I keep reading about podman, yet it doesm’t FEEL as mature to me as docker for a normal user like me. What’s your opinion? Did you already switch or do you keep waiting for … for what? When will you switch?
well I’ve been using both for quite a while. If you just want something that works, stick with docker. There is nothing wrong with docker in the homelab scenario and podman has rough edges that cringes you. If you are a control freak like me who wants to control every aspect of container running, then podman is a great tool that forces you into the habbit of learning and tinkering. It helped me understand a hell lot of things.
utilize the Systemd integration for container management.
The systemd integration is probably the thing I dislike most about it. 😆 Systemd has no business managing containers IMO, it should manage podman and podman should manage the containers. It’s a completely gratuitous mix of concerns but it seems that podman is set on becoming a systemd subsystem… so I’ll probably never use it.
On a related note, the systemd expansion is getting ridiculous. It’s gotten to the point if you read one day that wayland is being merged into systemd you wouldn’t even know if it’s a joke.
It’s not the “official” way to do it, but you can make systemd run Docker Compose (talking to Podman instead of Docker), which is pretty close to what you’re talking about. And then you don’t have to write stinky systemd INI files for each container.
But you don’t need to have systemd run anything (except docker or podman itself). Just run containers with “restart: always” and docker/podman will start them on boot, restart them of they fail, and leave them alone if they’re manually stopped.
You only need to run compose when you are [re]provisioning a container.
When you used the Podman systemd integration it starts containers on boot just fine. You can even configure it to auto-update containers. Very hassle free.
One of the reasons I use containers instead of installing things directly is that i can completely uninstall a service by deleting a single directory (that contains a compose.yml and any necessary volumes) and running a docker/podman system prune -a
or that i can back up everything by backing up a single “containers” dir, which i could have on a subvolume and snapshot if i wanted to
systemd/quadlet on the other hand makes me throw files in /etc (which is where you’re supposed to put them, but ends up resulting in them being tangled together with base system configuration often partially managed by the package manager)
The Solution™ to this is configuration management like ansible or whatnot, which needlessly overcomplicates things for the use cases i need (though they’re still useful for getting a base system “container ready” wrt ssh hardening and such)
tldr: i want my base system to be separated from my services, and systemd integration is the exact wrong tool for this job
You technically can bind ports <1024 to unprivileged containers. echo 'net.ipv4.ip_unprivileged_port_start=0' | sudo tee /etc/sysctl.d/50-unprivileged-ports.conf; sudo sysctl --system. Though this will allow any user to bind ports below 1024, so it’s not very clean.
Another workaround is to redirect port 80 to 8080 (or other) through iptables and have your proxy listen on this port. sudo iptables -t nat -A PREROUTING -p tcp --dport 80 -j REDIRECT --to-ports 8080. Same thing for 443.
As far as I know granting the CAP_NET_BIND_SERVICE capability to /usr/bin/podman does not work.
Also podman-compose implementation is still incomplete, and I prefer using systemd units to start and manage containers. Check man podman-generate-systemd
You shouldn’t need root to bind to privileged ports. I use k8s but I assume you can set sys capabilities for containers in a similar way I do for pods.
I only tried running rootless when i set them up several years ago and i was completely green, so it was probably me who was the problem.
Regarding podman-compose, Fedora repos has a a package that aliases podman -> docker and the regular docker-compose package, which i used before migrating to podman+systemd. It worked flawlessly unless i did networking shenanigans because Podman and Docker differs (/differed?) in so some thing simply couldn’t be brought over.
Edit: i found the docker-compose and Podman alias thingies in a Fedora Magazine post.
However, unless you use docker-compose a lot for other stuff, learning to use Podmans systemd integration (also called quadlet) is very much worth it.
They’re just a really powerful combo and systemd has a ton of nice features for making stuff run and keep running.
For what it’s worth, I just wrote up a compose.yaml file as I’d write it for Docker Compose and it just worked. See the bottom of my comment on this GitHub issue for an example. I think the team’s intention is for it to transparently support whatever you’d write for a standard Compose file but of course we don’t have things like the brand new Docker watch. They do point to the Compose spec in the Podman Compose README. Bind mounts are good enough for me, thus far.
