Dockers manipulation of nftables is pretty well defined in their documentation

Documentation people don’t read. People expect, that, like most other services, docker binds to ports/addresses behind the firewall. Literally no other container runtime/engine does this, including, notably, podman.

As to the usage of the docker socket that is widely advised against unless you really know what you’re doing.

Too bad people don’t read that advice. They just deploy the webtop docker compose, without understanding what any of it is. I like (hate?) linuxserver’s webtop, because it’s an example of the two of the worst footguns in docker in one

To include the rest of my comment that I linked to:

Do any of those poor saps on zoomeye expect that I can pwn them by literally opening a webpage?

No. They expect their firewall to protect them by not allowing remote traffic to those ports. You can argue semantics all you want, but not informing people of this gives them another footgun to shoot themselves with. Hence, docker “bypasses” the firewall.

On the other hand, podman respects your firewall rules. Yes, you have to edit the rules yourself. But that’s better than a footgun. The literal point of a firewall is to ensure that any services you accidentally have running aren’t exposed to the internet, and docker throws that out the window.

You originally stated:

I think from the dev’s point of view (not that it is right or wrong), this is intended behavior simply because if docker didn’t do this, they would get 1,000 issues opened per day of people saying containers don’t work when they forgot to add a firewall rules for a new container.

And I’m trying to say that even if that was true, it would still be better than a footgun where people expose stuff that’s not supposed to be exposed.

But that isn’t the case for podman. A quick look through the github issues for podman, and I don’t see it inundated with newbies asking “how to expose services?” because they assume the firewall port needs to be opened, probably. Instead, there are bug reports in the opposite direction, like this one, where services are being exposed despite the firewall being up.

(I don’t have anything against you, I just really hate the way docker does things.)

Probably not an issue, but you should check. If the port opened is something like 127.0.0.1:portnumber, then it’s only bound to localhost, and only that local machine can access it. If no address is specified, then anyone with access to the server can access that service.

An easy way to see containers running is: docker ps, where you can look at forwarded ports.

Alternatively, you can use the nmap tool to scan your own server for exposed ports. nmap -A serverip does the slowest, but most indepth scan.

Yes it is a security risk, but if you don’t have all ports forwarded, someone would still have to breach your internal network IIRC, so you would have many many more problems than docker.

I think from the dev’s point of view (not that it is right or wrong), this is intended behavior simply because if docker didn’t do this, they would get 1,000 issues opened per day of people saying containers don’t work when they forgot to add a firewall rules for a new container.

My problem with this, is that when running a public facing server, this ends up with people exposing containers that really, really shouldn’t be exposed.

Excerpt from another comment of mine:

It’s only docker where you have to deal with something like this:

---
services:
  webtop:
    image: lscr.io/linuxserver/webtop:latest
    container_name: webtop
    security_opt:
      - seccomp:unconfined #optional
    environment:
      - PUID=1000
      - PGID=1000
      - TZ=Etc/UTC
      - SUBFOLDER=/ #optional
      - TITLE=Webtop #optional
    volumes:
      - /path/to/data:/config
      - /var/run/docker.sock:/var/run/docker.sock #optional
    ports:
      - 3000:3000
      - 3001:3001
    restart: unless-stopped

Originally from here, edited for brevity.

Resulting in exposed services. Feel free to look at shodan or zoomeye, internet connected search engines, for exposed versions of this service. This service is highly dangerous to expose, as it gives people an in to your system via the docker socket.

If you need public access:

https://github.com/anderspitman/awesome-tunneling

From this list, I use rathole. One rathole container runs on my vps, and another runs on my home server, and it exposes my reverse proxy (caddy), to the public.

I’m not too well versed in rustdesk, but it seems that they use end to end encryption (is it good? Idk).

https://github.com/rustdesk/rustdesk/discussions/2239#discussioncomment-5647075

I have experience with a similar software that uses relays, syncthing. With syncthing, everything is e2ee, so there’s no concern about whether or not the relay’s are trustworthy, and you can even host your own public relay server.

I find it hard to believe that rustdesk, another relay based software, wouldn’t have a similar architecture.

edit: typo

Nothing that is more questionable than lxd, which now requires a contributor license agreement, allowing canonical to not open source their hosted versions, despite lxd being agpl.

Thankfully, it’s been forked as incus, and debian is encouraging users to migrate.

But yeah. They haven’t said what makes proxmox’s license questionable.

Someone recommended ssh, which is good, but it can’t do udp connections.

https://github.com/anderspitman/awesome-tunneling

From this list, I selected rathole since they claimed to be more performant than frp, the most popular solution.

Don’t do unattended upgrades. Neither host nor containers. Do blind or automated updates if you want but check up on them and be ready to roll back if something is wrong.

Those issues are only common on rolling releases. On stable distros, they put tape between breaking changes, test that tape, and then roll out updates.

