Regarding Rowsplitter

eostic · Post by **eostic** » Sat Jan 20, 2007 4:27 pm

For what it's worth, the RowSplitter (and twin RowMerger) stages were written by the RTI development team to allow dynamic re-assignment of metadata definitions without having to land data to disk. There are times where developers (in DS Server) drop a passive stage in the middle of a job, for the purpose of re-defining the meta data. No comment here as to whether this is right or wrong, but it is/was often easier and more manageable (and at times faster) than doing lots of substrings, etc. in a Transformer..........except there's a catch --- an RTI Server Job can't have a Passive Stage with Input links AND Output Links (directly in the flow -- a job like that won't compile if RTI enabled). So the rowSplitter does it "on the fly," alleviating that issue, and giving you an option for RTI with an existing Job that uses the Passive Stage technique.

Ernie

ray.wurlod · Post by **ray.wurlod** » Sat Jan 20, 2007 7:03 pm

For those of you thinking about making the brain transition to parallel jobs, the same restriction applies - no stage that accesses persistent data is permitted to have both an input and an output link (apart from reject-handling links).

chulett · Post by **chulett** » Sat Jan 20, 2007 10:38 pm

Brain and brain, what is brain?

ray.wurlod · Post by **ray.wurlod** » Sun Jan 21, 2007 1:19 am

The same word, but in italics.

chulett · Post by **chulett** » Sun Jan 21, 2007 9:02 am

Correct - but it is, of course, also Spock's Brain. :D