Chameth.com - posts like docker-automatic-nginx-proxy, reverse-engineering-arctis-pro-wireless-headset, understanding-docker-volume-mounts, why-you-should-be-using-https but not modern-css-is-fun

The Case of the Unchanging Config

2026-04-09T00:00:00Z

Last week I was attempting to make it so I could share pictures on IRC directly from my client. This sounds simple, but it involves a bouncer¹ that proxies the request to a standalone image hosting service that I had to modify to be compatible. At one point my testing loop was:

Make a change to the hosting service
Commit it
Tag a new release
Wait for it to build
Update the version I’m running on my server
Reconfigure the bouncer to pass the new parameter or change the URL or whatever
Try uploading a photo from my phone
Realise I’ve overlooked something and go back to step one

I could have set up a local copy of everything and tested it sensibly. I should have set up a local copy of everything and tested it sensibly. But it seemed like such a trivial change, and setting up the whole environment seemed like such a pain. After the third or so iteration of failure I was pretty annoyed with myself, computers, and basically everything.

My biggest annoyance was that my bouncer would not pick up the new URL from the config when I changed it. It’s meant to reload the config when it receives a SIGHUP, and it claimed to in the logs, but I could clearly see it was still hitting the old URL. Restarting the bouncer to update the config is a pain, as it disconnects me from all the IRC networks, and has to reconnect to them all, reauthenticate, etc. It also mildly spams everyone who shares a channel with me².

When I finally got everything working I had a look at the bouncer source, and thought I’d spotted the issue. I raised a bug report, ending in this remark:

It looks like the config is reloaded properly, but the handler for uploads is created once at startup and has its own copy of the uploader, so effectively snapshots the config to whatever it is at startup:

fileUploadHandler := http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
    cfg := srv.Config()
    h := fileupload.Handler{
        Uploader:    cfg.FileUploader,
        DB:          db,
        Auth:        cfg.Auth,
        HTTPOrigins: cfg.HTTPOrigins,
    }
    h.ServeHTTP(w, r)
})

I’m not quite sure what I thought I saw there, but that’s not right. I want to call it a hallucination, but that feels like a very overloaded word these days. I guess I was just seeing what I wanted to see, and that was a good excuse to stop investigating. Still, even if I misidentified the cause, the bug was still present, right? … Right?

The Science of Deduction

When the developer got back and said he couldn’t reproduce my issue, I went back to my install and immediately reproduced it. With a calmer head, I figured it was probably something with my particular set up. First thing to check: am I actually running the version I think I am?

Unfortunately the bouncer doesn’t actually seem to expose the version anywhere that I can see. It’s not in the logs, it doesn’t have a -version flag, and none of the IRC-based status commands seem to include it. But I know it’s a Go app, and I know Go embeds the version information. go version -m <binary> will dump it all out, but the binary is inside a Docker image, and the Docker image is one of my nice, minimal, artisanal ones so doesn’t ship a go binary. No problem, docker compose cp bouncer:/bnc ./bnc yoinks the binary out of the container, and then dumping the version shows that, yes, I am running the version I thought I was. Hmm.

The next tool I reached for in my toolbox was strace. Maybe it’s not actually reading the file for some reason? I immediately executed strace -p <pid> -e trace=openat,open,read,pread64 -f without having to look any part of that up. Yep. Definitely. Then I edited the config, HUP’d the bouncer, and saw that it was… reading the config file. As it’s meant to. By default strace truncates strings to 32 bytes, so I couldn’t actually see the line I’d changed. Some more definitely-not-RTFMing later, and rerunning it with an extra -s 65536 let me see the full config. Surprise! The config hadn’t changed!

To confirm my findings, I used docker compose cp again, this time yoinking the config file from inside the container. The inside config file was definitely different to the outside config file. What? My hypothesis at this point was “something something Docker nonsense”. I mount the config as read-only, and was wondering if that meant that Docker was doing something weird instead of just bind mounting it. A quick trip to /proc/<pid>/mounts showed that it was, in fact, not doing anything weird, and was just bind mounting it.

If the file is bind mounted, then surely it’s the same file? I ran stat on the file on the host, noted the inode number, then pondered how to actually run stat on the file inside the container, given the aforementioned awkwardly minimal image. The solution was easy: access it via /proc/<pid>/root/. I could’ve saved myself a bunch of docker compose cp if I’d thought about that earlier. Oh well. The inode of that file was different. What?

The answer was DNS. It’s always DNS. Oh, sorry, force of habit. I meant the answer was vim. I was editing the config in vim, and when it saves files, by default, it writes the new content to a temporary file and does an atomic rename. That’s normally a good thing: it prevents corruption if the write fails midway through³. In this case, though, that means the new file has a new inode. Bind mounting a file binds to the inode, so the container just keeps clutching onto the original config from when it was started, blissfully unaware that the party has relocated down the street.

