Folks, I have a node.js script running on my Windows machine that uses the dockerode npm package to talk to docker on said box and starts and kills docker containers.
However, after the containers have been killed off, docker still holds on to the memory that it blocked for those containers and this means downstream processes fail due to lack of RAM.
To counter this, I have powershell scripts to start docker desktop and to kill docker desktop.
All of this is a horrid experience.
On my Mac, I just use Colima with Portainer and couldn’t be happier.
I’ve explored some options to replace Docker Desktop and it seems Rancher Desktop is a drop-in replacement for Docker Desktop, including the docker remote API.
- Is this true? Is Rancher Desktop that good of a drop-in replacement?
- Does Rancher Desktop better manage RAM for containers that have been killed off? Or does it do the same thing as Docker Desktop and hold on to the RAM?
Are there other options which I’m not thinking of which might solve my problems? I’ve seen a few alternatives but haven’t tried them yet -
moby,
containerd,
podman
I don’t actually need the Docker Desktop interface. So pure CLI docker would also just work. How are you all running pure docker on Windows boxes?
First, it’s not possible to use “pure docker” on Windows. Docker is for running additional user mode environments under the same kernel. You can’t run Linux applications under the Windows kernel without WSL1, and WSL1’s Linux implementation does not support the features required for Docker. This is also possible in limited cases with Windows Server, but because of differences in the way Windows works you almost always end up running a second kernel.
WSL2 can be used to run Docker, and in fact that’s how Docker Desktop works since years ago. When you start Docker Desktop it starts a WSL2 distribution under which the containers run. Running Docker from the command line only will not positively change the performance of your containers.
Running other virtualization software, especially VirtualBox, to start a separate Linux VM and running your containers in there is going to be more complicated and give worse performance unless you disable all virtualization-based features of Windows, such as WSL2 and security isolation.
The solution to your memory problem is most likely one of the following:
- Don’t disable the pagefile. Windows uses a weird memory model where all virtual memory must be backed by physical memory. Certain software will allocate virtual memory without using it, and Windows will require that the sum of the physical memory size and the page file size be adequate to use all of that virtual memory. Disabling the pagefile or limiting it to small sizes because you “have enough RAM” will cause out of memory errors while you still have plenty of RAM available.
- Reduce the amount of memory that Docker is allowed to use to a level that your Windows software can tolerate. You may need to switch Docker Desktop to Hyper-V mode for this option to be available, which isn’t an option if you’re on Windows Home, and this may reduce compatibility.
- After stopping your containers, run
echo 1 > /proc/sys/vm/compact_memoryat a WSL2 prompt orwsl -u root -- bash -c 'echo 1 > /proc/sys/vm/compact_memory'from a Windows prompt. See Memory Reclaim in the Windows Subsystem for Linux 2 for details about what this does.
Wsl2 and podman.
Far from a nice overall experience compared to using docker on a real OS, but the best I could get on windows
I despise Docker Desktop. Before I knew anything about docker or containers, all I knew was that it was in the required software list for my work for building our software. All I knew was that if it wasn’t open, my build would fail and if it was open, my laptop would slow down to a crawl.
Eventually I took classes on Docker for work and learned quite a bit about it. I learned that I could use docker from command line with no UI, and I wouldn’t take anywhere near the performance hit. I eventually linked my IDE docker runtime to podman running on WSL2. Now I take pretty much no noticable performance hit.
TL;DR: you can replace Docker Desktop with WSL2 command line commands and have no UI.
The integration of Docker for windows with wsl2 is an abomination that breaks just about every time I update either ddw or windows. Also the fact that it is tied to my user account ( both ddw and wsl2) means that it is not a great choice for persistent services. I still use it to provide monitoring agents for Prometheus and portainer, but otherwise everything runs on Linux vms on my homelab xenserver cluster.
It is possible to install docker without ddw. It’s documented for server versions of windows, but is basically only for running windows containers. The only use case for that is windows build agents as far as I can tell.
Docker can be installed standalone on wsl2 and would be more reliable.
thanks for that :)
BTW, if I fire up a bunch of docker containers in WSL2 using podman or native docker, and then kill them, does WSL2 release the RAM it acquired to run those containers?
At this point i would just install a linux distro like ubuntu/debian/arch as a VM on virtualbox/vmware/hyperv and do it from there, WSL2 is supposed to be custom ubuntu minus gui running on hypervany way.
WSL2 runs ubuntu on Hyper-V. It isn’t really as custom as you’d think. You can install other distros besides ubuntu. Or you can install regular ubuntu instead of LTS.
Take a look at this for running other distros: https://learn.microsoft.com/en-us/windows/wsl/use-custom-distro
And as such it’s perfectly suited for what OP is trying to achieve. A custom VM will be a lot more work than using WSL2.
If you don’t need UI, I prefer Podman. Rancher Desktop is good though.
When I had a look on it a while ago they didn’t support Docker Compose. But except this it’s a drop in replacement.
There’s now https://github.com/containers/podman-compose which seems to work okay, haven’t tested heavily though
This is a common misconception. Podman has similar commands to Docker CLI but it’s not a “drop-in replacement”. Depending on your usage, you might run into things that don’t work the same.
When I had Windows I ran WSL2 + standard Linux docker, worked flawlessly. If you have all your files in the WSL volume, it’s also really fast compared to Docker Desktop on Windows or Mac. I found it almost as fast as a native Linux version.
I thought WSL2 made things slow because of some stupidity they did with the code? Maybe they fixed it.
Anyways, is it able to take as much resources as it needs from the host? Unrestricted in terms of RAM and CPU?
is it able to take as much resources as it needs from the host? Unrestricted in terms of RAM and CPU?
https://learn.microsoft.com/en-us/windows/wsl/wsl-config#configuration-setting-for-wslconfig
It’s slow when you go cross-filesystem, meaning accessing WSL2 files from Windows, or accessing Windows files from WSL2. If you keep all related files in WSL2, it’s really comparable to native Linux experience (with a small penalty due to being ran in a VM, but it’s not noticeable by a human eye).
As far as I know, yes, it can take all the resources it needs.
made things slow
That’s probably referring to how file systems are handled. Going from WSL to windows file system is slower than using the “proper” mount point
Unrestricted
yes
