Re: [BUGS] BUG #14155: bloom index error with unlogged table

On Wed, Jun 8, 2016 at 1:23 AM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> "David G. Johnston" <david(dot)g(dot)johnston(at)gmail(dot)com> writes:
>> On Tue, Jun 7, 2016 at 1:35 AM, Michael Paquier <michael(dot)paquier(at)gmail(dot)com>
>> wrote:
>>> I have finally given a shot at improving the docs with the attached.
>>> Comments are welcome.
>
>> [ assorted comments ]
>
> I adopted most of David's suggestions, whacked it around a bit further
> myself, and committed. See what you think.

That looks better, thanks.

>> It would be nice to give guidance on selecting a bit size for columns and
>> a signature length. Yes, Wikipedia covers the topic but to get the reader
>> started some discussion of the relevant trade-offs when using larger
>> numbers than the default would be nice. I don't suspect using smaller the
>> default values is apt to be worthwhile...
>
> Agreed, but I didn't want to write such text myself. There's room for
> further improvement here. I did add a note in the main example about
> what happens with a non-default signature length, but that hardly
> constitutes guidance.
>
> BTW, it seemed to me while generating the example that the planner's
> costing for bloom index searches was unduly pessimistic; maybe there's
> work to do there?

I wanted them to do so to prove that index rechecks are necessary as
false positives can be returned when scanning the index. We could add
an extra example with an index that has a longer signature size... I
am not sure that's worth the complication though.

I am marking this item as closed, in my view things are looking far better.
--
Michael