I tried switching a while back, but I found a bunch of stuff didn’t work properly, and wasn’t considered supported. I don’t remember what it was exactly.
I might try it again once there’s been a bit more development and community use. Docker isn’t ideal, but at least it works and there’s a lot of community support.
I’ve been using podman instead of Docker for a couple years now. I’m not a heavy user, but it doesn’t ever break for me and I appreciate the pods and ease of turning pod config into a kubernetes deployment.
Only thing I miss is proper support for some services I use. Minikube is afaik still a pain with podman, at least rootless. Gitlab runner still doesn’t support podman completely imho. But a plus to docker is that they still build packages for EL 7 while the podman version in EL 7 is pretty damn old. Besides from that I went podman all the way.
It changes packet routing because you’re asking it to map a container port to the host public interface. How else would that occur? And what would be the point in blocking access to it?
Do you want to write routing rules, and keep track of container interfaces, and to grant access manually, for each and every port you expose?
When i have port 8888 not allowed on my firewall then no other program should ‘open’ that port. If i map a port with the syntax ‘8888:8888’ exaclty that happens with docker. Not with podman.
Also this is the default syntax you will find in any guide and docs there is.
To prevent this happening with docker you will have to specify the localhost with ‘127.0.0.1:8888:8888’
When you check the internet for this subject you will notice that this behavior catches a lot of ppl by surpirise.
It catches people by surprise because they don’t have a clue. If someone is binding to localhost just to avoid opening the firewall it means that either they don’t understand how packet routing works, or that they should have used a docker network.
If you use 8888:8888 it means you want the port to be open on the host’s external interface. Which means it needs to be forwarded as well as accessible. Docker does this for you so you don’t have to write the rules by hand, you don’t have to keep track of container interfaces, and you don’t have to remember to take the rules up and down whenever you start or stop the container.
Out of curiosity, how do you do all this with podman? By hand?
Just alone the fact that podman, a drop in replacement for docker, does implement this in another way proofs that it is bad practice to implement the way docker did.
If you use 8888:8888 it means you want the port to be open on the host’s external interface
I am sorry. That is just bs. When i install apache and start the service and let it listen on port 80 and 443 i still have to add the firewall rule to allow it. This is the default behavior of every other programm.
When you install Apache on the host then yes, you just need a rule to allow the port. But a container runs on a separate network interface, which means you also (1) need to know the name of the interfaces involved, which change for each container and their network mode and the host’s network adapter and so on, and (2) set up rules to forward the port traffic between the container interface and the host interface.
I’m guessing you haven’t checked to see what Docker is actually doing, or you’ve looked only at the OUTPUT and INPUT chains.
I’m also guessing that when you stop Apache you leave the ports open in the firewall. Which is bad practice. Docker takes down the rules when the container stops, which ends up as better security.
I sincerely hope you’re not actually running a machine that’s just a firewall away from the Internet, because blaming Docker for bad security in that case is very shortsighted. Docker is actually helping you by making correct and complete firewall rules.
podman, a drop in replacement for docker, does implement this in another way
Technically I do, maybe? My home server is running ProxMox which virtualizes PFSense. My docker install is on a separate VM, but same physical device. Not sure if that counts lol
I fully admit I’m slow sometimes. I could never understand the podman solution for that common scenario when podman runs as one user ID on the host and the image uses a completely different user ID that doesn’t even exist on the host. And no, unfortunately I can’t always go look for a better image, even though I agree that images should be written to allow for ID selection. But they aren’t.
Tried switching some time back, didn’t take long to go back to docker. Podman does not have the polish that docker has taken years to perfect and as much as I love systemd, managing containers in docker is 10x better.
I switched a year ago to podman and had some trouble to get everything running. But it is possible. I’m not running anything rootful and everything works.
Read the docs, use podman-compose (this sadly has no good docs, but works quit well when you got it) and get ready to play around with permissions and file ownership.
My only serious complaint with docker is the quality of their updates. They keep breaking stuff.
If podman supported all docker functionality including compose based stacks, I’d consider switching, but last time I looked it didn’t.
If you’re only on Linux and don’t ever touch containers on Windows or Mac, podman can work fairly well. You need to be comfortable with orchestration tools like k8s to replace compose (or just do a ton of containers) and you can’t use a lot of COTS that has hardcoded dockerisms (localstack, for example, does not work well with podman).