This problem is likely to happen whenever you bind mount a file into a container. When I mentioned this to a friend, he immediately responded “oh yeah, never do that”, and went on to describe the horrible hacks he’s had to add to Ansible to sidestep the issue. The nicer solution is to just bind mount an entire directory if you can, as then it doesn’t matter what happens to the files within it. I really like having the config files sat alongside the Docker compose files, though; having to create a directory just to work around some bind mount weirdness upsets me.

Now I knew what the problem was, I found there was an issue raised against Docker twelve years ago⁴. The first response was “that’s expected”, and I fully agree with the author: “respectfully, that might be expected by you, but it was not expected by me”!⁵ As a result of the issue they added a nice note to the docs:

Note: Many tools used to edit files including vi and sed --in-place may result in an inode change. Since Docker v1.1.0, this will produce an error such as “sed: cannot rename ./sedKdJ9Dy: Device or resource busy”. In the case where you want to edit the mounted file, it is often easiest to instead mount the parent directory.

But in the intervening twelve years, both the note and the functionality described have gone missing. Ho-hum.

I found it interesting how I’ve only just hit this problem, given how long I’ve used Docker. But I realised that almost everything I run I’m happy to just restart. Cattle, not pets, and so on. My IRC bouncer is one of the few exceptions to that. The only other thing I regularly hot reloaded was Centauri, my reverse proxy, but that had a whole config directory mounted because it was shared between containers, so nicely sidestepped the foot-gun.

So lesson learnt: check for weird bind mount issues before raising issues about config hot reloading. It’ll join the esteemed company of “maybe the drive is full and causing completely unrelated problems?”, “perhaps everything is dog slow because the kernel ran out of entropy?”, and “did systemd sneakily take over that functionality while you weren’t looking?” in the troubleshooting checklist.

An IRC bouncer is basically an always-on proxy. It connects to the IRC networks for you, then your clients connect to your bouncer. The bouncer can then send incoming messages to all your different clients, cache them when you’re offline and replay them later, and lots of other nice things people take for granted in their chat apps these days. ↩︎
Join/part/quit spam is part of IRC, and clients have ways of handling it, but I still feel bad about doing it excessively. ↩︎
You can disable this in vim by setting backupcopy=yes. Good luck redoing all the debugging if you ever accidentally remove that from your vimrc, though! ↩︎
Which is weird, because Docker can’t possibly be that old. That would make me much older than I’m prepared to accept. ↩︎
If you read the thread they didn’t actually mean to sound so dismissive, but it’s still pretty funny. “Yes, it’s expected that the foot-gun causes your foot to hurt. Duh. What did you expect?” ↩︎

An interesting Tailscale + Docker gotcha

2026-01-29T00:00:00Z

As I’ve written about before, I use Tailscale for a lot of things. I thought I had it set up in a reasonably secure manner, but I recently noticed a problem.

I use Tailscale’s ACLs to limit what each node can access, based on the tags I apply to it. So an app node can’t access anything via Tailscale, while an integration or server node can access things tagged with either app or integration. This is expressed pretty simply in the Tailscale ACL JSON:

{
	"tagOwners": {
		// Servers that can be SSH'd into
		"tag:server": [],
		// Applications that are exposed on tailscale but never connect out
		"tag:app": [],
		// Things which talk to other services over tailscale (connecting to apps etc)
		"tag:integration": [],
	},
	"grants": [
		{
			// Users can access everything
			"src": ["autogroup:member"],
			"dst": ["*"],
			"ip":  ["*"],
		},
		{
			// Servers and integrations can access integrations and apps
			"src": ["tag:server", "tag:integration"],
			"dst": ["tag:integration", "tag:app"],
			"ip":  ["*"],
		},
	],
}

Over the past week I’ve been setting up a private Forgejo instance behind Tailscale, complete with an actions runner that runs things using a docker-in-docker container. I didn’t want the runner knowing anything about Tailscale, so I had it configured to speak to Forgejo direct over HTTP (forgejo:3000) instead of using full Tailscale HTTPS URL that I use when accessing it (http://git.example-net.ts.net/)¹.

Everything was going fine, until I forgot to do that translation… and it worked. My Forgejo action runners could access anything on my tailnet. I run several things on Tailscale that just have authentication turned off, on the basis that only authorised devices can access them. Things like the admin interface for this website. I definitely didn’t intend for any workflow I run on my git server to have access to edit my website!

