Re: [PATCHES] Proposed patch: synchronized_scanning GUC variable

On Sun, 2008-01-27 at 21:04 -0500, Tom Lane wrote:

[ redirecting thread to -hackers ]

Neil Conway <neilc@samurai.com> writes:

On Sun, 2008-01-27 at 21:54 +0000, Gregory Stark wrote:

I liked the "synchronized_sequential_scans" idea myself.

I think that's a bit too long. How about "synchronized_scans", or

"synchronized_seqscans"?

We have enable_seqscan already, so that last choice seems to fit in.

If we're going to have a GUC, we may as well make it as useful as

possible.

Currently we set synch scan on when the table is larger than 25% of

shared_buffers. So increasing shared_buffers can actually turn this

feature off.

Rather than having a boolean GUC, we should have a number and make the

parameter "synchronised_scan_threshold". This would then be the size of

a table above which we would perform synch scans. If its set to -1, then

this would be the same as "off" in all cases. The default value would be

25% of shared_buffers. (Think we can only do that at initdb time

currently).

If we do that, its clearly different from the enable_* parameters, so

the name is easier to decide ;-)