Thanks for the confirmation - that's what I suspected. Edmond
On Thu, Feb 25, 2010 at 7:00 PM, Jonathan Ellis <jbel...@gmail.com> wrote: > Yes, that's going to hurt forward scans with no start column. > (Reverse scans, or scans that start with a known live column, will > still be fast b/c of the per-row column indexes.) > > On Thu, Feb 25, 2010 at 8:56 PM, Edmond Lau <edm...@ooyala.com> wrote: >> Given that Cassandra needs to maintain tombstones to handle >> distributed deletes, does the existence of deleted columns slow down >> slices? >> >> To be more concrete, suppose I used a row as a queue. I keep adding >> columns to the end of the sort order of a column family, and I keep >> deleting columns from the start of the sort order. After some time, >> the row would have a large number of deleted columns followed by a >> number of undeleted columns in the column family. Does slicing for >> the first N columns from the row now require scanning over all the >> initial deleted columns (meaning reads would get more expensive as >> time goes on), or are the deleted columns stored separately to enable >> Cassandra to skip over deleted columns when processing reads? >> >> Edmond >> >