[Xen-users] Unstable/inconsistent auto-start of domUs under Xen-4.8/Linux-4.9

[Daniel Widenfalk <Daniel@xxxxxxxxxxxx>]

Hi,

I'm having consistency issues on reboot of my Debian (Stretch) Xen-
based server. It is currently running Xen-4.8 with Dom0 using Linux-4.9.

The issue I'm seeing is that not all the domUs linked to from
/etc/xen/auto get started on reboot and while the first machine always
start (The links in /etc/xen/auto are named 001-... 002-... etc) it is
semi-random what other machines that manage to boot - about every other
machine starts ok. I can find no traces of logs for the failed-to-start
domUs and there are no gaps in the domU numbering sequence of those
that managed to start.

Once the Dom0 has booted and I can log in I can manually start the
missing domUs without issue. Well, I sometimes get the message

  "failed to free memory for the domain"

when issuing the "xl create ..." command. Re-running the command
launches the domU. Side-note: After starting all my nine domUs
I still have approximately 22GB available to the Dom0.

Are these issues related? Maybe the Dom0 is too heavily loaded
during boot (the Dom0 hosts the raid-1/raid-6 arrays) to relinquish
memory fast enough for the auto-start domUs and they fail to start
because of that?

As the machine is in a semi-production state (some of the domUs
are actively used) I cannot just randomly try stuff out and am thus
seeking some sage advice here.

Some ideas that I've been thinking of trying:

1) Somehow stagger the auto start-up of the domUs. They're all booting
   through pygrub so I can just go in and stagger the boot menus in
   steps of 30 seconds as a slap-on duct-tape band aid.

2) Disable balooning on Dom0. My Dom0 will never run anything memory
   intensive so limiting it to, say, 4GB would not be a severe
   restriction.

Advice and comments are much welcome!

Regards
/Daniel W

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxx
https://lists.xen.org/xen-users