Debian, and many other distros support it officially: https://wiki.debian.org/UnattendedUpgrades. It’s not just a cronjob running “apt install”, but an actual process, including automated checks. You can configure it to not upgrade specific packages, or stick to security updates.

As for containers, it is trivial to rollback versions, which is why unattended upgrades are ok. Although, if data or configuration is corrupted by a bug, then you probably would have to restore from backup (probably something I should have suggested in my initial reply).

It should be noted that unattended upgrade doesn’t always mean “upgrade to the latest version”. For docker/podman containers, you can pin them to a stable release, and then it will do unattended upgrades within that release, preventing any major breaking changes.

Similarly, on many distros, you can configure them to only do the minimum security updates, while leaving other packages untouched.

People should use what distro they know best. A rolling distro they know how to handle is much better than a non-rolling one they don’t.

I don’t really feel like reinstalling the bootloader over ssh, to a machine that doesn’t have a monitor, but you do you. There are real significant differences between stable and rolling release distros, that make a stable release more suited for a server, especially one you don’t want to baby remotely.

I use arch. But the only reason I can afford to baby a rolling release distro is because I have two laptops (both running arch). I can feel confident that if one breaks, I can use the other. All my data is replicated to each laptop, and backed up to a remote server running syncthing, so I can even reinstall and not lose anything. But I still panicked when I saw that message suggesting that I should reinstall grub.

That remote server? Ubuntu with unattended upgrades, by the way. Most VPS providers will give you a linux distro image with unattended security upgrades enabled because it removes a footgun from the customer. On Contabo with Rocky 9, it even seems to do automatic reboots. This ensures that their customers don’t have insecure, outdated binaries or libraries.

Docker doesn’t “bypass” the firewall. It manages rules so the ports that you pass to host will work. Because there’s no point in mapping blocked ports. You want to add and remove firewall rules by hand every time a container starts or stops, and look up container interfaces yourself? Be my guest.

Docker is a way for me to run services on my server. Literally every other service application respects the firewall. Sometimes I want services to be exposed on my home network, but not on a public wifi, something docker isn’t capable of doing, but the firewall is. Sometimes I may want to configure a service while keeping it running. Or maybe I want to test it locally. Or maybe I want to use it locally

It’s only docker where you have to deal with something like this:

---
services:
  webtop:
    image: lscr.io/linuxserver/webtop:latest
    container_name: webtop
    security_opt:
      - seccomp:unconfined #optional
    environment:
      - PUID=1000
      - PGID=1000
      - TZ=Etc/UTC
      - SUBFOLDER=/ #optional
      - TITLE=Webtop #optional
    volumes:
      - /path/to/data:/config
      - /var/run/docker.sock:/var/run/docker.sock #optional
    ports:
      - 3000:3000
      - 3001:3001
    restart: unless-stopped

Originally from here, edited for brevity.

Resulting in exposed services. Feel free to look at shodan or zoomeye, internet connected search engines, for exposed versions of this service. This service is highly dangerous to expose, as it gives people an in to your system via the docker socket.

Do any of those poor saps on zoomeye expect that I can pwn them by literally opening a webpage?

No. They expect their firewall to protect them by not allowing remote traffic to those ports. You can argue semantics all you want, but not informing people of this gives them another footgun to shoot themselves with. Hence, docker “bypasses” the firewall.

On the other hand, podman respects your firewall rules. Yes, you have to edit the rules yourself. But that’s better than a footgun. The literal point of a firewall is to ensure that any services you accidentally have running aren’t exposed to the internet, and docker throws that out the window.

A tip I have is to move away from manjaro.

When you use a rolling release, you lose one of the main features of stable release distros: Automatic, unattended upgrades. AFAIK, every stable release distro has those, and none of the rolling releases do (except maybe opensuses’s new slowroll and centos rolling, but I wouldn’t recommend or use them).

Manjaro has other issues too, but that’s the big one.

Although I use arch on my laptop, I run debian on my server because I don’t want to have to baby it, especially since I primarily access it remotely. Automatic upgrades are one less complication removed, allowing me to focus on my server itself.

As for application deployment itself, I recommend using application containers, either via docker or podman. There are many premade containers for those platforms, for apps like jellyfin, or the various music streaming apps people use to replace spotify (I can’t remember any of the top of my head, but I know you have lots of options).

However, there are two caveats to docker (not podman) people should know:

• Docker containers don’t auto update. Although you can use something like watchtower to automatically update them. As for podman, podman has an auto update command you can probably configure to run regularly.
• Docker bypasses your firewall. If you forward port 80, docker will go around the firewall and publish it. The reason for this is that most linux firewalls work by using iptables or nftables behind the hood, but docker also edits those directly… this has security implications, I’ve seen many container services people didn’t intend to put on the public internet, on there.

Podman, however, respects your firewall rules. Podman isn’t perfect though, there are some apps that won’t run in podman containers, although my use case is a little more niche (greenbone service and vulnerability scanner).

