I made a suggestion on the issue. It involves a change to the underlying system iterator that I think might work. No change to RowDeletingIterator will work, because it has no ability to inform the underlying system iterator outside of changing the column families.
-- Christopher L Tubbs II http://gravatar.com/ctubbsii On Wed, Jun 17, 2015 at 4:02 PM, Josh Elser <[email protected]> wrote: > Keith Turner wrote: >> >> >> >> On Wed, Jun 17, 2015 at 12:28 PM, Christopher <[email protected] >> <mailto:[email protected]>> wrote: >> >> Can you provide sample data which corresponds to these scan commands >> to help reproduce it? Also, are you using any other iterators which >> might be hiding the delete markers from this iterator at scan time? >> >> >> I think an Accumulo system iterator is hiding the delete markers. I >> made a comment on ACCUMULO-3905 about this. > > > Assuming that this is the case (perhaps boldly assuming this), that would > make the second user to be bitten by this. How can we improve this iterator > to prevent this from happening in the future? > > Change RowDeletingIterator so that it always includes the empty cf/cq column > in the fetched list?
