Fix: LockManagerTest.updateValue is flaky

[shibd]

Motivation

#13663 Flaky-test: org.apache.pulsar.metadata.LockManagerTest.updateValue

The reasons are discussed in detail in this PR(#13725)

This is mainly caused by the callback method of MetadataStore and the concurrent access of the notification thread to the MetadataCache#refresh.

We should queue them in the metada-store thread pool(SingleThread)

Modifications

1. In AbstractMetadataStore method, all future callback opt are executed through metada-store thread.

Documentation

• [ x] no-need-doc

[shibd]

@Jason918 @Technoboy- @codelipenghui Hi, Can you help me review it again? Thanks.

[nodece]

This PR looks the same as #13809, we can find this solution together.

[shibd]

This PR looks the same as #13809, we can find this solution together.

Thank you reply, I saw you fix the callback problem of ZKMetadata. I'll cancel this commit.

However, the essential reason why this unit test will report an error is LocalMemoryMetadataStore that does not callback through metada-store thread pool.

I think it can be handled like this:

In AbstractMetadataStore method, all future callback opt are executed through metada-store thread.

[shibd]

/pulsarbot run-failure-checks

[shibd]

/pulsarbot run-failure-checks

[shibd]

@Jason918 @codelipenghui @nodece Sorry,The current changes have caused other problems. I will resubmit the PR in another way