On Sunday, February 23, 2014 1:55:42 PM CEST, [email protected]
wrote:
You don't care if other ids become candidates while all this is
processing - they get caught next time around.
Right. What I care about ids that cease being candidates between the
identification and deletion phases, because when that happens there's a fat
ugly error message in the pg logs and the delete fails.
IIRC the problem with doing small batches was that when Abhijit and I
tried, we ended up doing a giant seqscan anyway to identify candidates.
I'll be very happy if Stephen has found a way that avoids that (or if pg
has grown more capable).
Arnt
