In case anyone is interested I received a pretty good answer from Oracle
support about this.
I think it boils down to the fact that the chrony service at times has been
seen to not correct the time immediately or within acceptable parameters and
this causes issues with RAC and multiple instances that require heavily on time
synchronization.
If there is a discrepancy reported from the NTP/GPS server to chrony it speeds
up or slows down the system clock, whereas ntp client will make the change
right away depending on the amount and whether or not slewing option is used
(which is recommended, btw)
For instance, say there was a 5 minute discrepancy, between the system clock
and the NTP/GPS reported time.
From my understanding, chrony would speed up the system clock until that 5min
gap was gone. That could take a significant amount of time depending on
threshold the application has.
Now my understanding of ntp client is that it will automatically increase the
time immediately (depending on slewing option) (+1min, +1min, etc.) until it is
synchronized with the server.
Doc ID 2068875.1 :
"The often misconception is that the chrony service is setting the time to the
one given by the NTP server. This is incorrect - what actually happens is that
based on the answer from the NTP server, chrony just tells the system clock to
go faster or slower. For this reason, sometimes even though the time is wrong
and the NTP server is working, the time does not get corrected immediately."
... unfortunately it appears that's the way(support for ntp client) it is going
to be for the foreseeable future. Unless of course customer demand and/or
necessity dictate otherwise.
Regards,
Dave
From: Herring, David
Sent: Tuesday, February 07, 2017 2:06 PM
To: oracle-l@xxxxxxxxxxxxx
Subject: RHEL 7 and Chrony
Folks,
Does anyone have insight as to why Oracle still relies on NTP even though
Redhat has moved on with Chrony as of v7? Various "cluvfy" commands still
assume the use of NTP and give warnings without it (not a huge deal) but we
found on RHEL 7.x with 12.1.0.2 after host reboots the CRS won't come back up
due to issues with NTP not being configured. For now our 7.x install
instructions require NTP but our Linux team isn't thrilled about the one-off
requirement for NTP.
Regards,
Dave