IGNITE-18554 Remove MetaStorage learners on topology events

[sashapolo]

https://issues.apache.org/jira/browse/IGNITE-18554

[sanpwc]

1. We do call onNodeValidated, ..., onNodeLeft are called within raft thread, is that correct?
2. If true
   2.a You didn't introduce any synchronous long running logic, e.g. network communication as a reaction to such events, did you? serializationFuture?
   2.b You didn't introduce any logic that may prevent CMG state-machine idempotentancy, did you? Meaning, that if we'll have an exception within onSmt on Node A and won't have such an exception on Node B, CMG listener won't be effected, is that true?
3. Who is responsible for exception handling for serializationFuture? E.g. logging.

[sashapolo]

We do call onNodeValidated, ..., onNodeLeft are called within raft thread, is that correct?

yes

You didn't introduce any synchronous long running logic, e.g. network communication as a reaction to such events, did you? serializationFuture?

How is it possible to introduce synchronous network calls in our code? All such stuff is done asynchronously. That's exactly why I need the serializationFuture - to chain a bunch of asynchronous calls to preserve Raft linearization guarantees. But I never wait on on this future.

You didn't introduce any logic that may prevent CMG state-machine idempotentancy, did you? Meaning, that if we'll have an exception within onSmt on Node A and won't have such an exception on Node B, CMG listener won't be effected, is that true?

Since all operations are asynchronous, I expect all exception to simply fail on the related thread and be possibly logged somewhere.

Who is responsible for exception handling for serializationFuture? E.g. logging.

Logging is done inside the action that is passed to executeIfLeaderImpl. For example, see the updateConfigUnderLock method, it should do all the logging.