Ok, now I got how it works. Thank you.

Regards,
Max




From:   ivan bella <[email protected]>
To:     [email protected], Massimilian Mattetti/Haifa/IBM@IBMIL
Date:   14/11/2016 14:41
Subject:        Re: use a single byte for split points



Just guessing here, but if your keys are uniformly distributed between 0 
and NTS-1, then perhaps none of your keys are actually going into 
(-inf,0].  The first byte of the key may be 0 but if you have anything 
else in the key then it would fall into (0, 1].
On November 14, 2016 at 6:14 AM Massimilian Mattetti <[email protected]> 
wrote:

Hi all,

I want to pre-split a table using the first byte of the key which is 
uniformly distributed between 0 to the number of tablet servers (NTS) - 1. 
I defined the split points as [0, 1, 2, ..., NTS - 2], in this way the 
split ranges should be distributed on the tablet servers as follows:  

(-inf ,0] - Server 1 
(0, 1] - Server 2
(1, 2] - Server 3
...
(NTS - 2, +inf) - Server NTS

Unfortunately it is not working as expected since there is always a tablet 
server that has ingestion rate equal to 0. Am I missing something?

Thanks,
Max



Reply via email to