If you have to use Windows or Mac, podman makes life really difficult because you’re running through a VM and it’s just not worth it yet.
Docker on Windows and Mac also runs containers through a VM though. (It’s more obvious on Windows, where you need WSL (powered by a VM) and Hyper-V (a way to run VMs on Windows). But on a Mac, VMs to run Linux are also used to run Docker containers inside the VM.)
For the command line, there’s “podman machine” to abstract away the VM. https://podman.io/docs/installation (installing on macOS is mentioned on that page and Windows has a link to more docs which also uses the podman machine command.)
As for Docker compose, you can use it directly with Podman too: https://www.redhat.com/sysadmin/podman-docker-compose (there’s also podman-compose as well). The only thing Docker compose doesn’t support with Podman is swarm functionality.
You are not logged in. However you can subscribe from another Fediverse account, for example Lemmy or Mastodon. To do this, paste the following into the search field of your instance: !selfhosted@lemmy.world
A place to share alternatives to popular online services that can be self-hosted without giving up privacy or locking you into a service you don’t control.
Rules:
Be civil: we’re here to support and learn from one another. Insults won’t be tolerated. Flame wars are frowned upon.
No spam posting.
Posts have to be centered around self-hosting. There are other communities for discussing hardware or home computing. If it’s not obvious why your post topic revolves around selfhosting, please include details to make it clear.
Don’t duplicate the full text of your blog or github here. Just post the link for folks to click.
Submission headline should match the article title (don’t cherry-pick information from the title to fit your agenda).
Why not try docker rootless? Been using it for 2 years and does everything docker does.
well I’ve been using both for quite a while. If you just want something that works, stick with docker. There is nothing wrong with docker in the homelab scenario and podman has rough edges that cringes you. If you are a control freak like me who wants to control every aspect of container running, then podman is a great tool that forces you into the habbit of learning and tinkering. It helped me understand a hell lot of things.
Podman is solid, just don’t use podman-compose but rather utilize the Systemd integration for container management.
With very few exceptions, it works just as well as Docker these days.
The systemd integration is probably the thing I dislike most about it. 😆 Systemd has no business managing containers IMO, it should manage podman and podman should manage the containers. It’s a completely gratuitous mix of concerns but it seems that podman is set on becoming a systemd subsystem… so I’ll probably never use it.
On a related note, the systemd expansion is getting ridiculous. It’s gotten to the point if you read one day that wayland is being merged into systemd you wouldn’t even know if it’s a joke.
A container is a service, makes perfect sense for me to manage that via Systemd like all other services.
Sure, anything can be a service if you want it to be hard enough. Like the bootloader.
Bet. Give me puppies as a service.
You might want to avoid looking into systemd-homed
It’s not the “official” way to do it, but you can make systemd run Docker Compose (talking to Podman instead of Docker), which is pretty close to what you’re talking about. And then you don’t have to write stinky systemd INI files for each container.
But you don’t need to have systemd run anything (except docker or podman itself). Just run containers with “restart: always” and docker/podman will start them on boot, restart them of they fail, and leave them alone if they’re manually stopped.
You only need to run compose when you are [re]provisioning a container.
Podman does not start your containrs on boot. You need to do some magic yoursefel. Like a cronjob that starts all containers at boot.
When you used the Podman systemd integration it starts containers on boot just fine. You can even configure it to auto-update containers. Very hassle free.
One of the reasons I use containers instead of installing things directly is that i can completely uninstall a service by deleting a single directory (that contains a compose.yml and any necessary volumes) and running a
docker/podman system prune -aor that i can back up everything by backing up a single “containers” dir, which i could have on a subvolume and snapshot if i wanted to
systemd/quadlet on the other hand makes me throw files in /etc (which is where you’re supposed to put them, but ends up resulting in them being tangled together with base system configuration often partially managed by the package manager)
The Solution™ to this is configuration management like ansible or whatnot, which needlessly overcomplicates things for the use cases i need (though they’re still useful for getting a base system “container ready” wrt ssh hardening and such)
tldr: i want my base system to be separated from my services, and systemd integration is the exact wrong tool for this job
Docker ever felt mature to you?
