Reduce dns lookup overhead on NodeScheduler

[guhanjie]

Description

Reduce dns lookup overhead on NodeScheduler

Additional context and related issues

It does not need to dns lookup for hostname if HostAddress has port, which incurred additional overhead when dns too slow on some host environment.

Release notes

(x) This is not user-visible or docs only and no release notes are required. In case someone things differently, here is a stub:

# release notes stub (but perhaps no Rn needed)

* Reduce dns lookup overhead on NodeScheduler.
   This can avoid unnecessary dns lookup for HostAddress has port, while previous would incurred additional overhead when dns too slow on some host environment.

[guhanjie]

@findepi I had one ci check(test-jdbc-compatibility) not passed, I want to re-run it, but can't find the trigger entry. Could you help me, thanks

[findepi]

Does this matter only when multiple worker nodes on single host?
If so, I don't think this is anyhow recommended scenario.
Am i reading this wrong?

[guhanjie]

Does this matter only when multiple worker nodes on single host? If so, I don't think this is anyhow recommended scenario. Am i reading this wrong?

@findepi It will must do host.toInetAddress(); whenever one/multi workers on host, but that is unnecessary when nodes has port.
And it would be too slow or hang if dns lookup has problem on some environment as possible as we have.

So, I think the block code before if (!host.hasPort()) { is absolutely unnecessary when host has port.
It should be optimized as pull request.

[guhanjie]

Does this matter only when multiple worker nodes on single host? If so, I don't think this is anyhow recommended scenario. Am i reading this wrong?

@findepi I suppose you are reading wrong, please check it again.

[guhanjie]

@findepi Does this commit have any problem?

[guhanjie]

@martint Could you have a review for this? It seems @findepi always missing my message...