Re: Index Skip Scan (new UniqueKeys)

On 3/17/21 6:02 PM, Dmitry Dolgov wrote:
>> On Wed, Mar 17, 2021 at 03:28:00AM +0100, Tomas Vondra wrote:
>> Hi,
>>
>> I took a look at the new patch series, focusing mostly on the uniquekeys
>> part. It'd be a bit tedious to explain all the review comments here, so
>> attached is a patch series with a "review" patch for some of the parts.
>
> Great, thanks.
>
>> Most of it is fairly small (corrections to comments etc.), I'll go over
>> the more serious part so that we can discuss it here. I'll keep it split
>> per parts of the original patch series.
>> I suggest looking for XXX and FIXME comments in all the review patches.
>>
>>
>> 0001
>> ----
>>
>> ....
>>
>> 0002
>> ----
>>
>
> In fact both 0001 & 0002 belong to another thread, which these days
> span [1], [2]. I've included them only because they happened to be a
> dependency for index skip scan following David suggestions, sorry if
> it's confusing.
>
> At the same time the author behind 0001 & 0002 is present in this thread
> as well, maybe Andy can answer these comments right here and better than me.
>

Ah, sorry for the confusion. In that case the review comments probably
belong to the other threads, so we should move the discussion there.
It's not clear to me which of the threads is the right one.

>> 0003
>> ----
>>
>> Just some comments/whitespace.
>>
>>
>> 0004
>> ----
>>
>> I wonder why we don't include this in explain TEXT format? Seems it
>> might make it harder to write regression tests for this? It's easier to
>> just check that we deduced the right unique key(s) than having to
>> construct an example where it actually changes the plan.
>
> Yeah, good point. I believe originally it was like that to not make
> explain too verbose for skip scans, but displaying prefix definitely
> could be helpful for testing, so will do this (and address other
> comments as well).
>

Cool. Thanks.

regards

--
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company