I guess you are refering to dbms_random?
I have some doubts that you can migrate this function to TD. Main issue is the parallel data distribution and parallel processing.
You clearly don't want to have a central maintained seed.
So alternative would be to initiallize this per vproc, which would still give you the problem that in case of sync scans the first row per SQL could be different even if you submit the same SQL twice.
What might work is to implement a pseudo random generator as UDF and you use the row input as seed - rowid would be best but can't be used any more. So you need to do it more complidated.
This would give you per row a different output number - still multiset tables would be an issue as same rows will get same output.
↧