Question

I wonder if it’s possible to pivot a table in one pass in Apache Pig.

Input:

Id    Column1 Column2 Column3
1      Row11    Row12   Row13
2      Row21    Row22   Row23

Output:

Id    Name     Value
1     Column1  Row11
1     Column2  Row12
1     Column3  Row13
2     Column1  Row21
2     Column2  Row22
2     Column3  Row23

The real data has dozens of columns.

I can do that with awk in one pass then run it with Hadoop Streaming. But majority of my code is is Apache Pig so I wonder if it’s possible to do it in Pig efficiently.

Answer 1

You can do it in 2 ways:
1. Write a UDF which returns a bag of tuples. It will be the most flexible solution, but requires Java code;
2. Write a rigid script like this:

inpt = load '/pig_fun/input/pivot.txt' as (Id, Column1, Column2, Column3);
bagged = foreach inpt generate Id, TOBAG(TOTUPLE('Column1', Column1), TOTUPLE('Column2', Column2), TOTUPLE('Column3', Column3)) as toPivot;
pivoted_1 = foreach bagged generate Id, FLATTEN(toPivot) as t_value;
pivoted = foreach pivoted_1 generate Id, FLATTEN(t_value);
dump pivoted;

Running this script got me following results:

(1,Column1,11)
(1,Column2,12)
(1,Column3,13)
(2,Column1,21)
(2,Column2,22)
(2,Column3,23)
(3,Column1,31)
(3,Column2,32)
(3,Column3,33)