QEMU: Pulling images is very slow

[jandubois]

Seems to happen with both docker pull and nerdctl pull.

Questions:

• Is this an upstream issue; can it be reproduced in a default Lima config with Ubuntu
• Is it related to user mode networking; we are still using slirp with vz.

[jandubois]

Even with QEMU the docker pull performance is pretty bad:

$ docker rmi ghcr.io/rancher-sandbox/busybox
[...]
$ time docker pull ghcr.io/rancher-sandbox/busybox
[...]
real	0m22.613s

Time with nerdctl is about 10s. Using docker with the Lima docker VM (based on Ubuntu) is around 12s.

This should take less than 2 seconds!