I mean, it was good enough for podman to copy its API and interface verbatim…
I don’t think Docker’s API and CLI are historically where it’s had problems…
Podman is CLI and API compatible with Docker (except where differences in implementation doesn’t allow it)
Running Podman as root is 99.9% the same as running Docker.
I have been running my homelab with Podman for several years and it is absolutely mature enough for a regular user.
Also, the docs are really good.
deleted by creator
You technically can bind ports <1024 to unprivileged containers.
echo 'net.ipv4.ip_unprivileged_port_start=0' | sudo tee /etc/sysctl.d/50-unprivileged-ports.conf; sudo sysctl --system. Though this will allow any user to bind ports below 1024, so it’s not very clean.Another workaround is to redirect port 80 to 8080 (or other) through iptables and have your proxy listen on this port.
sudo iptables -t nat -A PREROUTING -p tcp --dport 80 -j REDIRECT --to-ports 8080. Same thing for 443.As far as I know granting the
CAP_NET_BIND_SERVICEcapability to/usr/bin/podmandoes not work.Also podman-compose implementation is still incomplete, and I prefer using systemd units to start and manage containers. Check
man podman-generate-systemdYou shouldn’t need root to bind to privileged ports. I use k8s but I assume you can set sys capabilities for containers in a similar way I do for pods.
I only tried running rootless when i set them up several years ago and i was completely green, so it was probably me who was the problem.
Regarding podman-compose, Fedora repos has a a package that aliases podman -> docker and the regular docker-compose package, which i used before migrating to podman+systemd. It worked flawlessly unless i did networking shenanigans because Podman and Docker differs (/differed?) in so some thing simply couldn’t be brought over.
Edit: i found the docker-compose and Podman alias thingies in a Fedora Magazine post.
However, unless you use docker-compose a lot for other stuff, learning to use Podmans systemd integration (also called quadlet) is very much worth it. They’re just a really powerful combo and systemd has a ton of nice features for making stuff run and keep running.
For what it’s worth, I just wrote up a
compose.yamlfile as I’d write it for Docker Compose and it just worked. See the bottom of my comment on this GitHub issue for an example. I think the team’s intention is for it to transparently support whatever you’d write for a standard Compose file but of course we don’t have things like the brand new Dockerwatch. They do point to the Compose spec in the Podman Compose README. Bind mounts are good enough for me, thus far.I tried switching a while back, but I found a bunch of stuff didn’t work properly, and wasn’t considered supported. I don’t remember what it was exactly.
I might try it again once there’s been a bit more development and community use. Docker isn’t ideal, but at least it works and there’s a lot of community support.
I’ve been using podman instead of Docker for a couple years now. I’m not a heavy user, but it doesn’t ever break for me and I appreciate the pods and ease of turning pod config into a kubernetes deployment.
Switched a long time ago - it’s just soooo much easier and I never have any problems
Only thing I miss is proper support for some services I use. Minikube is afaik still a pain with podman, at least rootless. Gitlab runner still doesn’t support podman completely imho. But a plus to docker is that they still build packages for EL 7 while the podman version in EL 7 is pretty damn old. Besides from that I went podman all the way.
Podman didnt silently rewrite my firewall rules upon install
10/10 would recommend
I do not even want to know how many databases are openly available because of that shit.
It changes packet routing because you’re asking it to map a container port to the host public interface. How else would that occur? And what would be the point in blocking access to it?
Do you want to write routing rules, and keep track of container interfaces, and to grant access manually, for each and every port you expose?
When i have port 8888 not allowed on my firewall then no other program should ‘open’ that port. If i map a port with the syntax ‘8888:8888’ exaclty that happens with docker. Not with podman.
Also this is the default syntax you will find in any guide and docs there is.
To prevent this happening with docker you will have to specify the localhost with ‘127.0.0.1:8888:8888’
When you check the internet for this subject you will notice that this behavior catches a lot of ppl by surpirise.
It catches people by surprise because they don’t have a clue. If someone is binding to localhost just to avoid opening the firewall it means that either they don’t understand how packet routing works, or that they should have used a docker network.
