Impala substr can't get utf8 character correctly - hadoop

Question

I am new to ETL and I was assigned with a task on sanitizing some sensitive information before giving the data to a client.
I am using HUE web client with Impala.
What I want to do is:
For example, a column info like '京客隆（三里屯店）', then I need to transform it into something like '京XXX店）' .
My query is:
select '京客隆（三里屯店）', concat(substr('京客隆（三里屯店）', 1, 3), 'XXX', substr('京客隆（三里屯店）', char_length('京客隆（三里屯店）') -6, 6));
But I get gibberish in the output:
'京客隆（三里屯店）' | concat(substr('京客隆（三里屯店）', 1, 3), 'xxx', substr('京客隆（三里屯店）', char_length('京客隆（三里屯店）') - 6, 6))
京客隆（三里屯店） | 京XXX�店�
The problem is that :
select '京客隆（三里屯店）', substr('京客隆（三里屯店）', char_length('京客隆（三里屯店）') -3 , 3);
output: 京客隆（三里屯店） ��
doesn't get the correct characaters. Why is that? I pasted the string in python shell and I can get the correct characters if I only take the last 3 bytes.

Answer 1

It turns out that I misunderstood the function substr.
substr(STRING a, INT start [, INT len]) :
It takes characters starting from (including) INT start. So for example my string '京客隆（三里屯店）' is 27 bytes long in total, and each utf8 char takes 3 bytes here. I need to take the last 3 bytes, which is the ） , then I need to write:
substr('京客隆（三里屯店）, 27 - 2 ,3 ) .
It then gets the 25, 26, 27 3 bytes and display the char ） correctly.
Updated:
I was told to use :
SELECT regexp_replace('京客隆（三里屯店）', '(.)(.*)(.{2})', '\\1***\\3');
works like an charm :P.