PHOENIX-6050 Set properties is invalid

[wangchao316]

[wangchao316]

@virajjasani @gjacoby126 could you please review this ? Thanks.

[stoty]

I am generally not convinced that overriding poolsize from the client side is a good idea. It is very easy to overload the cluster by specifying a too large poolsize. IMO it's better to leave this knob in the cluster administrator's hand.

[stoty]

Also see my objection from the ticket that this will cause discrepancy between the poolsize used by the client, and by the Index rebuilding initiated by MetaDataObserver.

[wangchao316]

Also see my objection from the ticket that this will cause discrepancy between the poolsize used by the client, and by the Index rebuilding initiated by MetaDataObserver.

Thanks @stoty , I deal this ，have a big reason. when have not use phoenix-server, compiler and hbase scan operate in client-side, and scan by poolsize in parallelIterator. I test in our prodect environment, this performance improvement more.

[wangchao316]

and I find from 4.x to 5.x, reduce performance in some changed, which consumer use in our product environment by upgrading. I will always deal this question later.

[stoty]

I agree @wangchao316 that tuning poolsize is crucial for performance optimization. It is just my opinion that the tuning should be done on the cluster size, in hbase-site.xml, rather than on the client side.

[wangchao316]

I agree @wangchao316 that tuning poolsize is crucial for performance optimization. It is just my opinion that the tuning should be done on the cluster size, in hbase-site.xml, rather than on the client side.

yes, I feel we provide this parameter, should be altered in effect performance place. phoenix deploy in phoenix-server mode and no phoenix-server mode.

1. if be no phoenix-server mode, client will be compiler, optimization，scan to hbase-regionserver side. so client will use this parameter in ParallelIterator.java， which effect scan.
2. if be phoenix-server mode, we will alter parameter in phoenix-server .