To give some perspective, we are using a transformer to split a data flow into 10 or more links, a lookup is executed on each data flow in parallel (some sparse, some normal) then the data is merged back together through a merge stage. When we notice the holes in the data, we strategically placed sequential files prior to the merge to ensure the lookups were indeed finding matches and compared the data to the output table on the other side of the Merge (where the 'holes' or blank columns were found).
We experiemented, and found the setting the partitioning on the input links to the Merge Stage to 'Entire' yeilded the correct result. According to the PX Guide:
Can anyone recommend best practices for partitioning with keyed input to a merge? Has anyone run into this same issue?The following partitioning methods are available:
�� (Auto). DataStage attempts to work out the best partitioning
method depending on execution modes of current and preceding
stages and how many nodes are specified in the Configuration
file. This is the default collection method for the Merge stage.
Thanks!