This wasn’t just limited to Forgejo, either. Any docker container I was running could access the tailnet. In hindsight it’s fairly obvious why: the host is running Tailscale, connected as a node tagged with my server tag. That creates a tailscale0 interface, and automagically sets up iptables rules to route Tailscale traffic over the interface. Docker also automagically sets up iptables rules to bridge traffic, and apparently these two sets of rules interact in such a way that traffic from Docker containers is allowed to route via the tailscale0 interface.

I say it’s fairly obvious in hindsight — there’s no reason why Docker would special case any particular host interface after all — but it still feels pretty surprising. Because both bits of software inject their own iptables rules, I never really had a good mental model for how they interact. The host Tailscale node was a completely separate building block to Docker. It would be a pain to use either of them if they didn’t do these rules, but it’s also one of the reasons I don’t really like “magical” things².

There are lots of ways to fix this, but none of them feel particularly great. You can configure both Tailscale and Docker to not automatically fiddle with iptables and handle the rules yourself, but I really hate dealing with iptables³. As a stopgap I did hold my nose and add some iptables rules to drop traffic to the tailscale0 interface if it originated from the IP ranges that Docker was configured to use:

iptables -I DOCKER-USER -s 192.168.0.0/16 -o tailscale0 -j DROP
iptables -I DOCKER-USER -s 172.17.0.0/12 -o tailscale0 -j DROP

The DOCKER-USER chain is a nice little escape hatch; it comes before the main auto-generated DOCKER chain, and Docker leaves the rules in it alone.

So I added these rules, and felt pretty good about myself, and then… everything started breaking in weird ways. After some debugging I realised the problem was DNS⁴. The server’s DNS resolver is 100.100.100.100, a special Tailscale address. This is so that it can resolve tailnet hostnames via MagicDNS (and so I don’t have to configure my custom DNS servers manually on each device; Tailscale does it for me). My new iptables rules inadvertently dropped all the DNS packets coming from docker containers. D’oh.

Obviously the solution here is to double down and add MORE iptables rules:

iptables -I DOCKER-USER -s 192.168.0.0/16 -d 100.100.100.100 -o tailscale0 -j ACCEPT
iptables -I DOCKER-USER -s 172.17.0.0/12 -d 100.100.100.100 -o tailscale0 -j ACCEPT

This got DNS working again, but it all feels a bit gross. Not least because to actually make these persist you need to use iptables-save, and then you also get all the rules that both Docker and Tailscale have inserted. I came across a script to filter out the Docker ones, but… yuck⁵.

I realised a cleaner idea might just be to lock down what server tagged nodes could access within Tailscale itself. That way I can avoid touching iptables at all. It’s handy to be able to curl services from the host when debugging, but it’s not really necessary. So I removed that access, and… stuff broke again. I’m using my Forgejo instance as a registry for some of the docker images I run, so the docker daemon needs to be able to reach it. I ended up making a new tag for infrastructure, which can be accessed from server devices. This does still allow all the docker containers to reach Forgejo, but I already have it set up with appropriate access controls and public/private repository splits. Forgejo is a service designed to run publicly, so this seems a reasonable trade-off for convenience. I used the tests feature of Tailscale’s ACL config to make sure I’d got the rules right:

{
	"tests": [
		{
			// Servers can only access infrastructure
			"src":   "tag:server",
			"proto": "tcp",
			"allow": [
				"tag:infrastructure:8080",
			],
			"deny": [
				"100.84.16.43:8080",
				"me@example.com:8080",
				"tag:server:8080",
				"tag:app:8080",
				"tag:integration:8080",
			],
		},
	],
}

The whole situation still feels a bit messy. If I ever get around to switching to nftables I might loop back and manually craft some rules for routing traffic, instead of leaving Tailscale and Docker to do their own thing.

I discovered later on that logging into the container registry at forgejo:3000 actually issued a redirect to https://git.example-net.ts.net/ so this was all basically for naught… ↩︎
I didn’t use Tailscale’s MagicDNS for a long time just because the word “magic” put me off. Only when I eventually got around to learning how it worked, and seeing that it wasn’t really that magical under the hood, did I change my mind. ↩︎
nftables seems far better in lots of ways, but I can’t really be bothered migrating. Maybe next time I reimage the server for whatever reason… ↩︎
It’s always DNS… ↩︎
Again, nftables would almost certainly help here. It actually has (gasp) configuration files. But again, I really didn’t want to spend the time migrating. ↩︎

Docker reverse proxying, redux

2022-12-06T00:00:00Z

Six years ago, I described my system for configuring a reverse proxy for docker containers. It involved six containers including a key-value store and a webserver. Nothing in that system has persisted to this day. Don’t get me wrong – it worked – but there were a lot of rough edges and areas for improvement.

