Join Not Working on multiple nodes

[this_is_venki]

Hi

In one of the parallel jobs I had been using the config file as single.apt(1 node).
The job uses a Oracle reads,Transfomers,JOIN stage and finally Oracle write.
This job has been tested on the testing Environment.

Now for Performance reasons I just tried changing the config File from single.apt to medium.apt(4 nodes)

This actually lead to some incorrect data in the final table.
When i investigated for a specific record, i was able to spot that the join was not working.

Again when i ran with single.apt (For that specific record) It ran fine.

This is really baffling....
Does or not, Join Work with multiple nodes....
Someone please suggest something

TIA,
Venky

[ray.wurlod]

There is a requirement for the Join and Merge stages that the input Data Sets be identically partitioned and sorted on all the join keys. Have you configured this? A Lookup stage has the same requirement or, if you do not want to sort the primary input Data Set, that the reference inputs use Entire partitioning.

Yes, they DO work in parallel execution environment. It would be a strange thing to call it a parallel execution environment if they did not!

[kumar_s]

Try to sort and do a hash partion based on Joing key prior to join or merge for the both the input. So that the joining key will be similarly distributed across nodes for both set of data.

[pavankvk]

yeah..a simple hash partition should solve this problem...u may lose records with out specifying correct partition options..we had the same problem 2 years ago when we tested the software in test env with single node and had to fix this issue later in uat env whioch is multi node