pycor

python2 rewrite of Container/Overlay.

 

What?

It leverages OverlayFS under unprivileged LXD containers to share a single r/o rootfs. Containers write their changes (i.e. local packages, drop-in configuration, application data) to their respective r/w overlay directories.

It can be useful in a development environment where resources are scarce. OverlayFS inherently saves diskspace, but also saves memory usage and disk I/O by utilizing the Linux Page Cache, essentially providing a single cached operating system shared by all containers.

Why?

I wrote it trying to simplify simulating a managed hosting environment.

It provides atomic centralized control over shared packages and configuration. Local configuration can usually be included as drop-in files written to r/w overlays, which allows configuration management with puppet, ansible, etc.

Instantly configure packages and middleware across all containers by configuring the base container, and remounting the overlays:

lxc start base
lxc shell base
  # update packages, run puppet, do whatever
lxc stop base

for overlay in $(awk '$1 == "overlay" {print $2}' /etc/mtab); do
  mount -o remount $overlay
done

Performance

OverlayFS, as a mainline kernel-level overlay implementation, is very efficient in terms of performance. Containers reading a file from the r/o filesystem will result in one disk I/O operation, adding the file(s) to the Linux Page Cache. Further read() calls will effectively read from RAM.

Processes across containers essentially reference a single cached operating system: binaries, shared libraries/objects, and default configuration can all be read from RAM. Local drop-in configuration can be maintained in individual r/w overlays and easily changed during runtime.

Containers suffer less I/O overhead since the operating system is mostly cached, reducing access to secondary storage. At scale, this significantly decreases IOPS on the storage cluster, adding another performance benefit.