On Tue, Nov 3, 2015 at 5:21 PM, Craig Ringer <craig@2ndquadrant.com> wrote:
> On 3 November 2015 at 23:04, Bill Moran <wmoran@potentialtech.com> wrote:
>>
>> Looking for feedback to see if anyone sees any issues or has any
>> suggestions on what I'm doing. The attached patch alters 3 things
>> with regard to TOAST behavior:
>
> COMPRESSION_TEST_SIZE (2) seems useful.
>
> The other two mostly seem like options nobody's going to know are
> there, or know how to sensibly set if they do notice them. What's the
> driving reason behind those, the problem you're trying to solve? Why
> make them configurable per-table (or at all)?
I currently have a table with one column which has a median width of
500 bytes, a 90th percentile of 650 bytes, and makes up 75% of the
table's size, and the column is rarely used, while the table itself is
frequently seq scanned. I'd very much like to drive that column out
of main and into toast. I think target_tuple_size would let me do
that.
(Per-column control would be even nicer, but I'd take what I can get)
Cheers,
Jeff