Microservices and their limitations

My goal was to follow the UNIX philosophy of “do one thing and do it well”. Unfortunately, that doesn’t really work when applied to network services that have to interact with one and other. UNIX tools are built upon a common file system and simple data passed over STDIN. Microservices don’t have that shared foundation. You could make one: companies that use microservices in anger often have a team that deals with the “developer experience” of creating and using microservices. But as a solo developer that’s not something I wanted to spend my time doing.

This became very apparent when trying to debug issues. In the UNIX world, if your series of commands piped together does something strange you can simply echo the data at various stages. Not so much when that data is flying around on a network, going into and out of things like etcd. Trying to figure out why a certificate hadn’t been acquired was a case of searching through logs from four containers, none of which had particularly good logging. There are many ways to get insight into what’s happening with microservices but, again, that’s not something I wanted to do myself.

Over time, and with experience in other projects, I came to realise that microservices only really make sense if you’re unable to deploy a monolith. For tech companies this naturally happens when different teams contribute to the same service: splitting it out into smaller services that are wholly owned by individual teams makes sense. For solo developers, that never happens. You can still gain the other benefits of microservices – such as code separation and having clearly defined APIs – by sticking to certain coding standards.

Proxy inconveniences

As well as being unhappy with the microservice nature of the solution, I wasn’t pleased with nginx. If you requested an unknown domain, nginx would use the first server block in its config to serve a response, instead of sending an “unrecognised name” alert as I wanted. It was a minor issue, but it irked me.

So from nginx I switched to haproxy. It has a strict-sni option when configuring TLS connections which makes it behave properly. It also performs a lot better for this type of workload than nginx. All was well for a while, but then I started getting alerts that requests were occasionally failing. I couldn’t reproduce the issue, but my nightly jobs to build and push containers managed to hit it nearly every night, causing them to fail.

After some investigation, I found that the haproxy developers had refactored the header parsing code, and neglected to properly reset flags when multiple requests were sent over the same connection. There was a patch, but it wasn’t released. No problem, I thought, I’ll just cherry-pick it onto the last release… Except that haproxy use Git in the most convoluted manner I’ve ever seen – they have one repository per release. This makes it harder to patch, but it also made me question whether I trusted them to ship stable software: there were no tests for the header parsing code (which is both fundamental and finicky, the perfect target for tests), the source code management was weird, and they didn’t seem in any rush to patch this bug.

Not long after that issue, Greg managed to encounter another bug where haproxy returned a 500 error whenever the upstream server replied with a particular, perfectly valid, header. The die was cast – it was time to move to something else.

Not Invented Here syndrome

Looking for a new solution, there were many more options than back in 2016. I’m still convinced, however, that anything exposed to the Internet should not have access to run docker containers. It’s the modern equivalent of running a CGI script as root. That single requirement eliminates most off-the-shelf solutions. What do you do when nothing quite meets your specific requirements? You make something yourself! My new solution has two components: Dotege and Centauri.

Dotege is a replacement for the microservices that monitored containers and obtained certificates. It’s fundamentally a templating engine - whenever the containers change, it evaluates a template and saves the result to disk. The template has access to details about the containers, their labels, ports, and so on. Dotege can also obtain certificates from Let’s Encrypt, and raise a signal against another process whenever the template or certificates change. I used this to generate the configuration and certificates used by haproxy for a while, and more recently changed the template so that it works for Centauri.

Centauri is my own reverse proxy. It’s configured using a simple text file and can also obtain certificates from an ACME provider. It doesn’t serve static content, has no knowledge about docker, and avoids the other bells and whistles that adorn most reverse proxies. It also has good test coverage to ensure that I don’t, say, accidentally break header parsing.

As a software engineer I enjoy writing software, but I also enjoy running simple, easy to understand software. That’s what I’ve achieved here: it’s very easy to identify where the problem is if anything goes wrong, both are small Go programs rather than vast sprawling C monstrosities, and their interaction is primarily through a file written to disk that can be inspected or edited as needed.

Reverse engineering an Arctis Pro Wireless Headset

2021-06-12T00:00:00Z

The Arctis Pro Wireless Headset.

For the last year and a bit, I’ve been using a SteelSeries Arctis Pro Wireless Headset for gaming and talking to friends. It’s a fine headset, but because there’s an always-on receiver there’s no way to detect if the headset is turned on or not from the desktop.

Whenever I start using the headset, I set my desktop’s sound to go to the headset, and then when I stop using the headset I set it to go back to speakers. It doesn’t take more than a second, but some days I might put the headset on a dozen times as I’m on calls, or if it’s noisy outside, etc. That means it’s probably worth at least a few hours of my time trying to automate it.

