DNS is not configured during provisioning

[bas-de-laine]

Description

Since v0.9.0, when running provisioning scripts (from the templates provisioning section), DNS is not configured/working. At this point in time, there is no /etc/resolv.conf file.

Once the VM is booted, /etc/resolv.conf is materialized. Maybe setupDNS (https://github.com/abiosoft/colima/blob/main/environment/vm/lima/dns.go#L26) is called too late?

Version

colima version 0.9.1
git commit: 0cbf719

runtime: docker
arch: x86_64
client: v28.5.0
server: v28.4.0
limactl version 1.2.1
qemu-img version 10.1.0
Copyright (c) 2003-2025 Fabrice Bellard and the QEMU Project developers

Operating System

• macOS Intel <= 13 (Ventura)
• macOS Intel >= 14 (Sonoma)
• Apple Silicon <= 13 (Ventura)
• Apple Silicon >= 14 (Sonoma)
• Linux

Output of colima status

INFO[0000] colima is running using macOS Virtualization.Framework
INFO[0000] arch: x86_64
INFO[0000] runtime: docker
INFO[0000] mountType: virtiofs
INFO[0000] docker socket: unix:///Users/bas-de-laine/.colima/default/docker.sock
INFO[0000] containerd socket: unix:///Users/bas-de-laine/.colima/default/containerd.sock

Reproduction Steps

1. Using the default template https://github.com/abiosoft/colima/blob/v0.9.1/embedded/defaults/colima.yaml, add a provisioning step to check whether /etc/resolv.conf exists and whether it is possible to resolve github.com:

provision:
  - mode: system
    script: |
      #!/usr/bin/env bash
      { 
        printf "Checking whether /etc/resolv.conf exists\n\n"
        [ -f /etc/resolv.conf ] && printf "🟢 /etc/resolv.conf exists\n" || printf "🔴 /etc/resolv.conf is missing\n" 
        printf "\nCalling getent ahosts github.com\n\n"
        getent ahosts github.com
      } > /Users/bas-de-laine/colima_resolv_conf_is_present.txt

2. put the template in ~/.colima/_templates/default.yaml
3. create the vm COLIMA_PROFILE=default colima start --template

Expected behaviour

I expect DNS to work when provisioning scripts run.

Additional context

No response

[abiosoft]

What are you trying to achieve?

[bas-de-laine]

I use the provision feature to install additional packages inside the VM (tcpdump, dnsutils, ping, mise, etc.) and setup a VPN.

I also use that moment to setup infra containers that provide DNS from the host to the containers, configure iptables and docker (allow-direct-routing) so that the host can route directly to containers instead of using ports on a host's IP.

It worked in 0.8.4. I understand the change from systemd-resolved to dnsmasq broke this (I was also using *.colima.internal which you now also use! :) ).

I already have a work around though. I simply don't use the provision feature anymore. I start the VM with colima start --template, wait until it is fully booted and then colima ssh -- provision.sh, etc.

[abiosoft]

In that case, I think it makes sense to provide an option to run provision scripts after Colima is done booting.

[bas-de-laine]

That would be great.

For the moment, I use Task to orchestrate the start, then provisioning, etc. It works well. But having this built-in to Colima could reduce the dependency to orchestrators like Task, just, etc.

This makes me wonder if colima status could also provide health information like dns is working, ip can route, etc. Or a new colima health subcommand that would provide that information. The VM could regularly check whether it can resolve, say github.com and ping github.com (or a user provided domain, ip to test) and report if that works. Colima could update a file in ~/.local/state/colima/<profile>/health and watchexec, entr, etc. could monitor for changes and let the user deal with health issues.

If we had that, we could use it as a gate before the provisioning scripts can run.

[abiosoft]

An health information would not be a bad idea. It should probably be included the status command output.

I would refrain from pinging external servers as that can annoy some users. I think there is enough information within the VM and the host to deduce if everything is fine.