Migrate from ntpd to Chrony #1852
Conversation
Signed-off-by: Saikrishna Arcot <[email protected]>
Signed-off-by: Saikrishna Arcot <[email protected]>
Signed-off-by: Saikrishna Arcot <[email protected]>
Signed-off-by: Saikrishna Arcot <[email protected]>
Signed-off-by: Saikrishna Arcot <[email protected]>
| ### Monitoring | ||
|
|
||
| For the purpose of making time synchronization issues more visible, a Monit | ||
| check will be added to verify that the time is currently synchronized to one or |
There was a problem hiding this comment.
It would be good if the minimum number of active servers could be defined so that the synchronization is considered correct. For high requirements, 3 servers should be the minimum requirement.
It would also be good if the details of the time synchronization station (e.g. offset, jitter, delay, count) could be monitored from the network. As far as I know there is currently no Prometheus exporter included in SONiC, SNMP would probably be the only option here.
There was a problem hiding this comment.
For the first part, while specifying 3 servers is required for a high level of time accuracy, I'd rather not make it a minimum requirement within SONiC, only because if it is used in an environment that either doesn't need a high level of time accuracy or doesn't have 3 servers available, then that requirement won't be met.
For the second part, I agree on exposing the metrics somehow. I think the current standard we have is to publish the data into STATE_DB, which should make it easier to get exported elsewhere. However, this would involve having (at minimum) another daemon polling chrony frequently, and publishing the data into STATE_DB, and the scope of this effort is already large enough. Can you open an enhancement request for this?
There was a problem hiding this comment.
About the minimum servers required: I fully agree that there are environments where having accurate time is not that important. My recommendation was to make the minimum numbers of servers "defineable", then its up to the operator what is needed.
Regarding metrics topic:
Yes of course, see #1857
|
|
||
| Examples: | ||
|
|
||
| ``` |
There was a problem hiding this comment.
Can you make sure Chrony works with mgmt VRF as well? Please add examples with mgmt VRF.
This HLD is to describe the migration from ntpd to Chrony, along with the reasons this is needed and what the changes would be.