[cfarm-users] Podman on supported cfarm machines

[Gregor Riepl]

Hi Bruno,

> Use of containers generally tends to consume more memory (both RAM and disk),
> compared to traditional deployment/installation methodology, right?
> At least that's the impression I get from the minimum system requirements
> of OpenShift [1].

That is generally true, but OpenShift is a full-blown container 
orchestrator with many baseline services and is not comparable to a 
local Podman runtime and a handful of containers launched directly by users.

Single container instances don't use massively more memory than running 
binaries with system libraries in shared memory, and there is normally 
no observable overhead CPU-wise (seccomp bugs notwithstanding). The only 
exception is when you run software with a huge dependency chain, 
requiring replicas of every system-installed library in memory, but that 
would very likely still not make much of a difference.

Take into account that cfarm boxes are intended to build/test software. 
Compiling a moderately large C++ project already consumes huge amounts 
of resources. Running the compiler in a container would add very little 
overhead to that.

There is, however, an overhead in disk usage, because each container has 
its own file system that is distinct from the host. In that respect it's 
similar to a chroot, but with some copy-on-write magic sprinkled on top.

> If that's the case, I would suggest to exclude Podman from boxes with ≤ 8 GB
> of RAM.

No objections here - although it might still be useful to have it 
installed on more exotic hardware if supported, even with less RAM.

Regards,
Greg