At first, I hoped I’d be able to tell from the state of the USB device whether there was a headset connected but nothing at all changed when flipping it on and off. Then I went hunting for existing open source tools that might work with it and found that while people have reverse engineered many of the older Arctis headsets, no one has done the same for the Pro Wireless. I finished off with a search to see if anyone had documented the wire protocol even if there was no nice open source software to go with it; I came up short there, too. Looks like I’d have to do it myself.

Capturing data with WireShark

The headset exposes a Human Interface Device (HID), and the wire protocols for earlier versions of the Arctis series looked to be very simple messages passed over the HID connection. It should therefore be fairly easy to use WireShark to capture the data being sent and received by the official SteelSeries application¹.

After setting up WireShark and making it record only the headset’s HID connection, it quickly became apparent that the software was sending three different requests each second, over and over again:

Wireshark, showing a packet capture of a SET_REPORT request from the host to the headset

The packets we’re interested in are the SET_REPORT Request frames sent from the host to the device. Wireshark understands the HID protocol, so it nicely shows us the raw HID data; in the screenshot this is 0x40AA (ignoring the trailing zero bytes). The other requests sent immediately after have 0x41AA and 0x42AA payloads — clearly the first byte is indicating which piece of data is being requested.

The responses to these requests come back in an URB_INTERRUPT in frame:

Wireshark, showing a packet capture of the headset’s response to the earlier request

So the answer to our first request appears to be 0x04. The second response is 0x0402 and the third is 0x04 again. The first thing I tried doing was popping the spare battery out of the charger. The response to the 0x42AA request changed from 0x04 to 0x00 - that’s certainly clear enough! After a bit of waiting around, the battery in my headset dropped from full on the display to three bars, and at that point the 0x40AA response dropped from 0x04 to 0x03. The display on the receiver shows battery state as four bars, and it appears the wire protocol directly corresponds to that particular representation.

That left the 0x41AA request as an unknown. I tried everything I could think of, but it stubbornly kept returning 0x0402. I enlisted a friend who has the same headset to run some hacky Go code and report his values, and he also got a 0x0402 response. As I explained that I couldn’t figure out what these values are, he reported back that turning his headset off made the response change to 0x0202. In all my testing, I’d forgotten to try turning the headset off! That’s the one thing I was actually trying to detect, as well. Thanks, Simon, for helping me get past that bit of stupidity!

I still don’t know what the second byte of the response is, or whether there are other values than 0x04 for on and 0x02 for off, but I’m happy enough to label it as “device status” and move on.

Exploring other features

The software allows you to tweak a bunch of different settings:

SteelSeries Good Game software, showing the advanced headset settings

I went through each one and fiddled with all the values, recording the requests in WireShark as I did so. It turns out the protocol is very simplistic - the wire protocol directly corresponds to the UI elements in the software (or on the receiver, if you navigate through its menus). For example, the software allows you to set the “Headset auto shutoff” value in increments of 10 minutes. You might usually expect this to be converted to seconds or something similar before being passed to the device, but on the wire it’s actually sent as 0x00 for off, 0x01 for 10 minutes, up through to 0x0C for the maximum of 120 minutes.

All the dropdown options seem to function this way — if you pick the 6th option then the request payload will be an 0x06 byte. The sliders have fixed positions they snap to and function similarly: for the two brightness sliders pictured they snap to 11 positions and on the wire these range from 0x00 for off to 0x0A for the maximum.

At this point I was well past my goal of being able to detect whether the headset was on or off, and I was now just trying to see if I could figure out enough of the protocol that I could reimplement the control software on Linux if I ever wanted to. Just as I was about to close the app, something caught my eye: there are integrations with games and other software that can display information on the receiver’s OLED display!

There’s an official API for this, but it involves sending JSON to a webserver which runs as part of their app². That doesn’t feel very nice, and definitely won’t work for me on Linux, so I started a new WireShark session and took some captures while I spoke to myself on Discord.

Decoding pixels

The frames sent to the device whenever Discord showed a notification had a 1060 byte payload. The display on the receiver is 140 pixels wide³ and each pixel can only seem to be on or off, so I figured each bit in the payload corresponded to one bit of the output. I exported the data to binary in the hope that I could visually see what was going on - in theory if I line wrap the data at 140 characters it should look vaguely like the final output. Unfortunately, it did not. There was roughly the right number of high bits, but no amount of fiddling with them in a text editor could give me a coherent picture.

