Simon Riggs wrote:
> And if you have a partitioned table with partitions inconveniently
> sized? You'd need to *reduce* shared_buffers specifically to get synch
> scans and BAS to kick in. Or increase partition size. Both of which
> reduce the impact of the benefits we've added.
>
> I don't think the argument that "a table is smaller than shared buffers
> therefore it is already in shared buffers" holds true in all cases. I/O
> does matter.
>
+1. If we go with 'enable_sync_seqcans' for 8.3, and in a future release
cycle we do test the cases Simon described above and we agree we need to
do a fine tune to benefit from this feature, we will need to deprecate
'enable_sync_seqscans' and invent another one (sync_seqscans_threshold).
Looking at this perpective, IMHO we should go with the number (0.25)
instead of the boolean.
-- Euler Taveira de Oliveira http://www.timbira.com/