hadoop-pig-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Mridul Muralidharan <mrid...@yahoo-inc.com>
Subject Re: [jira] Created: (PIG-723) Pig generates incorrect schema for generated bags after FOREACH.
Date Wed, 18 Mar 2009 11:24:42 GMT


   I think the schema generated is right.
It might a problem with declaring complex types (bags only ?) in terms 
of scalars in the script which you might be hitting (iirc there is a bug 
for it already).

 From what I recall, even I have faced issues with things like "foreach 
A generate $0 as f1, {(1.0, 'str')} as f1;" type declarations - pig 
seems to not be able to handle them currently.

If this is the issue, rest of the mail might help you -

The workaround we have is to have a dummy udf which returns this value - 
and use that udf.

An example snippet which we use :

--- A custom udf which generates bag with tuples (0, '')
define GENERATE_EMPTY_BAG1 myudf.GenerateEmptyBag('long, chararray');
--- A custom udf which generates bag with tuples (0, '', 0.0f)
define GENERATE_EMPTY_BAG1 myudf.GenerateEmptyBag('long, chararray, float');

grp_op = COGROUP inp1 by eid, inp2 by eid6;

res = FOREACH grp_op { GENERATE FLATTEN((COUNT(inp1) != 0 ? inp1 : 
GENERATE_EMPTY_BAG1(0L))) AS (eid:long, query:chararray), 
FLATTEN((COUNT(inp2) != 0 ? inp2 : GENERATE_EMPTY_BAG2(0L))) AS 
(eid6:long, candidate_url6:chararray, rank:float); };


The idea above is to have some default values when the bags are empty 
(that is, no data for a particular eid/eid6).
Note the exact syntax in foreach (defends against parser bugs in pig), 
and the use of the udf - the generate the bag.


Dhruv M (JIRA) wrote:
> Pig generates incorrect schema for generated bags after FOREACH.
> ----------------------------------------------------------------
>                  Key: PIG-723
>                  URL: https://issues.apache.org/jira/browse/PIG-723
>              Project: Pig
>           Issue Type: Bug
>     Affects Versions: 0.1.0
>          Environment: Linux
> $pig --version
> Apache Pig version 0.1.0-dev (r750430)
> compiled Mar 07 2009, 09:20:13
>             Reporter: Dhruv M
>             Priority: Critical
> grunt> rf_src = LOAD 'rf_test.txt' USING PigStorage(',') AS (lhs:chararray, rhs:chararray,
r:float, p:float, c:float);                            
> grunt> rf_grouped = GROUP rf_src BY rhs;                                         
> grunt> lhs_grouped = FOREACH rf_grouped GENERATE group as rhs, rf_src.(lhs, r) as
lhs, MAX(rf_src.p) as p, MAX(rf_src.c) AS c;                
> grunt> describe lhs_grouped;
> lhs_grouped: {rhs: chararray,lhs: {lhs: chararray,r: float},p: float,c: float}
> I think it should be:
> lhs_grouped: {rhs: chararray,lhs: {(lhs: chararray,r: float)},p: float,c: float}
> Because of this, we are not able to perform UNION on 2 sets because union on incompatible
schemas is causing a complete loss of schema information, making further processing impossible.
> This is what we want to UNION with:
> grunt> asrc = LOAD 'atest.txt' USING PigStorage(',') AS (rhs:chararray, a:int);  
> grunt> aa = FOREACH asrc GENERATE rhs, (bag{tuple(chararray,float)}) null as lhs,
-10F as p, -10F as c;
> grunt> describe aa;
> aa: {rhs: chararray,lhs: {(chararray,float)},p: float,c: float}
> If there is something wrong with what I am trying to do, please let me know.

View raw message