[PIG-4007] GROUP ALL on an empty table produces no output

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Labels: cogroup empty filtered group group_all no records rows
• Assignee: Unassigned
• Reporter: Flip Kromer
• Created: 11/Jun/14 06:23
• Updated: 12/Jun/14 04:16

Description

Using GROUP ALL on an empty table produces no output. I would expect it to produce a single row with key 'all' and an empty bag. This seems inconsistent with PIG-514.

vals = LOAD 'test/data/pigunit/top_queries_input_data.txt' AS (name:chararray, val:int); 
empty = FILTER vals BY (1 == 2); 
empty_g_1 = GROUP empty ALL; 
empty_g_2 = GROUP empty BY 1;
empty_g_1_stats = FOREACH empty_g_1 GENERATE 
  COUNT_STAR(empty); 
DUMP empty_g_1; 
DUMP empty_g_2; 
DUMP empty_g_1_stats; 

None of the previous statements produce output. My workaround is to COGROUP with a one-line table:

 
one_line = FOREACH (LIMIT vals 1) GENERATE 1 AS uno; 
empty_cog = COGROUP one_line BY uno, empty BY 1; 
DUMP empty_cog;

A practical example of where it complicates things is set equality. You'd like to do this by testing whether the symmetric difference has zero size, but this bug prevents that method:

vals_a = LOAD 'test/data/pigunit/top_queries_input_data.txt' AS (name:chararray, val:int); 
vals_b = LOAD 'test/data/pigunit/top_queries_input_data.txt' AS (name:chararray, val:int); 
a_xor_b = FILTER (COGROUP vals_a BY name, vals_b BY name)
  BY ((COUNT_STAR(vals_a) == 0L) OR (COUNT_STAR(vals_b) == 0L));
-- doesn't work, a_xor_b has no rows if the sets are equal
a_equals_b = FOREACH (GROUP a_xor_b ALL) GENERATE
  ((COUNT_STAR(a_xor_b) == 0) ? 1 : 0) AS is_equal; 
DUMP a_equals_b;