On Tue, Jan 10, 2017 at 09:25:05PM -0500, Bruce Momjian wrote:
> Thank you for the summary. I think we have to consider two things with
> indirect indexes:
>
> 1. What percentage speedup is the _average_ user going to get? You
> have to consider people who will use indirect indexes who get no benefit
> or a net slowdown, and users who will get a benefit.
>
> 2. What percentage of users are going to use indirect indexes?
>
> So, for #1 you might have users who are getting +1%, +50%, and -20%, so
> maybe +10% average, and for #2 you might have 0.1%. When you multiply
> them out, you get 0.01% average improvement per installation, which is
> very small. Obviously, these are just wild guesses, but this is just to
> make a point. If you assume WARM has been optimized, #1 gets even
> lower.
Sorry to have to reply to my own email but I need some of the text
above.
Basically, with WARM, the adoption rate (#2) is 100%.
I am asking what instructions we will give users for #1 to prevent
people from using indirect indexes and getting worse performance. Are
we going to say, "Use indirect indexes on columns that are updated
frequently?" Actually, that seems pretty clear and would be easy for
users to follow.
I think the big question is that we will not know the benefits of
indirect indexes over WARM until we have implemented WARM, and if the
benefits of indirect indexes over WARM are small, and considering #2, we
might decide that it isn't worth adding it, for the reasons I already
outlined.
Therefore, I think we need WARM done first, then we can test indirect
indexes to see if they are a sufficient win to add it for the small
percentage of users who will use it.
In general, Postgres doesn't support ever possible performance tuning
option, and I think we are better for that because Postgres is simpler
to use. Going back to my blog post, if you add a feature, every user
who is considering tuning Postgres has to understand the feature and
decide if they should use it, so even for people who don't user the
feature, there is a cost, however small.
In summary, I love WARM, and might love indirect indexes too, but I need
to feel that indirect indexes are a clear win for the added complexity,
both in our code, and for the user API.
-- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB
http://enterprisedb.com
+ As you are, so once was I. As I am, so you will be. +
+ Ancient Roman grave inscription +