Is it correct that the example use of a timestamp is for a message that will only be sent by one agent, i.e. the nova-compute instance? If so then I don't think it is a comparable scenario.
#1 I agree that typically synchronized timing is expected, but I wouldn't expect timing to be used as the basis of versioning data that can be updated at high frequency as it can be non-deterministic due to clock jitter etc.
#2 I missed the detail of a DB-derived timestamp, which means there is a "master lock". I have not tested this but use of MariaDB Galera Cluster might return the timestamp of whatever node executes the query?
I am not familiar with the nova code; if the use of timestamps is prevalent in this scenario then this particular bug is unlikely to be severe enough to warrant a change in approach.
Is it correct that the example use of a timestamp is for a message that will only be sent by one agent, i.e. the nova-compute instance? If so then I don't think it is a comparable scenario.
#1 I agree that typically synchronized timing is expected, but I wouldn't expect timing to be used as the basis of versioning data that can be updated at high frequency as it can be non-deterministic due to clock jitter etc.
#2 I missed the detail of a DB-derived timestamp, which means there is a "master lock". I have not tested this but use of MariaDB Galera Cluster might return the timestamp of whatever node executes the query?
I am not familiar with the nova code; if the use of timestamps is prevalent in this scenario then this particular bug is unlikely to be severe enough to warrant a change in approach.