If you use
8888:8888it means you want the port to be open on the host’s external interface. Which means it needs to be forwarded as well as accessible. Docker does this for you so you don’t have to write the rules by hand, you don’t have to keep track of container interfaces, and you don’t have to remember to take the rules up and down whenever you start or stop the container.Out of curiosity, how do you do all this with podman? By hand?
Just alone the fact that podman, a drop in replacement for docker, does implement this in another way proofs that it is bad practice to implement the way docker did.
I am sorry. That is just bs. When i install apache and start the service and let it listen on port 80 and 443 i still have to add the firewall rule to allow it. This is the default behavior of every other programm.
When you install Apache on the host then yes, you just need a rule to allow the port. But a container runs on a separate network interface, which means you also (1) need to know the name of the interfaces involved, which change for each container and their network mode and the host’s network adapter and so on, and (2) set up rules to forward the port traffic between the container interface and the host interface.
I’m guessing you haven’t checked to see what Docker is actually doing, or you’ve looked only at the OUTPUT and INPUT chains.
I’m also guessing that when you stop Apache you leave the ports open in the firewall. Which is bad practice. Docker takes down the rules when the container stops, which ends up as better security.
I sincerely hope you’re not actually running a machine that’s just a firewall away from the Internet, because blaming Docker for bad security in that case is very shortsighted. Docker is actually helping you by making correct and complete firewall rules.
Yeah? How does it do that?
Who the hell runs Docker on an edge device?
Everyone who runs it on a root server that is not part of some larger private network at that hoster?
Technically I do, maybe? My home server is running ProxMox which virtualizes PFSense. My docker install is on a separate VM, but same physical device. Not sure if that counts lol
That doesn’t count. You’re still externally firewalling it, which is good procedure.
Firewall rules shouldn’t be your only line of defense
It wouldn’t rewrite them if you didn’t have a firewall to begin with.
I fully admit I’m slow sometimes. I could never understand the podman solution for that common scenario when podman runs as one user ID on the host and the image uses a completely different user ID that doesn’t even exist on the host. And no, unfortunately I can’t always go look for a better image, even though I agree that images should be written to allow for ID selection. But they aren’t.
Tried switching some time back, didn’t take long to go back to docker. Podman does not have the polish that docker has taken years to perfect and as much as I love systemd, managing containers in docker is 10x better.
I switched a year ago to podman and had some trouble to get everything running. But it is possible. I’m not running anything rootful and everything works.
Read the docs, use podman-compose (this sadly has no good docs, but works quit well when you got it) and get ready to play around with permissions and file ownership.
My only serious complaint with docker is the quality of their updates. They keep breaking stuff. If podman supported all docker functionality including compose based stacks, I’d consider switching, but last time I looked it didn’t.
There is a wrapper for podman supporting compose.
But maybe it’s time to use kubernetes deployments or pods instead of compose files…
Yeah, the constant Docker breakage was one of the main reasons I switched to Podman. FYI you can use Docker Compose directly with Podman.
If you’re only on Linux and don’t ever touch containers on Windows or Mac, podman can work fairly well. You need to be comfortable with orchestration tools like k8s to replace compose (or just do a ton of containers) and you can’t use a lot of COTS that has hardcoded dockerisms (localstack, for example, does not work well with podman).
If you have to use Windows or Mac, podman makes life really difficult because you’re running through a VM and it’s just not worth it yet.
Docker on Windows and Mac also runs containers through a VM though. (It’s more obvious on Windows, where you need WSL (powered by a VM) and Hyper-V (a way to run VMs on Windows). But on a Mac, VMs to run Linux are also used to run Docker containers inside the VM.)
Podman Desktop helps to abstract VMs away on Windows and macOS: https://podman-desktop.io/
For the command line, there’s “podman machine” to abstract away the VM. https://podman.io/docs/installation (installing on macOS is mentioned on that page and Windows has a link to more docs which also uses the podman machine command.)
As for Docker compose, you can use it directly with Podman too: https://www.redhat.com/sysadmin/podman-docker-compose (there’s also podman-compose as well). The only thing Docker compose doesn’t support with Podman is swarm functionality.
Docker compose can even work with rootless Podman containers on a user account. It requires an environment variable. https://major.io/p/rootless-container-management-with-docker-compose-and-podman/ (it’s basically enabling the socket for podman and using the environment variable to point at the user podman socket)