SOLVED, Re: Weird problem with scatter-gather I/O

They say that everything can be found in the last place you look. It wasn't the scatter-gather; it was a failed update.

As it happens, I hadn't checked the job log for duplicate key collisions on the update. When I did, I found that every affected record had one, on a logical that was uniquely keyed on a generated key field that was regenerated in my program.

This particular key was designed, at least in part, to enforce uniqueness on the records. And some outside program writing and/or updating the file had been modifying that generated key in order to subvert the enforced uniqueness.

--
JHHL