Hi,
Some thoughts that might help you to make a decision:
- IgniteAtomicLong and IgniteAtomicSequence are both public interfaces
but their implementations are inside the internal part of Ignite. It is a
bad practice to reference Ignite internals so you cannot configure them in
the XML file.
- You create both atomic long and sequence as Ignite#atomicLong(name,
initialValue) and Ignite#atomicSequence(name, initialValue). Thus, if you
really want to make them Spring beans you could create an annotated spring
configuration (@Configuration MyConfiguration { ... }) and provide factory
methods for them (like @Bean public requestId() { return
ignite.atomicLong("requestId", initialValue); }
- In my opinion both the methods currently have this design flaw: even
if you enable native persistence, you still cannot get the last sequence
value since you are forced to provide initial value in the factory methods.
Thus, you have to provide your custom mechanism to restore last sequence
value after the cluster restart and pass it as initialValue if you want the
value to be unique through the cluster restarts.
- Please note differences between atomic long and sequence: both of them
provide unique values but the "long" is sequential but slower and the
"sequence" is not sequential but faster.
- Are you sure simple Java UUID.randomUUID() will not work for you?