Instead, I wrote some code to write different values to the device. Starting with a payload of all zeroes and gradually increasing a bit at a time every second. After the first few seconds, I saw a line of pixels being drawn downwards on the left of the screen; had I just got the axes the wrong way around? After the first eight pixels lit up, though, they jumped over to the next column. The actual addressing scheme looks something like this:

Diagram of how pixels are addressed on the receiver’s OLED display

So the first 140 bytes given the pixels for the first 8 rows, the next 140 bytes fill in the 8 rows below that, and so on and so forth. Armed with this information I wrote a simple program to read an image and output it to the display:

Headset receiver displaying a custom Hello World message

At this point I was thinking about trying to get Doom rendering on the screen, but I couldn’t find anything nicely hackable that would let me grab the output and pass it on to the receiver. Instead, I decided to try a GIF decoder and before very long had a nice little animated display:

Protocol reference

All the HID messages have a single byte that determines the message type, then an 0xAA byte, then any payload required by the command. These are the ones I’ve figured out:

Byte	Command	Payload
`0x09`	Save changes	None.
`0x10`	Request firmware version(?)	None.
`0x27`	Volume limiter	`0x00` for off, `0x01` for on.
`0x2E`	Equalizer preset	ID of the equalizer preset to use.
`0x39`	Sidetone level	`0x00` for lowest to `0x09` for highest.
`0x3C`	Set headset timeout	Timeout, as a number of 10 minutes. `0x00` for off to `0x0C` for 120 mins.
`0x3E`	Mic mute LED brightness	`0x00` for lowest to `0x0A` for highest.
`0x40`	Request headset battery	None.
`0x41`	Request device status	None.
`0x42`	Request receiver battery	None.
`0x51`	Surround sound mode	`0x00` for off, `0x01` for on.
`0x62`	Auto-start Bluetooth	`0x00` for off, `0x01` for on.
`0x63`	Auto-mute game audio during calls	`0x00` for off, `0x01` for on.
`0x85`	OLED brightness	`0x00` for lowest to `0x0A` for highest.
`0x83`	Equalizer	(Not yet decoded)
`0x89`	Screensaver mode	`0x00` to dim, `0x01` for off, `0x02` for screensaver.
`0xD2`	Render image	Pixel array as described above.

Some of these I’ve not dug too much into because they didn’t seem very interesting. If you have an Arctis Pro Wireless and figure anything more out, let me know, and I’ll update the list.

If you just want to check device state like I originally did, I’ve contributed support for this headset to the excellent HeadsetControl project. It should be in the next release.

Unfortunately (as you’d expect) their application only runs on Windows, so this process involved an annoying amount of rebooting to Windows, fleeing back to Linux, and then realising I hadn’t actually recorded enough to figure it out and repeating. ↩︎
You can also send lisp to the webserver, and it will execute it. No comment. ↩︎
I know this not because it’s mentioned in the technical specs (it’s not), but because I took a photo and counted them out one by one. ↩︎

An introduction to containers

2020-03-01T00:00:00Z

So. Many. Containers.

I’m a huge fan of (software) containers. Most people I know fall in to one of two camps: either they also use, and are fans of, containers, or they haven’t yet really figured them out and view them as some kind of voodoo that they don’t really want or need.

I’m writing this short guide to explain a little how containers work - and how running something in a container isn’t really that much different to running it normally - to hopefully enable more people in that second group to give them a try. It’s aimed at people who have a fairly good grasp of how Linux works.

Containers are often mentioned in the same breath as VMs, which is not a helpful comparison or analogy. Think of containers as standard units of software, much like how Intermodal containers are standard units of freight transport across the world. When a company internationally ships goods in volume there isn’t a question about how they’re packaged - they go in an intermodal container. The same container can be deployed on a freight train, a lorry, or a ship. The haulage company doesn’t need to care what’s in the container because they’re completely standardised. Likewise, with software containers you don’t really need to care about what’s inside: the software you’re deploying could be written in Go, Python2, Python3, Bash, PHP, LOLCODE, or anything¹.

What does a running container look like?

When you run a container, you’re just running a process. In a lot of ways it’s not very different to what happens if you were to start the same process on the host computer.

For example I have a container that just runs cat(1). With no arguments, cat will read from stdin until it receives an EOF, so it’s handy to test with. If I run ps a on my computer, I can see the cat process in amongst everything else I’m currently running:

    PID TTY      STAT   TIME COMMAND
   7199 pts/1    Ss     0:01 /usr/bin/zsh
 323806 pts/0    Ss+    0:00 /bin/cat
 324120 pts/4    R+     0:00 ps a

The /bin/cat process is in a container, and the ps a underneath it is just running like normal on my desktop. They look very similar, right? If I look under /proc/323806 I can see all the usual attributes, the same as any other process running.

