On 3/10/07, Tom Lane <tgl@sss.pgh.pa.us> wrote:
Also, we know this case works because it already is working: in the
situation where VACUUM happens to visit and remove the DEAD tuple(s)
before reaching the RECENTLY_DEAD tuples that link forward to them,
it treats the RECENTLY_DEAD tuples as a disconnected chain and moves
them as-is. I saw tons of this in the traces I was making today, and
it doesn't seem to create any bad effects. (My attention was drawn to
it because I saw move_chain_tuple being used to move single-member
chains, which looks impossible when you first look at the code --- the
is-it-a-chain test seems to ensure that we can link either forward or
backward. But not so if t_ctid points to an already-removed tuple.)
Oh. So thats the corner case which I missed. This would probably
explain how we could miss marking an offset free and thus not remove
the corresponding index entry.
Thanks,
Pavan
--
EnterpriseDB
http://www.enterprisedb.com