As for where to start, projects like linuxserver provide podman/docker containers, which you can use to deploy many apps fairly easily, once you learn how to launch apps with the compose file. Check out this nextcloud dockerized, they provide. Nextcloud is a google drive alternative, although sometimes people complain about it being slow… I don’t know about the quality of linuxserver’s nextcloud, so you’d have to do some research for that, and find a good docker container.

your typical manga/light novel weebo

No chinese support :(

I read a ton of web novels translated from Chinese, and reading the untranslated versions would be a fun way to learn Chinese. Or Korean.

I don’t really like the Japanese light novels as much.

Edit: hmmm, it seems like their are similar projects, and some have custom language support. I may need to look into those into the future.

The tldr as I understand it is that Mac M1/M2 devices are unique in that the vram (gpu ram) is the same as the normal ram. This sharing allows LLM models to run on the gpu of those chips, and in their “vram” as well, allowing you to run bigger models on smaller devices.

Llama.cpp was the software that users did this with originalky. I can’t find the original guide/article I looked at, but here is a github gist, where the commenters have done benchmarks:

https://gist.github.com/cedrickchee/e8d4cb0c4b1df6cc47ce8b18457ebde0

If I run two mysql containers, it won’t necessarily take twice the resources of a single mysql containers

It’s complicated, but essentially, no.

Docker images, are built in layers. Each layer is a step in the build process. Layers that are identical, are shared between containers to the point of it taking up the ram of only running the layer once.

Although, it should be noted that docker doesn’t load the whole container into memory, like a normal linux os. Unused stuff will just sit on your disk, just like normal. So rather, binaries or libraries loaded twice via two docker containers will only use up the ram of one instance. This is similar to how shared libraries reduce ram usage.

Docker only has these features, deduplication, if you are using overlayfs or aufs, but I think overlayfs is the default.

https://moonpiedumplings.github.io/projects/setting-up-kasm/#turns-out-memory-deduplication-is-on-by-default-for-docker-containers

Should you run more than one database container? Well I dunno how mysql scales. If there is performance benefit from having only one mysqld instance, then it’s probably worth it. Like, if mysql uses up that much ram regardless of what databases you have loaded in a way that can’t be deduplicated, then you’d definitely see a benefit from a single container.

What if your services need different database versions, or even software? Then different database containers is probably better.

Nginx and nginx proxy manager are two different things, although nginx proxy manager uses nginx underneath the hodd.

Nginx is a lightweight reverse proxy and http(s) server configured via config files.

https://nginx.org/en/

Nginx proxy manager is a docker container that runs nginx, but also had a webui on top of it to make it much, much easier to configure.

Sometimes abbreviated as NPM.

https://nginxproxymanager.com/

That’s why people keep asking you for your nginx config since when you just say nginx, people are expecting that you are using just nginx, and configuring it through text files.

I heard obfsproxy

Yeah, tor obs4 bridges.

But somehow, my high school managed to block those. My high school was literally more locked down than the great firewall of China.

I set up: https://github.com/cognetwork-dev/Metallic

At first, then I eventually switched to https://github.com/v2ray/v2ray-core as metallic struggled on some things. Both v2ray and xray are built for the great firewall of China, and iirc, they use the same tech.

It’s not too fast though. That privacy comes at a price. This may be the slowest proxy/vpn out there (although it’s speedy enough for normal web browsing), whereas wireguard is the fastest. Maybe you want something in between? It depends on your threat model.

You can just run it from your local computer. I did that because I wanted it to be available offline.

Somewhat related, there is a site I follow called royalroad. Royalroad is a site for web serials, which are basically books uploaded to the internet chapter by chapter.

Although royalroad used to be only google ads, at some point they started accepting user submitted ads. (Also, ads on that site have always been unobtrusive).

I like these ads much better because they are more privacy respecting (literally an a image and a link).

Also, they are really funny. User’s with no art skills will make memes, or doodle stick figures, and I clicked on that one anyways, and the story was soooo good.

You want webtop: https://docs.linuxserver.io/images/docker-webtop

But just like with kasm, not all software will work, although I think most will.

About kasm:

Not really. I don’t thing the default kasm images come with sudo or a root password, so you cant “sudo apt” or the like.

If you do create a software image with sudo, them yes, but only for a single session, if you keep it long running. Every time you destroy the session completely it will be reset.

Although, If you need software in your images, it’s better to just build your own docker images for use with kasm, that have everything you want.

Also relevant: https://wiki.archlinux.org/title/Internet_sharing

Important to note from that article: docker (the “docker” one, but not podman) edits iptables rules so you have to run different iptables commands if you want it to work right.

Can you elaborate on what you found lacking in kasm? Because afaik, kasm is one of the best solutions for this, giving you a full desktop session inside a docker container.

The chances I am going to manage a linux distro without systemd are low, but some systems (arch for example) don’t have cron out of the box.

Not that big of a deal since it’s easy to translate them all, but that’s one of the reasons why I default to systemd/timer units.