Join stage is not giving expected result.

[kumar_s]

Vivek, I dont find a real need to clear the partition before sending it to Sort stage. If the Sort stage has partition in it, the existing partition will obviouly be re partitioned. The partition prior to that will not take effect.

[vivekreddy]

I was advised to do this when I had a similar problem with Join Stage. At that time, I understood the rationale for doing so as that when I clear the partitioning before sorting the data, I am sending the data to be sorted as a single block, and not as disjoint block.
After sorting, when the dataset is again partitioned, data that should be logically in the same neighbourhood on the basis of the key value would be placed in the same partition, something that I think might not happen if the partitions are not cleared before sorting?
I am not sure, but then this helped me, so thought it might work in this case as well.

[kumar_s]

If you are talking about Preserver Partition, there are three options.
Default is Propagate. This will try to use the same option what its previous stage is used. If the previous stage Preserver Partition has 'Set' this stage will also try to have 'Set' in it.
If Preserver Partition is set to hold the value 'Set', it instructs the following stage to follow the current partition as possible.
And if its set to 'Clear', the following stage should decide the partition of its own.
This will play a big role when used in Schema file with RCP enabled in Auto partition.
But specifying partition option in that stage will override this option.
If the case of OP, the first stage is active stage, where the option of Preserver partition default value will be 'Clear'.
Even if the it is set to 'Set' and if the consecutive stage is partitioned with a specific partition, the option will be over ridden.