If I run ps in a container, though, it’s a different story:

    PID TTY      STAT   TIME COMMAND
      1 pts/0    Rs+    0:00 ps a

So inside the container it looks like there’s only one process running. It can’t see anything running “outside” on my desktop. The secret here is that this isn’t a special container trick: this is just a feature of the Linux kernel called namespacing.

If we go back to procfs and look at the ns/pid node we can see the process in the container is in a separate PID (process ID) namespace to the one on my desktop:

# readlink /proc/323806/ns/pid  
pid:[4026534564]
# readlink /proc/7199/ns/pid
pid:[4026531836]

Almost all² processes running ’normally’ on my desktop have the same PID namespace, whereas each container gets their own by default. PID namespaces are hierarchical: a new process is assigned a PID in its own namespace, and the parent namespace, and the grandparent namespace, and so on. That’s why I can see the process running in the container from my normal shell - the container’s namespace is a child of the main namespace all of my desktop software is running in.

Linux supports - and container software makes use of - a bunch of other namespaces too: mount points, network, UTS³, cgroups, and more. These all play a part in isolating a container from the system it is running on.

You can manually run a process with unshare(1) to “unshare” some namespaces from the parent process. For example if I run unshare -fp --mount-proc ps a, it looks very similar to running ps instead the container:

    PID TTY      STAT   TIME COMMAND
      1 pts/4    R+     0:00 ps a

So: a process running inside a container is just a heavily namespaced process running otherwise normally in the operating system. No voodoo magic here!

What about the filesystem? What are ‘images’?

Containers run in their own mount namespace meaning mount points can be different inside the container to those on the host. This means the container can have a different / mounted to the host, effectively giving it its own filesystem.

The root filesystem of the container is defined in the container’s image. If I use Docker to run a container using the Ubuntu image (docker run ubuntu), the root filesystem inside that container will resemble a minimal ubuntu install. Note that this is just the filesystem: the container doesn’t have its own kernel.

You might be thinking that sounds pretty inefficient. Downloading Ubuntu is definitely not instant, and doing it for every application you run would be insane! Quite. Containers solve this by using filesystem layers. These are stacked on top of one another to create the final filesystem. Each layer can be retrieved and cached independently of all others.

Say (for simplicity) that the Ubuntu image is a single layer. If I run one container with that image, then the layer will be downloaded and cached once. If I run three hundred containers with that image then the layer will be downloaded and cached once. Even better, if I use another image that’s based on Ubuntu but adds some software on top, only the “on top” layer will be downloaded if I already have the relevant “Ubuntu” layer cached.

If all the layers are cached, what happens when you change a file? This is dealt with using the copy-on-write technique: when you modify a file it is copied from the source layer and the changes are only made in a new layer. This is handled by the OverlayFS filesystem which is part of the mainline kernel.

When a container is running, changes made to its filesystem are temporary, and do not persist across container restarts. To persist data - or introduce new data to a container - you can mount volumes. How this works varies depending on how you’re running your container, but at the basic level it is pretty much the same as bind-mounting (mount -o bind).

You may be familiar with using chroots to change the apparent root directory of processes, perhaps with full-blown “jails” built on top. Containers offer much better isolation thanks to the use of namespaces. Instead of being constrained to a portion of the host’s filesystem, they don’t even have it mounted! Containers also get to specify their environment - if they expect in certain places, for example - instead of the sysadmin having to manually set up the chroot. Finally, containers offer much more fine-grained control over what processes can do (if you want it), and allow much more advanced use-cases such as inter-container networking.

Images and filesystems employ a little magic to ensure that layers are reusable and cacheable, but again there’s nothing terribly special about them: a container has a filesystem that appears to it to work the same way as a filesystem on the host, and it’s using a standard filesystem shipped with the kernel.

How about networking?

Again, networking is namespaced, so a container has its own network stack, its own virtual network interface, its own IP address and so on. How that network interacts with your real network depends on how you’re running the container. Docker, for example, can add iptables rules to NAT traffic between containers' networks and the outside world.

Containers can generally be connected into networks, and can communicate amongst themselves without the traffic actually leaving the host machine. This allows you to, for example, run a SQL database and connect it to a web application without ever exposing the database to the outside world. Moreover, as well as being isolated from the outside world, it’s isolated from other containers in other networks. If one of your applications has a crazy bug or is compromised, this significantly limits the damage it can do.

You have to explicitly opt in to “publishing” ports from a container, which exposes them to the outside world (either directly, or via a load balancer or some other middle-man, depending on how you’re running the container). This means you can pick and chose how the outside world sees the app you’re deploying: if it’s a web service that listens on both port 443 and port 80, you can chose to only expose the encrypted port.

