I have a love/hate relationship with docker. On one side it’s convenient to have a single line start for your services. On the other side as a self-hoster it made some developers rely only on docker meaning that deploying the stack from source is just an undocumented mess.
Also following the log4j vulnerability I tend to prioritize building from source as some docker package were updated far later than the source code was.
The Dockerfile is essentially the instructions for deploying from scratch. Sure, they most likely only exist for one distro but adapting isn’t a huge chore.
You can also clone the repo and build the container yourself. If you want to update say, log4j, and then attempt to build it, that’s still entirely possible and easier than from scratch considering the build environment is consistent.
If I’m updating the source code already I might as well build my service from it, I really don’t see how building a docker container afterward makes it easier considering the update can also break compatibility with the docker environment.
Also adapting can be a pita when the package is built around a really specific environment. Like if I see that the dockerfile installs a MySQL database can I instead connect it to my PostgreSQL database or is it completely not compatible? That’s not really something the dockerfile would tell me.
I really don’t see how building a docker container afterward makes it easier
What it’s supposed to make easier is both sandboxing and reuse / deployment. For example, Docker + Traefik makes some tasks so incredibly easy and secure compared to running them on bare metal. Or if you need to spin up multiple instances, they can be created and destroyed in seconds. Without the container, this just isn’t feasible.
The dockerfile uses MySQL because it works. If you want to know if the core service works with PostgreSQL, that’s not really on the guy who wrote the dockerfile, that’s on the application maintainer. Read the docs, do some testing, create your own container using its own PostgreSQL or connecting to an external database if that suits your needs better.
Once again the flexibility of bind mounts means you could often drop that external database right on top of the one in the container. That’s the real beauty of Docker IMO, being able to slot the containers into your system seamlessly due to the mount system.
adapting can be a pita when the package is built around a really specific environment
That’s the great thing about Docker, it lets you bring that really specific environment anywhere and in an incredibly lightweight manner compared to the old days of heavyweight VMs. I’ve even got Docker containers running on a Raspberry Pi B+ that otherwise is so old that it would be nearly impossible to install the libraries required to run modern software.
I love Docker because it is the only sane method to selfhost shit with my Synology NAS, and I love my Synology NAS because it is the only Linux interaction that I have (from my old MacBook Pro).
Also I created this repo to create a reproducible sec environment for myself. I added other languages, but personally work mostly with python. It is basically resonating for handling all the boiler plate:
For packaging in docker I started to use nix2container project as it gives me a greater control over layers. So for example when I package my phyton app I typically use 3 layers:
python and it’s dependencies
my application dependencies
my application, which is very tiny compared to other two, so there is great reuse of the layers
The algorithm mentioned in the video also helps a lot with reuse, but the above is more optimized by frequency of how things typically change.
BTW: today I discovered this https://github.com/astro/microvm.nix I haven’t play with it yet, but in theory it would let me generate a microvm image (in similar fashion to generate a docker container) which would let me to run my app natively as a tiny VM on EC2 for example, and use only minimum necessary of a typical OS to run it.
Literally why docker was invented
I have a love/hate relationship with docker. On one side it’s convenient to have a single line start for your services. On the other side as a self-hoster it made some developers rely only on docker meaning that deploying the stack from source is just an undocumented mess.
Also following the log4j vulnerability I tend to prioritize building from source as some docker package were updated far later than the source code was.
The Dockerfile is essentially the instructions for deploying from scratch. Sure, they most likely only exist for one distro but adapting isn’t a huge chore.
You can also clone the repo and build the container yourself. If you want to update say, log4j, and then attempt to build it, that’s still entirely possible and easier than from scratch considering the build environment is consistent.
If I’m updating the source code already I might as well build my service from it, I really don’t see how building a docker container afterward makes it easier considering the update can also break compatibility with the docker environment.
Also adapting can be a pita when the package is built around a really specific environment. Like if I see that the dockerfile installs a MySQL database can I instead connect it to my PostgreSQL database or is it completely not compatible? That’s not really something the dockerfile would tell me.
What it’s supposed to make easier is both sandboxing and reuse / deployment. For example, Docker + Traefik makes some tasks so incredibly easy and secure compared to running them on bare metal. Or if you need to spin up multiple instances, they can be created and destroyed in seconds. Without the container, this just isn’t feasible.
The dockerfile uses MySQL because it works. If you want to know if the core service works with PostgreSQL, that’s not really on the guy who wrote the dockerfile, that’s on the application maintainer. Read the docs, do some testing, create your own container using its own PostgreSQL or connecting to an external database if that suits your needs better.
Once again the flexibility of bind mounts means you could often drop that external database right on top of the one in the container. That’s the real beauty of Docker IMO, being able to slot the containers into your system seamlessly due to the mount system.
That’s the great thing about Docker, it lets you bring that really specific environment anywhere and in an incredibly lightweight manner compared to the old days of heavyweight VMs. I’ve even got Docker containers running on a Raspberry Pi B+ that otherwise is so old that it would be nearly impossible to install the libraries required to run modern software.
I love Docker because it is the only sane method to selfhost shit with my Synology NAS, and I love my Synology NAS because it is the only Linux interaction that I have (from my old MacBook Pro).
Yeah, it “solved” the “it works on my machine” by bundling the machine with the code.
https://youtu.be/0uixRE8xlbY
Man, I really was interested in that topic, but that guy really can’t do talks.
What about this? https://youtu.be/5XY3K8DH55M
Also I created this repo to create a reproducible sec environment for myself. I added other languages, but personally work mostly with python. It is basically resonating for handling all the boiler plate:
https://github.com/takeda/nix-cde
For packaging in docker I started to use nix2container project as it gives me a greater control over layers. So for example when I package my phyton app I typically use 3 layers:
The algorithm mentioned in the video also helps a lot with reuse, but the above is more optimized by frequency of how things typically change.
BTW: today I discovered this https://github.com/astro/microvm.nix I haven’t play with it yet, but in theory it would let me generate a microvm image (in similar fashion to generate a docker container) which would let me to run my app natively as a tiny VM on EC2 for example, and use only minimum necessary of a typical OS to run it.
Docker has been a savior.
Now we just need to run docker inside the browser