Join stage output quantity mismatch query

[vivekreddy]

I have a job wherein I am joining two datasets using a join stage and join type as left outer.
I have sorted and partitioned both datasets on the join key.
However, still I am not getting the desired output.

I have around 6670 rows as input to the join stage, but as output, I get only 2099. I thought that 6670 remains the minimum possible when performing left-outer join.

Any suggestions on what I should do?

[vijayrc]

I have a job wherein I am joining two datasets using a join stage and join type as left outer.
I have sorted and partitioned both datasets on the join key.
However, still I am not getting the desired output.

I have around 6670 rows as input to the join stage, but as output, I get only 2099. I thought that 6670 remains the minimum possible when performing left-outer join.

Any suggestions on what I should do?

Though you have mentioned LEFT OUTER, make sure the files are Link Ordered properly [Left and Right]

[ray.wurlod]

It may also be possible that your data are not partitioned based upon the join key (in addition to any other potential cause). Check that also.

[vivekreddy]

All done, still not working

[kumar_s]

What type of partition is used?
Just be aware that, 6670 will be output, only if its a Outer Join on the dataset which has 6670 number of records.

[vivekreddy]

Entire

[kumar_s]

Entire wont be a prescribed partition for Join stage. But still this will increase the number of resultant rows and not decrease.
Now explain more on what are the keys, and what is the partition that used on which stage and especially on the join stage, for both the input.
Basically need more details on job design.

[vivekreddy]

The key is a character field of length 2. In one dataset, partitioning method is Auto, whereas from the other, the left link, it is entire.

[vivekreddy]

The partitioning method on the Join stage is Entire.

[kumar_s]

I would suggest, do a hash partition on the Key, well before join, i.e,. the stages where you sort the data. And use same partition till Join stage.
Check if by any chance, you have any unique sort option enabled, and it removes duplicates.