0

I'm attempting to test my system monitoring by temporarily disabling NTP and using "date -s '5 minutes ago'" to change time enough to trigger my alerts. This test works fine on SLES 11 and SLES 12 SP1. On SLES 12 SP2 the date command reverts within seconds. What could be causing this? Here's my sample problem:

sr-0f6a00494095:/ # date
Tue Nov 29 19:59:12 UTC 2016
sr-0f6a00494095:/ # date -s '5 Minutes Ago'
Tue Nov 29 19:54:18 UTC 2016
sr-0f6a00494095:/ # date
Tue Nov 29 19:54:19 UTC 2016
sr-0f6a00494095:/ # date
Tue Nov 29 19:54:20 UTC 2016
sr-0f6a00494095:/ # date
Tue Nov 29 19:54:21 UTC 2016
sr-0f6a00494095:/ # date
Tue Nov 29 19:59:22 UTC 2016

To be clear: I see no ntpd process running, and the service is shutdown. No logs in /var/log/messages. No NTP logs either. This is a VM running on Azure, which is the same use case as my SLES 11 SP3 and SLES 12 SP1 VMs where this example works.

Minus Ex
  • 1
  • 1
  • SLES 11 and SLES 12 use different clocks when running as VMs. Specifically, SLES 12 supports most paravirtualized realtime clocks. I don't have any experience with azure, but you should be able to change the emulated or paravirtualized clock on the VM itself to something a little more dumb. – Spooler Nov 29 '16 at 20:28

1 Answers1

0

For the next poor soul that deals with this: Azure has released "Windows Server 2016 Accurate Time" on SLES 12 SP2. It's built into Hyper-V. This explains why even with NTP off my time was being forcibly adjusted.

Minus Ex
  • 1
  • 1