Shared Service - Vanilla Hazelcast

[charliecatney]

Hi,

I work in an environment where we have a managed Hazelcast cluster, consequently it is not straightforward to add Bucket4J jars to the cluster nodes class path.

Does bucket4j offer a distributed model with all serialisation etc happening on the client side, or using Hazelcast built in serializers?

Thanks

[vladimir-bukhtoyarov]

Hello @charliecatney

Currently, there is no such feature. But I can add it. However, you need to get a deep understanding that client-side implementations based on CAS or locking will show less performance than current implementation because bucket4j-hazelcast uses EntryProcessors, see https://hazelcast.com/blog/an-easy-performance-improvement-with-entryprocessor/

If you agree with performance degradation, I will add such a feature(most likely at weekend).

[charliecatney]

Hi @vladimir-bukhtoyarov,

That would be great. I appreciate the principals and understand the EntityProcessor model gives significant performance gains.

It would be interesting to do some performance benchmarking of the 2 approaches, to understand the material impact. In the Bucket4j Token Bucket implementation, does every request for a permit make a call to the backing store?

Thanks

[vladimir-bukhtoyarov]

Yes. One permit check leads to one interaction with store until one of optimizations is applied during bucket construction.

[charliecatney]

Thank You, @vladimir-bukhtoyarov, let me know once you have had a chance to look at the client side implementation, and I really appreciate the quick response to my query.

[vladimir-bukhtoyarov]

@charliecatney

Desired functionality has been released with version 7.1.0

See documentation https://bucket4j.com/7.1.0/toc.html#support-for-externally-managed-hazelcast-without-classpath-access

[charliecatney]

Awesome, thanks @vladimir-bukhtoyarov. I'll integrate with this and let you know if I encounter any issues.