Hi all,
following my installation (details here) : of Ceph on Proxmox for multi purpose (object bucket & HCI VM for K8s cluster), I'm facing well skewing :
clock skew detected on mon.srv158ilo151, mon.srv161ilo150mon.srv158ilo151 clock skew 0.79889s > max 0.05s (latency 0.000515921s)
mon.srv161ilo150 clock skew 0.637552s > max 0.05s (latency 0.00053284s)
on 2 of my 3 monitors. It's not a final installation just first POC iteration of a future product. But it's on a production environnement and it's not possible to use internet NTP server for security concern. So my NTP server a inside server, furnished by IT and... I expect it's an Active Directory server so drifting is the rule.
I've read quite a few post relative to problems relative to this time shifting on proxmox - Ceph and resolution is a bit confusing for a noob like me (noob at debian stuff not HCI). And some people I know that are working w/ NTP on production datacenter gave me other advices, powering up the confusion. So since I cannot directly connect to internet on this network, I have 2 choices as far as I understand :
moving to Chrony (but how ? should I put another NTP server try likne this w/ my AD server ?) or
moving to NTPD : using my 3 nodes to mount a proximity NTPD architecture with my AD alone as a "root" for those server. This way even if it's drifting on AD side, my cluster will be synchrone for itself which is the most important thing to achieve.
Did I miss someting ? Do you have some pros and cons on thoses scenarii ? should I give a try to chrony at first and then NTPD ?
following my installation (details here) : of Ceph on Proxmox for multi purpose (object bucket & HCI VM for K8s cluster), I'm facing well skewing :
clock skew detected on mon.srv158ilo151, mon.srv161ilo150mon.srv158ilo151 clock skew 0.79889s > max 0.05s (latency 0.000515921s)
mon.srv161ilo150 clock skew 0.637552s > max 0.05s (latency 0.00053284s)
on 2 of my 3 monitors. It's not a final installation just first POC iteration of a future product. But it's on a production environnement and it's not possible to use internet NTP server for security concern. So my NTP server a inside server, furnished by IT and... I expect it's an Active Directory server so drifting is the rule.
I've read quite a few post relative to problems relative to this time shifting on proxmox - Ceph and resolution is a bit confusing for a noob like me (noob at debian stuff not HCI). And some people I know that are working w/ NTP on production datacenter gave me other advices, powering up the confusion. So since I cannot directly connect to internet on this network, I have 2 choices as far as I understand :
moving to Chrony (but how ? should I put another NTP server try likne this w/ my AD server ?) or
moving to NTPD : using my 3 nodes to mount a proximity NTPD architecture with my AD alone as a "root" for those server. This way even if it's drifting on AD side, my cluster will be synchrone for itself which is the most important thing to achieve.
Did I miss someting ? Do you have some pros and cons on thoses scenarii ? should I give a try to chrony at first and then NTPD ?
