs when int variable is not large enough to handle the result of some arithmetic calculations. Such errors may not be detected when the software is only tested on 64-bit platform.
This board also contains a built-in WiFi (and I’m lucky that it did not burn together with the I2C part during my previous project) - once set up properly it’s enough to just power it on and have it available on the network.
As a distribution I’ve used Raspbian based on debian bullseye:
|
|
La Frite (AML-S805X-AC)
The La Frite board is not that popular as Raspberry Pi one but has some pretty interesting specs too. It includes a 4-core 64-bit ARM Cortex-A53 cpu and thus is much faster than the Raspberry one. It’s an 64-bit ARM architecture thus another interesting case other than the mainstream x86 world.
Because is not as popular as the Raspberry Pi family its early-boot software stack is not that mature and can cause some headaches. I managed to get a recent ubuntu (22.04) running on it using unofficial builds. In the end this board was running debian-like environment similarly to the Raspberry PI so it was very easy to set it up from that point.
I got this board from its Kickstarter campaign few years ago. It managed to gather some dust already but I knew it would be a perfect Cinode cluster addition once I realized it should still be somewhere around 😉. It lacks a built-in WiFi but has 100Mb RJ45 port instead - that’s more than enough for my tests.
You can still buy it for around 20$ on the loveRPI page so it looks like a great (and still cheap) alternative for Raspberries. And if you’re interested, checkout other boards from the same producer.
|
|
Cinode docker images
It took me some time and a lot of github actions minutes but there’s now a nice CI for building and releasing multi-arch Cinode docker images. Those images are not only built in the CI pipeline but are also tested on different CPU architectures in the same CI workflow. For CPUs other than x86 which are not natively supported on github runners I’m using Qemu emulation. You can read some more details on how it works in my blog post about it.
There was one issue building docker images for the Raspberry board due to the the armv6 architecture. I’m using Google’s distroless images as a base layer for the final Cinode images. That way those images are kept small and yet still contain all that’s needed to run go binaries (ca certificates, timezone info etc.). And it turns out that distroless images do not support this CPU.
Cinode binaries are statically linked and they don’t require any OS-level libraries though. For that reason, we can use the static variant of the base image from any platform because it is CPU agnostic:
|
|
Bring it on
Enough talking about specs, let’s see how our two boards impersonate as Cinode workers.
I’ll be using docker to run the official github-baked container images there thus the first thing to get into our boards is the docker runtime. It’s pretty straightforward on debian-based distributions due to an official deb repository that supports all the CPU architectures we want.
Here’s the run trace from the Raspberry Pi board:
|
|
The last log line outputs some information about the Cinode binary itself, and we can clearly see that the test above was running on an arm architecture.
Ok, so it did start, but did it Cinode? Let’s try it out.
So here’s the Cinode blog viewed through our almighty little fella, Raspberry Pi Zero W 🎉:
First thing I have noticed when running with that small board is that the docker itself is very slow. Starting up the docker image, even if it is already downloaded, takes like 15 to 20 seconds. I believe this is mainly limited by the speed of the attached SD card that suddenly has to digest huge amount of write requests. It could also be the result of running on a low-spec 32-bit CPU. But once up and running, everything remains in the memory thus the speed is not an issue anymore.
Let’s bring the second star onto the stage - La Frite. The setup and run command is almost identical:
|
|
This one proudly shows the arm64 architecture. Also docker startup times are really good compared to the previous one. This may be caused by either a faster, multi-core CPU or it could also be a better and faster USB storage. Didn’t drill that down though.
Let’s see how it works then, here’s the blog page:
Jay, no issues running Cinode 👍. CI pipeline with its rigorous unit and docker image tests did its job.
What’s to come?
I already have few ideas on how to use our boards effectively. Still too early to go into more details but it’s time to start building a bit larger Cinode clusters to show what it can be used for.
I also wanted to do some tests of the RISC-V architecture. There are some nice small boards with this CPU, e.g. MangoPI Pro. Those may be a bit challenging. RISC-V CPUs are still not that popular so the software will obviously be more problematic. But let’s wait for one and try it out, I already ordered the board 😁.
Author BYO
LastMod 2023-03-29