If you run some containers and create some networks, you can see the interfaces and bridges on the host:

$ ip l
...
6: br-2405a8cc0445: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP mode DEFAULT group default 
    link/ether 02:42:3e:fa:23:62 brd ff:ff:ff:ff:ff:ff
8: veth8ed0735@if7: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue master br-2405a8cc0445 state UP mode DEFAULT group default 
    link/ether b2:c1:5d:55:26:9b brd ff:ff:ff:ff:ff:ff link-netnsid 2
10: veth541d84b@if9: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue master br-9d7bc4024c1a state UP mode DEFAULT group default 
    link/ether 86:aa:5f:ee:da:1a brd ff:ff:ff:ff:ff:ff link-netnsid 1

This shows a bridge and two virtual NICs, just the same as if you’d manually created them (albeit with seemingly random names). So: as before, nothing special here.

Docker? Compose? K8s? Floccinaucinihilipilification?

(OK, Floccinaucinihilipilification isn’t actually a container technology, as far as I’m aware.)

Docker is the most popular container engine - that is, the bit of software that actually runs containers. It’s responsible for setting up all those namespaces we found, downloading and caching image layers, and actually starting and stopping the processes. Docker runs as a system-wide daemon - when you run a command like docker run ubuntu it actually just instructs the daemon to do the work.

There are several alternatives to Docker for running containers; one interesting one is Podman which runs containers without a daemon. Container engines have all standardised around the same image format looked after by the Open Container Initiative, so an image you build in Docker can be used in Podman, or pretty much any other engine.

Docker compose is a tool for defining and running multiple-container applications. I mentioned earlier running a database alongside a webapp - in practice to do this you’re going to have to configure a network for them, configure a mount point for the database to persist its data on, pass credentials in to both the database and the application, and so on. Doing all that by hand is tedious and error prone.

Docker compose lets you write “compose files”, which are simple yaml descriptions of the containers you wish to run, their properties, and details about any volumes or networks you may want. Out of the box, docker-compose will create a default network for each compose file you run so the containers within it can communicate.

Kubernetes is a container orchestrator, designed to automate deployment and management of large numbers of containers. It works with Docker under the hood, but provides a huge amount of tooling on top to allow you to deploy applications and manage their dependencies. It runs across multiple physical (or virtual) machines (while still allowing containers to communicate privately), and can support massive workloads by scaling out services (running multiple copies of a container on different hosts) and load balancing. Kubernetes is sometimes shorted to k8s (as in K - 8 elided letters - s) because computer people don’t like long words.

OK, they make sense now. But why bother?

Hopefully if you’ve read this far you’ve already picked up on some of the potential benefits, but this is my personal list:

Isolation. If I run some software in a container, there is very little it can do to upset me. It’s not going to leave bits of itself all over my filesystem, it can’t steal all of the secrets in my home directory, I can even limit its CPU and memory resources if I want. If I decide to stop running it, I just delete the container and it is completely gone: no trace remains.

Ease of use. If you give me a container image I have a very good idea of how to run it already. I might need to do some minor configuration to expose ports or mount volumes, but there’s no question about how to run it, how to make it automatically start, and there’s no “installation” procedure. If I want to then swap it with an alternative (say, move from MySQL to MariaDB), it’s potentially just a case of changing the name of the image I pull.

Dependencies included. Containers just run. Python 2 software includes Python 2 and just work. Python 3 software includes Python 3 and just work. I don’t have a massive headache trying to run both at the same time, because they take care of their own messes. Similarly I’m not going to have to install npm or cargo or composer to pull in dependencies for an application: that’s going to have been done in the build process.

Reproducibility. As a fallout from having dependencies included and being isolated from everything else, containers give you amazing reproducibility. If it “works on your machine” in a container, it’ll almost certainly work in production because it’s the exact same environment.

Standardisation. At the start of this article I called containers standard units of software. One of my favourite advantages of containers is that you basically get an API to list all the software you’re running. Most container engines let you supply labels attached to containers as well, so you can add your own annotations. I use this to annotate services which expose HTTP endpoints, and I have a tool that automatically generates SSL certificates for them and configures haproxy to route traffic to them. I can’t imagine how I’d do this without containers - I imagine it’d involve a lot of manual work.

Hopefully this has helped demystify containers a little. If you feel like I’ve missed something important out, or I’ve left you more confused than when you started, feel free to drop me a note using the feedback form below.

OK, maybe you should care if you’re deploying something written in crazy languages like PHP. ↩︎
Some multi-process apps, such as web browsers, are starting to use namespaces to enhance security, as do certain package systems like Flatpak ↩︎
“Unix timesharing system”; in practice this means having a separate hostname ↩︎