Description of problem: When we deploy OpenStack we don't set up an NTP server on the nodes. This can lead to the system times being wildly off, which causes HA to not function. We can use the Undercloud to do NTP sync for the overcloud. Version-Release number of selected component (if applicable): How reproducible: 100% Steps to Reproduce: 1. Deploy HA overcloud 2. If overcloud servers can't reach NTP, time will be off 3. Actual results: HA fails to achieve quorum Expected results: Time should sync, and HA should achieve quorum Additional info: This particularly affects proof-of-concept deployments, where the cloud might not be connected to the Internet. If an internal NTP server is not configured, there is no way to sync the time.
I might suggest a somewhat different approach. Let the admin specify an NTP server. If they don't, configure the undercloud as a dummy ntp server. If the do, register the undercloud and the overcloud nodes to the ntp server specified. You could also configure the undercloud to sync from the external server, then act as the server for the overcloud nodes.
(In reply to Mike Burns from comment #3) +1
this is an RFE to support environments where overcloud nodes do not have a routable network to an ntp server. is this really a blocker?
Hi James, is this still an issue/rfe? I know we were dealing with bug regarding syncing time across nodes before. Thanks, Jarda
(In reply to Jaromir Coufal from comment #6) > Hi James, is this still an issue/rfe? I know we were dealing with bug > regarding syncing time across nodes before. Thanks, Jarda Yes, this is still a serious shortcoming. We require an NTP server, but if the customer doesn't have one on their network, then NTP sync has to happen over the Internet, which is much less reliable. Considering that a time drift between controllers can bring the entire HA cluster down, I think we need to provide a reliable backup option for customers that don't have an infrastructure NTP server. It's not like it's a big change, we just have to configure NTP on the undercloud (probably add an upstream NTP server setting in undercloud.conf and use that as the authoritative time source for the undercloud). Really, though, as long as the controllers are syncing to the same source (Undercloud), we don't even care if the Undercloud has the correct time, it's just important that all servers remain in sync. Of course, the recommendation is still to use an infrastructure NTP server, ideally one with redundancy/HA.
this is still an open rfe
This bugzilla has been removed from the release and needs to be reviewed and Triaged for another Target Release.
Obsolete RFE without enough of backed demand.