Need sequential startup of VMs at boot to avoid race condition

[tonsimple]

Qubes OS version:

R 3.2

Affected component(s):

Fedora based VMs, Fedora 23, 25, 26

---

Steps to reproduce the behavior:

Basically, I have a dedicated VPN VM and a fedora-minimal proxy-VM that is attached to it so that everything that connects to the proxyvm goes through VPN.

All VMs (the proxyvm, the vpnvm, and the vpns that connect to them) are set to autostart on boot

approximately 19 reboots out of 20 everything works "as expected"

about 1 reboot out of 20 the proxyvm just can't send a packet out

This suggests a race condition that I can't quite catch (and rebooting the machine constantly is kind of annoying)

Expected behavior:

the config always works like it does most of the time

Actual behavior:

about 1 reboot out of 20 the proxyvm just can't send a packet out

General notes:

When I start manually the issue only arises (about 1 trial out of 40 or so) when I shut them all down, then try to bring up a VM connected to proxyvm (thus qubes try to bring everything up automatically)

However, if I bring them up manually in sequence, the issue never arises (>100 attempts)

Thus, the easiest fix would be to have Qubes bring them up slowly, one by one, in order of

[start up the main net vm]
[start up completes]
[start up vpnvm]
[start up completes]
[start up proxyvm]
[start up completes]
[start up other qubes else]

However, I can't for the life of mine figure out how to edit VM startup order (ideally, I'd like to have a small pause between them as well, just in case)

---

Related issues: